update

.buildkite/pipeline.yml

@@ -0,0 +1,6 @@
+steps:
+  - label: "Simple Unit Test"
+    commands:
+      - ".buildkite/scripts/simple_test.sh"
+    agents:
+      queue: "cpu_queue_premerge"

.buildkite/scripts/simple_test.sh

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd -- "${SCRIPT_DIR}/../.." && pwd)"
+UV_BIN="uv"
+PYTHON_BOOTSTRAP="${PYTHON:-python3}"
+VENV_DIR="${VENV_DIR:-${REPO_ROOT}/.venv-simple-test}"
+TARGET_PY_VERSION="3.12"
+
+if ! command -v "${PYTHON_BOOTSTRAP}" >/dev/null 2>&1; then
+  PYTHON_BOOTSTRAP="python"
+fi
+
+cd "${REPO_ROOT}"
+
+# Ensure pip-installed scripts (including uv) are on PATH
+PYTHON_SCRIPTS="$(${PYTHON_BOOTSTRAP} -c 'import sysconfig; print(sysconfig.get_path("scripts"))')"
+USER_BASE="$(${PYTHON_BOOTSTRAP} -m site --user-base 2>/dev/null || true)"
+PATH="${PYTHON_SCRIPTS}:${PATH}"
+if [[ -n "${USER_BASE}" ]]; then
+  PATH="${USER_BASE}/bin:${PATH}"
+fi
+
+if ! command -v "${UV_BIN}" >/dev/null 2>&1; then
+  "${PYTHON_BOOTSTRAP}" -m pip install --upgrade pip
+  "${PYTHON_BOOTSTRAP}" -m pip install uv
+  hash -r
+fi
+
+TARGET_PY_PATH="$(${UV_BIN} python find "${TARGET_PY_VERSION}" 2>/dev/null | head -n1 || true)"
+if [[ -z "${TARGET_PY_PATH}" ]]; then
+  "${UV_BIN}" python install "${TARGET_PY_VERSION}"
+  TARGET_PY_PATH="$(${UV_BIN} python find "${TARGET_PY_VERSION}" 2>/dev/null | head -n1 || true)"
+fi
+
+if [[ -z "${TARGET_PY_PATH}" ]]; then
+  echo "Failed to locate python ${TARGET_PY_VERSION}" >&2
+  exit 1
+fi
+
+if [[ ! -d "${VENV_DIR}" ]]; then
+  "${UV_BIN}" venv --python "${TARGET_PY_PATH}" "${VENV_DIR}"
+fi
+
+VENV_PYTHON="${VENV_DIR}/bin/python"
+[[ -x "${VENV_PYTHON}" ]] || { echo "Python not found in ${VENV_DIR}"; exit 1; }
+
+"${UV_BIN}" pip install --python "${VENV_PYTHON}" vllm==0.11.0
+"${UV_BIN}" pip install --python "${VENV_PYTHON}" -e ".[dev]"
+"${VENV_PYTHON}" -m pytest -v -s tests/test_omni_llm.py

.gitattributes

@@ -0,0 +1,44 @@
+# Set default behavior to automatically normalize line endings
+* text=auto
+
+# Explicitly declare text files you want to always be normalized and converted
+# to native line endings on checkout
+*.py text eol=lf
+*.yaml text eol=lf
+*.yml text eol=lf
+*.md text eol=lf
+*.txt text eol=lf
+*.json text eol=lf
+*.toml text eol=lf
+*.sh text eol=lf
+*.bat text eol=crlf
+*.cmd text eol=crlf
+
+# Declare files that will always have CRLF line endings on checkout
+*.bat text eol=crlf
+*.cmd text eol=crlf
+
+# Denote all files that are truly binary and should not be modified
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.gif binary
+*.ico binary
+*.mov binary
+*.mp4 binary
+*.mp3 binary
+*.flv binary
+*.fla binary
+*.swf binary
+*.gz binary
+*.zip binary
+*.7z binary
+*.ttf binary
+*.eot binary
+*.woff binary
+*.woff2 binary
+*.pyc binary
+*.pdf binary
+*.ai binary
+*.psd binary
+*.xcf binary

.github/CODEOWNERS

@@ -0,0 +1,17 @@
+# CODEOWNERS is used to automatically assign PR reviewers
+# Format: path pattern @username or @team
+# More info: https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/customizing-your-repository/about-code-owners
+
+# Default owners (fallback reviewers for all files)
+# later, we can add more reviewers here
+* @hsliuustc0106
+
+# Assign reviewers by directory
+/.github/ @owner1
+/tests/ @owner1
+/docs/ @owner1
+
+# Example: reviewers for specific directories
+# /examples/offline_inference/ @offline-inference-team
+# /vllm_omni/core/ @core-team
+# /vllm_omni/model_executor/ @model-team

.github/ISSUE_TEMPLATE/100-documentation.yml

@@ -0,0 +1,29 @@
+name: 📚 Documentation
+description: Report an issue related to documentation
+title: "[Doc]: "
+labels: ["documentation"]
+
+body:
+- type: textarea
+  attributes:
+    label: 📚 The doc issue
+    description: >
+      A clear and concise description of what content in the documentation is an issue.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Suggest a potential alternative/fix
+    description: >
+      Tell us how we could improve the documentation in this regard.
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/200-installation.yml

@@ -0,0 +1,47 @@
+name: 🛠️ Installation
+description: Report an issue here when you hit errors during installation.
+title: "[Installation]: "
+labels: ["installation"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm-omni/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Your current environment
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm-omni/main/vllm_omni/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+      It is suggested to download and execute the latest script, as vllm-omni might frequently update the diagnosis information needed for accurately and quickly responding to issues.
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: How you are installing vllm-omni
+    description: |
+      Paste the full command you are trying to execute.
+    value: |
+      ```sh
+      pip install -vvv vllm-omni
+      ```
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/400-bug-report.yml

@@ -0,0 +1,111 @@
+name: 🐛 Bug report
+description: Raise an issue here if you find a bug.
+title: "[Bug]: "
+labels: ["bug"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm-omni/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: markdown
+  attributes:
+    value: |
+      ⚠️ **SECURITY WARNING:** Please review any text you paste to ensure it does not contain sensitive information such as:
+      - API tokens or keys (e.g., Hugging Face tokens, OpenAI API keys)
+      - Passwords or authentication credentials
+      - Private URLs or endpoints
+      - Personal or confidential data
+
+      Consider redacting or replacing sensitive values with placeholders like `<YOUR_TOKEN_HERE>` when sharing configuration or code examples.
+- type: textarea
+  attributes:
+    label: Your current environment
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm-omni/main/vllm_omni/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+      It is suggested to download and execute the latest script, as vllm-omni might frequently update the diagnosis information needed for accurately and quickly responding to issues.
+    value: |
+      <details>
+      <summary>The output of <code>python collect_env.py</code></summary>
+
+      ```text
+      Your output of `python collect_env.py` here
+      ```
+
+      </details>
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: 🐛 Describe the bug
+    description: |
+      Please provide a clear and concise description of what the bug is.
+
+      If relevant, add a minimal example so that we can reproduce the error by running the code. It is very important for the snippet to be as succinct (minimal) as possible, so please take time to trim down any irrelevant code to help us debug efficiently. We are going to copy-paste your code and we expect to get the same result as you did: avoid any external data, and include the relevant imports, etc. For example:
+
+      ```python
+      from vllm_omni import OmniLLM, create_ar_stage_config, create_dit_stage_config
+
+      # Create stage configurations
+      ar_config = create_ar_stage_config(
+          stage_id=0,
+          model_path="Qwen/Qwen3-0.6B",
+          input_modalities=["text"],
+          output_modalities=["text"]
+      )
+
+      dit_config = create_dit_stage_config(
+          stage_id=1,
+          model_path="stabilityai/stable-diffusion-2-1",
+          input_modalities=["text"],
+          output_modalities=["image"]
+      )
+
+      # Initialize OmniLLM
+      omni = OmniLLM([ar_config, dit_config])
+
+      # Generate
+      outputs = omni.generate(prompt="A scenic watercolor painting of a lighthouse at sunset")
+      ```
+
+      If the code is too long (hopefully, it isn't), feel free to put it in a public gist and link it in the issue: https://gist.github.com.
+
+      Please also paste or describe the results you observe instead of the expected results. If you observe an error, please paste the error message including the **full** traceback of the exception. It may be relevant to wrap error messages in ```` ```triple quotes blocks``` ````.
+
+      Please set the environment variable `export VLLM_OMNI_LOGGING_LEVEL=DEBUG` to turn on more logging to help debugging potential issues.
+
+      If you experienced crashes or hangs, it would be helpful to run vllm-omni with `export VLLM_OMNI_TRACE_FUNCTION=1` . All the function calls in vllm-omni will be recorded. Inspect these log files, and tell which function crashes or hangs.
+    placeholder: |
+      A clear and concise description of what the bug is.
+
+      ```python
+      # Sample code to reproduce the problem
+      ```
+
+      ```
+      The error message you got, with the full traceback and the error logs with [dump_input.py:##] if present.
+      ```
+  validations:
+    required: true
+- type: markdown
+  attributes:
+    value: |
+      ⚠️ Please separate bugs of `transformers` implementation or usage from bugs of `vllm-omni`. If you think anything is wrong with the model's output:
+
+      - Try the counterpart of `transformers` first. If the error appears, please go to [their issues](https://github.com/huggingface/transformers/issues?q=is%3Aissue+is%3Aopen+sort%3Aupdated-desc).
+
+      - If the error only appears in vllm-omni, please provide the detailed script of how you run `transformers` and `vllm-omni`, also highlight the difference and what you expect.
+
+      Thanks for reporting 🙏!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/500-feature-request.yml

@@ -0,0 +1,38 @@
+name: 🚀 Feature request
+description: Submit a proposal/request for a new vllm-omni feature
+title: "[Feature]: "
+labels: ["feature request"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm-omni/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: 🚀 The feature, motivation and pitch
+    description: >
+      A clear and concise description of the feature proposal. Please outline the motivation for the proposal. Is your feature request related to a specific problem? e.g., *"I'm working on X and would like Y to be possible"*. If this is related to another GitHub issue, please link here too.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Alternatives
+    description: >
+      A description of any alternative solutions or features you've considered, if any.
+- type: textarea
+  attributes:
+    label: Additional context
+    description: >
+      Add any other context or screenshots about the feature request.
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/600-new-model.yml

@@ -0,0 +1,45 @@
+name: 🤗 Support request for a new model
+description: Submit a proposal/request for a new model support
+title: "[New Model]: "
+labels: ["new model"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm-omni/issues?q=is%3Aissue+sort%3Acreated-desc+).
+
+      #### We also highly recommend you read the contributing guidelines first to understand how to add a new model.
+- type: textarea
+  attributes:
+    label: The model to consider.
+    description: >
+      A huggingface url, pointing to the model, e.g. https://huggingface.co/openai-community/gpt2 .
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: The closest model vllm-omni already supports.
+    description: >
+      Here is the list of models already supported by vllm-omni. Which model is the most similar to the model you want to add support for?
+- type: textarea
+  attributes:
+    label: What's your difficulty of supporting the model you want?
+    description: >
+      For example, any new operators or new architecture? Is it a multimodal model? Does it require special preprocessing?
+- type: textarea
+  attributes:
+    label: Use case and motivation
+    description: >
+      What is the primary use case for this model? Why is it important to support in vllm-omni?
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/700-performance-discussion.yml

@@ -0,0 +1,59 @@
+name: ⚡ Discussion on the performance of vllm-omni
+description: Submit a proposal/discussion about the performance of vllm-omni
+title: "[Performance]: "
+labels: ["performance"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm-omni/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Proposal to improve performance
+    description: >
+      How do you plan to improve vllm-omni's performance?
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Report of performance regression
+    description: >
+      Please provide detailed description of performance comparison to confirm the regression. You may want to run the benchmark script at https://github.com/vllm-project/vllm-omni/tree/main/tests/benchmarks .
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Misc discussion on performance
+    description: >
+      Anything about the performance.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Your current environment (if you think it is necessary)
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm-omni/main/vllm_omni/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+      It is suggested to download and execute the latest script, as vllm-omni might frequently update the diagnosis information needed for accurately and quickly responding to issues.
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: false
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/750-RFC.yml

@@ -0,0 +1,52 @@
+name: 💬 Request for comments (RFC).
+description: Ask for feedback on major architectural changes or design choices.
+title: "[RFC]: "
+labels: ["RFC"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Please take a look at previous [RFCs](https://github.com/vllm-project/vllm-omni/issues?q=label%3ARFC+sort%3Aupdated-desc) for reference.
+- type: textarea
+  attributes:
+    label: Motivation.
+    description: >
+      The motivation of the RFC.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Proposed Change.
+    description: >
+      The proposed change of the RFC.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Feedback Period.
+    description: >
+      The feedback period of the RFC. Usually at least one week.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: CC List.
+    description: >
+      The list of people you want to CC.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Any Other Things.
+    description: >
+      Any other things you would like to mention.
+  validations:
+    required: false
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://vllm-omni.readthedocs.io), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,5 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Questions
+    url: https://github.com/vllm-project/vllm-omni/discussions
+    about: Ask questions and discuss with other vLLM-Omni community members

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,21 @@
+<!-- markdownlint-disable -->
+PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.
+
+## Purpose
+
+## Test Plan
+
+## Test Result
+
+---
+<details>
+<summary> Essential Elements of an Effective PR Description Checklist </summary>
+
+- [ ] The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
+- [ ] The test plan, such as providing test command.
+- [ ] The test results, such as pasting the results comparison before and after, or e2e results
+- [ ] (Optional) The necessary documentation update, such as updating `supported_models.md` and `examples` for a new model.
+- [ ] (Optional) Release notes update. If your change is user facing, please update the release notes draft.
+</details>
+
+**BEFORE SUBMITTING, PLEASE READ <https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md>** (anything written below this line will be removed by GitHub Actions)

.github/scripts/detect_changed_tests.sh

@@ -0,0 +1,89 @@
+#!/bin/bash
+# Script to detect changed files and find corresponding test files
+# Usage: ./detect_changed_tests.sh [base_sha] [head_sha] [event_type]
+#
+# Outputs to GITHUB_OUTPUT:
+#   test_files: Space-separated list of test files to run
+#   has_tests: 'true' if test files found, 'false' otherwise
+
+set -e
+
+BASE_SHA="${1:-}"
+HEAD_SHA="${2:-}"
+EVENT_TYPE="${3:-push}"
+
+# Determine base and head SHA based on event type
+if [ "$EVENT_TYPE" = "pull_request" ]; then
+    BASE_SHA="${BASE_SHA:-$GITHUB_BASE_REF}"
+    HEAD_SHA="${HEAD_SHA:-$GITHUB_SHA}"
+elif [ -z "$BASE_SHA" ] || [ -z "$HEAD_SHA" ]; then
+    # For push events, use git to find the previous commit
+    BASE_SHA="${BASE_SHA:-$(git rev-parse HEAD~1 2>/dev/null || echo HEAD)}"
+    HEAD_SHA="${HEAD_SHA:-HEAD}"
+fi
+
+echo "Event type: $EVENT_TYPE"
+echo "Base SHA: $BASE_SHA"
+echo "Head SHA: $HEAD_SHA"
+
+# Get changed Python files
+CHANGED_FILES=$(git diff --name-only --diff-filter=ACMRT "$BASE_SHA" "$HEAD_SHA" | grep -E '\.(py)$' || true)
+
+if [ -z "$CHANGED_FILES" ]; then
+    echo "No Python files changed"
+    echo "test_files=" >> "$GITHUB_OUTPUT"
+    echo "has_tests=false" >> "$GITHUB_OUTPUT"
+    exit 0
+fi
+
+echo "Changed Python files:"
+echo "$CHANGED_FILES"
+
+# Extract related test files
+TEST_FILES=""
+for file in $CHANGED_FILES; do
+    # If changed file is a test file, add it directly
+    if [[ "$file" == tests/* ]]; then
+        TEST_FILES="$TEST_FILES $file"
+    # If changed file is source code, find corresponding test file
+    elif [[ "$file" == vllm_omni/* ]]; then
+        # Convert vllm_omni/path/to/module.py to possible test file paths
+        REL_PATH=$(echo "$file" | sed 's|^vllm_omni/||' | sed 's|\.py$||')
+        MODULE_NAME=$(basename "$REL_PATH")
+        DIR_PATH=$(dirname "$REL_PATH")
+
+        # Try multiple possible test file naming patterns
+        POSSIBLE_TESTS=(
+            "tests/test_${REL_PATH//\//_}.py"
+            "tests/${REL_PATH//\//_}_test.py"
+        )
+
+        # If multi-level directory, also try directory structure
+        if [ "$DIR_PATH" != "." ]; then
+            POSSIBLE_TESTS+=(
+                "tests/${DIR_PATH}/test_${MODULE_NAME}.py"
+                "tests/${DIR_PATH}/${MODULE_NAME}_test.py"
+            )
+        fi
+
+        for test_file in "${POSSIBLE_TESTS[@]}"; do
+            if [ -f "$test_file" ]; then
+                TEST_FILES="$TEST_FILES $test_file"
+                break
+            fi
+        done
+    fi
+done
+
+# Deduplicate and format
+TEST_FILES=$(echo "$TEST_FILES" | tr ' ' '\n' | grep -v '^$' | sort -u | tr '\n' ' ' | sed 's/^ *//;s/ *$//')
+
+if [ -z "$TEST_FILES" ]; then
+    echo "No related test files found, will run all tests"
+    echo "test_files=" >> "$GITHUB_OUTPUT"
+    echo "has_tests=false" >> "$GITHUB_OUTPUT"
+else
+    echo "Found test files: $TEST_FILES"
+    echo "test_files=$TEST_FILES" >> "$GITHUB_OUTPUT"
+    echo "has_tests=true" >> "$GITHUB_OUTPUT"
+fi

.github/workflows/pre-commit.yml

@@ -0,0 +1,26 @@
+name: pre-commit
+
+on:
+  pull_request:
+    branches: [main]
+  push:
+    branches: [main]
+  workflow_dispatch: # for manual trigger
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+permissions:
+  contents: read
+
+jobs:
+  pre-commit:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+          cache: 'pip'
+      - uses: pre-commit/action@v3.0.1

.gitignore

@@ -0,0 +1,237 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+vllm/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# cursor
+.cursor/
+
+# docker
+docker/
+
+# scripts
+/scripts/
+
+# tests
+tests/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be added to the global gitignore or merged into this project gitignore.  For a PyCharm
+#  project, it is recommended to include the following files:
+#  .idea/
+#  *.iml
+#  *.ipr
+#  *.iws
+#  .idea/
+
+# vLLM-Omni specific
+# Model files and checkpoints
+checkpoints/
+*.bin
+*.safetensors
+*.pt
+*.pth
+
+# Cache directories
+cache/
+.cache/
+diffusion_cache/
+kv_cache/
+
+# Logs
+logs/
+*.log
+
+# Temporary files
+tmp/
+temp/
+.tmp/
+
+# IDE files
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# OS files
+.DS_Store
+.DS_Store?
+._*
+.Spotlight-V100
+.Trashes
+ehthumbs.db
+Thumbs.db
+
+# Data files
+data/
+datasets/
+*.csv
+*.json
+*.jsonl
+*.parquet
+
+# Output files
+outputs/
+results/
+generated/
+
+# Configuration overrides
+configs/local.yaml
+configs/production.yaml
+configs/development.yaml
+
+# Docker
+.dockerignore
+Dockerfile.dev

.pre-commit-config.yaml

@@ -0,0 +1,60 @@
+default_install_hook_types:
+  - pre-commit
+  - commit-msg
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.6.0
+    hooks:
+      # list of supported hooks: https://pre-commit.com/hooks.html
+      - id: check-yaml
+        args: ["--unsafe"]
+      - id: debug-statements
+      - id: end-of-file-fixer
+      - id: mixed-line-ending
+        args: ["--fix=lf"]
+      - id: trailing-whitespace
+        args: ["--markdown-linebreak-ext=md"]
+
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.14.0
+    hooks:
+    - id: ruff-check
+      args: [--output-format, github, --fix]
+    - id: ruff-format
+
+  - repo: https://github.com/crate-ci/typos
+    rev: v1.35.5
+    hooks:
+      - id: typos
+        # only for staged files
+
+  - repo: https://github.com/rhysd/actionlint
+    rev: v1.7.7
+    hooks:
+      - id: actionlint
+        files: ^\.github/workflows/.*\.ya?ml$
+
+
+  - repo: local
+    hooks:
+    - id: signoff-commit
+      name: Sign-off Commit
+      entry: bash
+      args:
+        - -c
+        - |
+          if ! grep -q "^Signed-off-by: $(git config user.name) <$(git config user.email)>" "$(git rev-parse --git-path COMMIT_EDITMSG)"; then
+            printf "\nSigned-off-by: $(git config user.name) <$(git config user.email)>\n" >> "$(git rev-parse --git-path COMMIT_EDITMSG)"
+          fi
+      language: system
+      verbose: true
+      stages: [commit-msg]
+
+    # Keep `suggestion` last
+    - id: suggestion
+      name: Suggestion
+      entry: bash -c 'echo "To bypass all the pre-commit hooks, add --no-verify to git commit. To skip a specific hook, prefix the commit command with SKIP=<hook-id>."'
+      language: system
+      verbose: true
+      pass_filenames: false
+    # Insert new entries above the `suggestion` entry

.readthedocs.yml

@@ -0,0 +1,24 @@
+# Read the Docs configuration file
+# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
+
+version: 2
+
+build:
+  os: ubuntu-22.04
+  tools:
+    python: "3.12"
+  jobs:
+    post_checkout:
+      - git fetch --unshallow || true
+
+mkdocs:
+  configuration: mkdocs.yml
+  fail_on_warning: true
+
+# Optionally declare the Python requirements required to build your docs
+python:
+  install:
+    - method: pip
+      path: .
+      extra_requirements:
+        - docs

CONTRIBUTING.md

@@ -0,0 +1,3 @@
+# Contributing to vLLM-Omni
+
+You may find information about contributing to vLLM-Omni on [Contributing](docs/contributing/)

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,78 @@
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" src="./docs/source/logos/vllm-omni-logo-text-dark.png">
+    <img alt="vllm-omni" src="./docs/source/logos/vllm-omni-logo-text-dark.png" width=55%>
+  </picture>
+</p>
+<h3 align="center">
+Easy, fast, and cheap omni-modality model serving for everyone
+</h3>
+
+<p align="center">
+| <a href="https://vllm-omni.readthedocs.io/en/latest/"><b>Documentation</b></a> | <a href="https://tinyurl.com/vllm-omni-meeting"><b>Weekly Meeting</b></a> |
+</p>
+
+---
+
+*Latest News* 🔥
+
+- [2025/11] vLLM community officially released [vllm-project/vllm-omni](https://github.com/vllm-project/vllm-omni) in order to support omni-modality models serving.
+
+---
+
+## About
+
+[vLLM](https://github.com/vllm-project/vllm) was originally designed to support large language models for text-based autoregressive generation tasks. vLLM-Omni is a framework that extends its support for omni-modality model inference and serving:
+
+- **Omni-modality**: Text, image, video, and audio data processing
+- **Non-autoregressive Architectures**: extend the AR support of vLLM to Diffusion Transformers (DiT) and other parallel generation models
+- **Heterogeneous outputs**: from traditional text generation to multimodal outputs
+
+<p align="center">
+  <picture>
+    <img alt="vllm-omni" src="./docs/source/architecture/omni-modality model architecture.png" width=55%>
+  </picture>
+</p>
+
+vLLM-Omni is fast with:
+
+- State-of-the-art AR support by leveraging efficient KV cache management from vLLM
+- Pipelined stage execution overlapping for high throughput performance
+- Fully disaggregation based on OmniConnector and dynamic resource allocation across stages
+
+vLLM-Omni is flexible and easy to use with:
+
+- Heterogeneous pipeline abstraction to manage complex model workflows
+- Seamless integration with popular Hugging Face models
+- Tensor, pipeline, data and expert parallelism support for distributed inference
+- Streaming outputs
+- OpenAI-compatible API server
+
+vLLM-Omni seamlessly supports most popular open-source models on HuggingFace, including:
+
+- Omni-modality models (e.g. Qwen-Omni)
+- Multi-modality generation models (e.g. Qwen-Image)
+
+## Getting Started
+
+Visit our [documentation](https://vllm-omni.readthedocs.io/en/latest/) to learn more.
+
+Visit our documentation to learn more.
+
+- [Installation](https://vllm-omni.readthedocs.io/en/latest/getting_started/installation/)
+- [Quickstart](https://vllm-omni.readthedocs.io/en/latest/getting_started/quickstart/)
+- [List of Supported Models](https://vllm-omni.readthedocs.io/en/latest/models/supported_models/)
+
+## Contributing
+
+We welcome and value any contributions and collaborations.
+Please check out [Contributing to vLLM-Omni](https://vllm-omni.readthedocs.io/en/latest/contributing/) for how to get involved.
+
+## Weekly Meeting
+
+- vLLM Omni Weekly Meeting: https://tinyurl.com/vllm-omni-meeting
+- Wednesday, 11:30 - 12:30 (UTC+8, [Convert to your timezone](https://dateful.com/convert/gmt8?t=15))
+
+## License
+
+Apache License 2.0, as found in the [LICENSE](./LICENSE) file.

docs/.nav.yml

@@ -0,0 +1,44 @@
+nav:
+  - Home: README.md
+  - User Guide:
+    - Getting Started:
+      - getting_started/quickstart.md
+      - getting_started/installation
+    - Examples:
+      - examples/README.md
+      - Offline Inference:
+        - Qwen2.5-Omni: user_guide/examples/offline_inference/qwen2_5_omni.md
+      - Online Serving:
+        - Qwen2.5-Omni: user_guide/examples/online_serving/qwen2_5_omni.md
+        - Qwen2.5-Image: user_guide/examples/offline_inference/qwen_image.md
+    - General:
+      - usage/*
+    - Configuration:
+      - configuration/README.md
+      - configuration/*
+    - Models:
+      - models/supported_models.md
+  - Developer Guide:
+    - General:
+      - contributing/README.md
+      - glob: contributing/*
+        flatten_single_child_sections: true
+    - Model Implementation:
+      - contributing/model/README.md
+    - CI: contributing/ci
+    - Design Documents:
+      - design/index.md
+      - design/architecture_overview.md
+      - design/vllm_omni_design.md
+      - design/mrs_design.md
+      - design/api_design_doc.md
+    - Docs Guide: contributing/DOCS_GUIDE.md
+  - API Reference:
+    - api/README.md
+    - api/vllm_omni
+  - CLI Reference: cli
+  - Community:
+    - community/*
+    - Slack: https://slack.vllm.ai
+    - Blog: https://blog.vllm.ai
+    - Forum: https://discuss.vllm.ai

docs/README.md

@@ -0,0 +1,61 @@
+---
+hide:
+  - navigation
+  - toc
+---
+
+# Welcome to vLLM-Omni
+
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" src="./source/logos/vllm-omni-logo-text-dark.png">
+    <img alt="vllm-omni" src="./source/logos/vllm-omni-logo-text-dark.png" width=55%>
+  </picture>
+</p>
+<h3 align="center">
+Easy, fast, and cheap omni-modality model serving for everyone
+</h3>
+
+<p style="text-align:center">
+<script async defer src="https://buttons.github.io/buttons.js"></script>
+<a class="github-button" href="https://github.com/vllm-project/vllm-omni" data-show-count="true" data-size="large" aria-label="Star">Star</a>
+<a class="github-button" href="https://github.com/vllm-project/vllm-omni/subscription" data-show-count="true" data-icon="octicon-eye" data-size="large" aria-label="Watch">Watch</a>
+<a class="github-button" href="https://github.com/vllm-project/vllm-omni/fork" data-show-count="true" data-icon="octicon-repo-forked" data-size="large" aria-label="Fork">Fork</a>
+</p>
+
+
+## About
+
+[vLLM](https://github.com/vllm-project/vllm) was originally designed to support large language models for text-based autoregressive generation tasks. vLLM-Omni is a framework that extends its support for omni-modality model inference and serving:
+
+- **Omni-modality**: Text, image, video, and audio data processing
+- **Non-autoregressive Architectures**: extend the AR support of vLLM to Diffusion Transformers (DiT) and other parallel generation models
+- **Heterogeneous outputs**: from traditional text generation to multimodal outputs
+
+<p align="center">
+  <picture>
+    <img alt="vllm-omni" src="./source/architecture/omni-modality model architecture.png" width=55%>
+  </picture>
+</p>
+
+vLLM-Omni is fast with:
+
+- State-of-the-art AR support by leveraging efficient KV cache management from vLLM
+- Pipelined stage execution overlapping for high throughput performance
+- Fully disaggregation based on OmniConnector and dynamic resource allocation across stages
+
+vLLM-Omni is flexible and easy to use with:
+
+- Heterogeneous pipeline abstraction to manage complex model workflows
+- Seamless integration with popular Hugging Face models
+- Tensor, pipeline, data and expert parallelism support for distributed inference
+- Streaming outputs
+- OpenAI-compatible API server
+
+vLLM-Omni seamlessly supports most popular open-source models on HuggingFace, including:
+
+- Omni-modality models (e.g. Qwen2.5-Omni, Qwen3-Omni)
+- Multi-modality generation models (e.g. Qwen-Image)
+
+For more information, checkout the following:
+- [Placeholder for vllm-omni blogpost]()

docs/api/README.md

@@ -0,0 +1,92 @@
+# Summary
+
+## Entry Points
+
+Main entry points for vLLM-Omni inference and serving.
+
+- [vllm_omni.entrypoints.async_omni_llm.AsyncOmniLLM][]
+- [vllm_omni.entrypoints.async_omni_llm.AsyncOmniStageLLM][]
+- [vllm_omni.entrypoints.chat_utils.OmniAsyncMultiModalContentParser][]
+- [vllm_omni.entrypoints.chat_utils.OmniAsyncMultiModalItemTracker][]
+- [vllm_omni.entrypoints.chat_utils.parse_chat_messages_futures][]
+- [vllm_omni.entrypoints.cli.serve.OmniServeCommand][]
+- [vllm_omni.entrypoints.log_utils.OrchestratorMetrics][]
+- [vllm_omni.entrypoints.omni.Omni][]
+- [vllm_omni.entrypoints.omni_llm.OmniLLM][]
+- [vllm_omni.entrypoints.omni_llm.OmniStageLLM][]
+- [vllm_omni.entrypoints.omni_stage.OmniStage][]
+- [vllm_omni.entrypoints.openai.serving_chat.OmniOpenAIServingChat][]
+
+## Inputs
+
+Input data structures for multi-modal inputs.
+
+- [vllm_omni.inputs.data.OmniEmbedsPrompt][]
+- [vllm_omni.inputs.data.OmniTokenInputs][]
+- [vllm_omni.inputs.data.OmniTokensPrompt][]
+- [vllm_omni.inputs.parse.parse_singleton_prompt_omni][]
+- [vllm_omni.inputs.preprocess.OmniInputPreprocessor][]
+
+## Engine
+
+Engine classes for offline and online inference.
+
+- [vllm_omni.diffusion.diffusion_engine.DiffusionEngine][]
+- [vllm_omni.engine.AdditionalInformationEntry][]
+- [vllm_omni.engine.AdditionalInformationPayload][]
+- [vllm_omni.engine.OmniEngineCoreRequest][]
+- [vllm_omni.engine.PromptEmbedsPayload][]
+- [vllm_omni.engine.arg_utils.AsyncOmniEngineArgs][]
+- [vllm_omni.engine.arg_utils.OmniEngineArgs][]
+- [vllm_omni.engine.output_processor.MultimodalOutputProcessor][]
+- [vllm_omni.engine.output_processor.OmniRequestState][]
+- [vllm_omni.engine.processor.OmniProcessor][]
+
+## Core
+
+Core scheduling and caching components.
+
+- [vllm_omni.core.dit_cache_manager.DiTCacheManager][]
+- [vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler][]
+- [vllm_omni.core.sched.output.OmniNewRequestData][]
+- [vllm_omni.core.sched.scheduler.OmniScheduler][]
+
+## Model Executor
+
+Model execution components.
+
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni.OmniOutput][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni.Qwen2_5OmniForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_talker.Qwen2_5OmniTalkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniAudioFeatureInputs][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniConditionalGenerationMixin][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerDummyInputsBuilder][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerMultiModalDataParser][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerMultiModalProcessor][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerProcessingInfo][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavBigVGANModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavDiTModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavForConditionalGenerationVLLM][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2EmbeddingModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2ForCausalLM][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2Model][]
+
+## Configuration
+
+Configuration classes.
+
+- [vllm_omni.config.model.OmniModelConfig][]
+
+## Workers
+
+Worker classes and model runners for distributed inference.
+
+- [vllm_omni.diffusion.worker.gpu_worker.GPUWorker][]
+- [vllm_omni.diffusion.worker.gpu_worker.WorkerProc][]
+- [vllm_omni.worker.gpu_ar_model_runner.GPUARModelRunner][]
+- [vllm_omni.worker.gpu_ar_worker.GPUARWorker][]
+- [vllm_omni.worker.gpu_diffusion_model_runner.GPUDiffusionModelRunner][]
+- [vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker][]
+- [vllm_omni.worker.gpu_model_runner.OmniGPUModelRunner][]

docs/cli/README.md

@@ -0,0 +1,24 @@
+# vLLM-Omni CLI Guide
+
+The CLI for vLLM-Omni inherits from vllm with some additional arguments.
+
+## serve
+
+Starts the vLLM-Omni OpenAI Compatible API server.
+
+Start with a model:
+
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni
+```
+
+Specify the port:
+
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091
+```
+
+If you have custom stage configs file, launch the server with command below
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --stage-configs-path /path/to/stage_configs_file
+```

docs/community/contact_us.md

@@ -0,0 +1,5 @@
+# Contact Us
+
+- For technical questions and feature requests, please use GitHub [Issues](https://github.com/vllm-project/vllm-omni/issues)
+- For coordinating contributions and development and discussing with other users and developers, please join `sig-omni` channel in our [Slack](https://slack.vllm.ai/) or use the [vLLM Forum](https://discuss.vllm.ai/)
+- For security disclosures, please use GitHub's [Security Advisories](https://github.com/vllm-project/vllm/security/advisories) feature

docs/community/meetups.md

@@ -0,0 +1 @@
+# Meetups

docs/configuration/README.md

@@ -0,0 +1,11 @@
+# Configuration Options
+
+This section lists the most common options for running vLLM-Omni.
+
+For options within a vLLM Engine. Please refer to [vLLM Configuration](https://docs.vllm.ai/en/v0.11.0/configuration/index.html)
+
+Currently, the main options are maintained by stage configs for each model.
+
+For specific example, please refer to [Qwen2.5-omni stage config](stage_configs/qwen2_5_omni.yaml)
+
+For introduction, please check [Introduction for stage config](./stage_configs.md)

docs/configuration/stage_configs.md

@@ -0,0 +1,269 @@
+# Stage configs for vLLM-Omni
+
+In vLLM-Omni, the target model is separated into multiple stages, which are processed by different LLMEngines, DiffusionEngines or other types of engines. Depending on different types of stages, such as Autoregressive (AR) stage or Diffusion transformer (DiT) stage, each can choose corresponding schedulers, model workers to load with the Engines in a plug-in fashion.
+
+Therefore, as a core part of vLLM-Omni, the stage configs for a model have several main functions:
+
+- Claim partition of stages and their corresponding class implementation in `model_executor/models`.
+- The disaggregated configuration for each stage and the communication topology among them.
+- Engine arguments for each engine within the stage.
+- Input and output dependencies for each stage.
+- Default input parameters.
+
+If users want to modify some part of it. The custom stage_configs file can be input as input argument in both online and offline. Just like examples below:
+
+For offline (Assume necessary dependencies have ben imported):
+```python
+model_name = "Qwen/Qwen2.5-Omni-7B"
+omni_llm = OmniLLM(model=model_name, stage_configs_path="/path/to/custom_stage_configs.yaml")
+```
+
+For online serving:
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
+```
+!!! important
+    We are actively iterating on the definition of stage configs, and we welcome all feedbacks from both community users and developers to help us shape the development!
+
+Below is a specific example of stage_configs.yaml in Qwen2.5-omni.
+```python
+# stage config for running qwen2.5-omni with architecture of OmniLLM.
+stage_args:
+  - stage_id: 0 # mark the unique id for each stage
+    runtime: # The disaggregated configuration
+      process: true  # Run this stage in a separate process
+      devices: "0" # Visible devices for this stage (CUDA_VISIBLE_DEVICES/torch.cuda.set_device)
+      max_batch_size: 1 # the batch_size for offline inference
+    engine_args: # Engine arguments for a certain engine
+      model_stage: thinker
+      model_arch: Qwen2_5OmniForConditionalGeneration # The model implementation registered in model_executor/models/registry.py
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker # The specific worker used
+      scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler # The specific scehduler used
+      gpu_memory_utilization: 0.8 # The gpu memory allocation for the stage within a single chip
+      enforce_eager: true  # Now we only support eager mode
+      trust_remote_code: true # Needed by huggingface config parsing
+      engine_output_type: latent  # It claims that the stage will input latent hiddenstates besides token ids
+      enable_prefix_caching: false # For request with hiddenstates output, the prefix caching is not supported now
+    is_comprehension: true # If the stage is a text or multimodal comprehension module. If it is, the AsyncOmniLLM will use its tokenizer as default
+    final_output: true # If the stage has output as part of final outputs. If it is false, which means that the stage only works as a intermediate role.
+    final_output_type: text # What is the final output type. It can be text and audio now.
+    default_sampling_params: # sampling parameters for the stage. Their meaning aligns with vLLM.
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+  - stage_id: 1
+    runtime:
+      process: true
+      devices: "1"
+      max_batch_size: 3
+    engine_args:
+      model_stage: talker
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler
+      gpu_memory_utilization: 0.8
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: latent
+    engine_input_source: [0]
+    custom_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen2_5_omni.thinker2talker
+    default_sampling_params:
+      temperature: 0.9
+      top_p: 0.8
+      top_k: 40
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.05
+      stop_token_ids: [8294]
+  - stage_id: 2
+    runtime:
+      process: true
+      devices: "0"            # Example: use a different GPU than the previous stage; use "0" if single GPU
+      max_batch_size: 1
+    engine_args:
+      model_stage: code2wav
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker
+      scheduler_cls: vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler
+      gpu_memory_utilization: 0.15
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: audio
+    engine_input_source: [1]
+    final_output: true
+    final_output_type: audio
+    default_sampling_params:
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+
+# Top-level runtime config (concise): default windows and stage edges
+runtime:
+  enabled: true
+  defaults:
+    window_size: -1             # Simplified: trigger downstream only after full upstream completion
+    max_inflight: 1             # Simplified: process serially within each stage
+  edges:
+    - from: 0                   # thinker → talker: trigger only after receiving full input (-1)
+      to: 1
+      window_size: -1
+    - from: 1                   # talker → code2wav: trigger only after receiving full input (-1)
+      to: 2
+      window_size: -1
+
+```
+
+## Stage Configuration Arguments
+
+Each stage in the `stage_args` list contains the following configuration options:
+
+### `stage_id`
+
+A unique identifier for each stage in the multi-stage pipeline. Stages are numbered sequentially starting from 0, and this ID is used to reference stages in inter-stage dependencies (e.g., `engine_input_source`).
+
+### `runtime`
+
+Configuration for disaggregated execution of the stage, controlling how the stage is deployed and executed.
+
+#### `runtime.process`
+
+Whether to run this stage in a separate process. When set to `true`, the stage will be executed in an isolated process, enabling better resource isolation and parallel execution across different stages. This is essential for multi-GPU deployments where different stages run on different devices.
+
+Default: `true`
+
+#### `runtime.devices`
+
+Visible devices for this stage, specified as a string. This controls which GPU devices are available to the stage process, similar to setting `CUDA_VISIBLE_DEVICES` or using `torch.cuda.set_device()`. For example, `"0"` uses GPU 0, `"1"` uses GPU 1, and `"0,1"` makes both GPUs 0 and 1 visible.
+
+Default: `"0"`
+
+#### `runtime.max_batch_size`
+
+The maximum batch size for offline inference in this stage. This limits how many sequences can be processed together in a single batch during offline inference operations.
+
+Default: `1`
+
+### `engine_args`
+
+Engine arguments for configuring the LLM engine, diffusion engine, or other engine types used by this stage.
+
+#### `engine_args.model_stage`
+
+The name identifier for this model stage within the multi-stage architecture. This is used internally to distinguish different stages of the same model (e.g., "thinker", "talker", "code2wav" in Qwen2.5-Omni).
+
+#### `engine_args.model_arch`
+
+The model architecture class name that is registered in `model_executor/models/registry.py`. This specifies which model implementation to use for this stage. The class must be registered in the model registry for vLLM-Omni to locate and instantiate it.
+
+#### `engine_args.worker_cls`
+
+The specific worker class to use for this stage. This determines how the model computations are executed. Examples include `vllm_omni.worker.gpu_ar_worker.GPUARWorker` for autoregressive stages and `vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker` for diffusion-based stages.
+
+#### `engine_args.scheduler_cls`
+
+The scheduler class to use for this stage. The scheduler manages request queuing, batching, and execution order. Examples include `vllm_omni.core.sched.scheduler.OmniScheduler` for standard stages and `vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler` for diffusion stages.
+
+#### `engine_args.gpu_memory_utilization`
+
+The fraction of GPU memory to allocate for this stage within a single GPU chip. This is a value between 0.0 and 1.0, where 0.8 means 80% of the GPU memory will be used by this stage. This allows fine-grained control over memory allocation when multiple stages share the same GPU or when reserving memory for other operations.
+
+Default: `0.8`
+
+#### `engine_args.enforce_eager`
+
+Whether to enforce eager execution mode. When set to `true`, the engine will run in eager mode without using CUDA graphs or other compilation optimizations. Currently, vLLM-Omni only supports eager mode.
+
+Default: `true`
+
+#### `engine_args.trust_remote_code`
+
+Whether to trust remote code when loading models from Hugging Face. This is required for models that use custom code in their configuration files. Set to `true` when loading models that require custom model implementations.
+
+Default: `true`
+
+#### `engine_args.engine_output_type`
+
+Specifies the type of output produced by this stage's engine. This determines what kind of data flows to downstream stages. Possible values include `latent` (hidden states), `text` (tokenized text), and `audio` (audio waveforms). When set to `latent`, the stage outputs latent hidden states in addition to token IDs, which are consumed by downstream stages.
+
+Default: `latent`
+
+#### `engine_args.enable_prefix_caching`
+
+Whether to enable prefix caching for this stage. Prefix caching can improve performance by caching KV cache for common prompt prefixes. However, for requests that output hidden states (when `engine_output_type` is `latent`), prefix caching is not currently supported and should be set to `false`.
+
+Default: `false`
+
+### `is_comprehension`
+
+Whether this stage is a text or multimodal comprehension module. When set to `true`, the stage acts as a comprehension module that processes input text or multimodal content. If this is the first comprehension stage, `AsyncOmniLLM` will use its tokenizer as the default tokenizer for the entire pipeline.
+
+Default: `true`
+
+### `final_output`
+
+Whether this stage produces output that is part of the final outputs returned to the user. When set to `false`, the stage only works as an intermediate stage, processing data that flows to downstream stages but not contributing directly to the final response.
+
+Default: `true`
+
+### `final_output_type`
+
+The type of final output produced by this stage. This specifies what format the output will be in when returned to the user. Currently supported values are `text` (for text generation) and `audio` (for audio generation).
+
+Default: `text`
+
+### `default_sampling_params`
+
+Default sampling parameters for this stage. These parameters control the generation behavior and align with vLLM's sampling parameter semantics. These defaults are used when no explicit sampling parameters are provided in the request.
+
+#### `default_sampling_params.temperature`
+
+Sampling temperature for controlling randomness. Lower values (e.g., 0.0) make the output more deterministic and focused, while higher values increase randomness.
+
+Default: `0.0`
+
+#### `default_sampling_params.top_p`
+
+Nucleus sampling parameter. Only tokens with cumulative probability mass up to `top_p` are considered. This helps filter out low-probability tokens while maintaining diversity.
+
+Default: `1.0`
+
+#### `default_sampling_params.top_k`
+
+Top-k sampling parameter. Only the top `k` most likely tokens are considered. Set to `-1` to disable top-k filtering and consider all tokens.
+
+Default: `-1`
+
+#### `default_sampling_params.max_tokens`
+
+Maximum number of tokens to generate in this stage. This limits the length of the output sequence.
+
+Default: `2048`
+
+#### `default_sampling_params.seed`
+
+Random seed for reproducible generation. When set, the random number generator will be initialized with this seed to ensure consistent outputs across runs.
+
+Default: `42`
+
+#### `default_sampling_params.detokenize`
+
+Whether to detokenize the output tokens into text. When set to `true`, token IDs are converted back to readable text strings.
+
+Default: `True`
+
+#### `default_sampling_params.repetition_penalty`
+
+Penalty applied to tokens that have already appeared in the generated sequence. Values greater than 1.0 discourage repetition, while values less than 1.0 encourage it. A value of 1.0 applies no penalty.
+
+Default: `1.1`

docs/configuration/stage_configs/qwen2_5_omni.yaml

@@ -0,0 +1,94 @@
+# stage config for running qwen2.5-omni with architecture of OmniLLM.
+stage_args:
+  - stage_id: 0
+    runtime:
+      process: true            # Run this stage in a separate process
+      devices: "0"            # Visible devices for this stage (CUDA_VISIBLE_DEVICES/torch.cuda.set_device)
+      max_batch_size: 1
+    engine_args:
+      model_stage: thinker
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler
+      gpu_memory_utilization: 0.8
+      enforce_eager: true  # Now we only support eager mode
+      trust_remote_code: true
+      engine_output_type: latent
+      enable_prefix_caching: false
+    is_comprehension: true
+    final_output: true
+    final_output_type: text
+    default_sampling_params:
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+  - stage_id: 1
+    runtime:
+      process: true
+      devices: "1"
+      max_batch_size: 1
+    engine_args:
+      model_stage: talker
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler
+      gpu_memory_utilization: 0.8
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: latent
+    engine_input_source: [0]
+    custom_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen2_5_omni.thinker2talker
+    default_sampling_params:
+      temperature: 0.9
+      top_p: 0.8
+      top_k: 40
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.05
+      stop_token_ids: [8294]
+  - stage_id: 2
+    runtime:
+      process: true
+      devices: "0"            # Example: use a different GPU than the previous stage; use "0" if single GPU
+      max_batch_size: 1
+    engine_args:
+      model_stage: code2wav
+      model_arch: Qwen2_5OmniForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker
+      scheduler_cls: vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler
+      gpu_memory_utilization: 0.15
+      enforce_eager: true
+      trust_remote_code: true
+      enable_prefix_caching: false
+      engine_output_type: audio
+    engine_input_source: [1]
+    final_output: true
+    final_output_type: audio
+    default_sampling_params:
+      temperature: 0.0
+      top_p: 1.0
+      top_k: -1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.1
+
+# Top-level runtime config (concise): default windows and stage edges
+runtime:
+  enabled: true
+  defaults:
+    window_size: -1             # Simplified: trigger downstream only after full upstream completion
+    max_inflight: 1             # Simplified: process serially within each stage
+  edges:
+    - from: 0                   # thinker → talker: trigger only after receiving full input (-1)
+      to: 1
+      window_size: -1
+    - from: 1                   # talker → code2wav: trigger only after receiving full input (-1)
+      to: 2
+      window_size: -1

docs/contributing/DOCS_GUIDE.md

@@ -0,0 +1,139 @@
+# Documentation Build Guide
+
+This directory contains the source files for the vLLM-Omni documentation.
+
+## Building Documentation Locally
+
+### Prerequisites
+
+Install documentation dependencies:
+
+```bash
+pip install -e ".[docs]"
+```
+
+### Build and Serve Documentation
+
+From the project root:
+
+```bash
+# Serve documentation locally (auto-reload on changes)
+# This starts a local web server at http://127.0.0.1:8000
+mkdocs serve
+
+# Build static site (generates HTML files in site/ directory)
+mkdocs build
+```
+
+When using `mkdocs serve`, the documentation will be automatically available at `http://127.0.0.1:8000`. The server will automatically reload when you make changes to the documentation files.
+
+## Auto-generating API Documentation
+
+The documentation automatically extracts docstrings from the code using mkdocstrings. To ensure your code is documented:
+
+1. Add docstrings to all public classes, functions, and methods
+2. Use Google or NumPy style docstrings (both are supported)
+3. Rebuild the documentation to see changes
+
+Example docstring:
+
+```python
+class Omni:
+    """Main entry point for vLLM-Omni inference.
+
+    This class provides a high-level interface for running multi-modal
+    inference with non-autoregressive models.
+
+    Args:
+        model: Model name or path
+        stage_configs: Optional stage configurations
+        **kwargs: Additional arguments passed to the engine
+
+    Example:
+        >>> llm = Omni(model="Qwen/Qwen2.5-Omni")
+        >>> outputs = llm.generate(prompts="Hello")
+    """
+```
+
+## Documentation Structure
+
+```
+docs/
+├── index.md              # Main documentation page
+├── getting_started/      # Getting started guides
+├── architecture/        # Architecture documentation
+├── api/                 # API reference (auto-generated from code)
+├── examples/            # Code examples
+└── stylesheets/         # Custom CSS
+```
+
+## Publishing Documentation
+
+### GitHub Pages (Recommended)
+
+The documentation is automatically deployed to GitHub Pages using GitHub Actions.
+
+1. **Enable GitHub Pages**:
+   - Go to repository `Settings` → `Pages`
+   - Set `Source` to `GitHub Actions`
+   - Save settings
+
+2. **Push changes**:
+   ```bash
+   git push origin main
+   ```
+
+3. **Documentation will be available at**:
+   - `https://vllm-project.github.io/vllm-omni/`
+
+The GitHub Actions workflow (`.github/workflows/docs.yml`) will automatically:
+- Build the documentation when you push to `main` branch
+- Deploy it to GitHub Pages
+- Update the documentation whenever you make changes
+
+
+### Read the Docs (Alternative)
+
+You can also use Read the Docs for hosting:
+
+1. Sign up at https://readthedocs.org/
+2. Import the `vllm-project/vllm-omni` repository
+3. Read the Docs will automatically build using `.readthedocs.yml`
+4. Documentation will be available at: `https://vllm-omni.readthedocs.io/`
+
+## Configuration
+
+The documentation configuration is in `mkdocs.yml` at the project root.
+
+## Tips
+
+- **API Documentation**: API docs are automatically generated using `mkdocs-api-autonav` and `mkdocstrings`
+  - No need to manually create API pages - they're generated automatically
+  - Use `[module.name.ClassName][]` syntax for cross-references in Summary pages
+- **Code Snippets**: Use `--8<-- "path/to/file.py"` for including code snippets
+- **Markdown**: Use Markdown for all documentation (no need for RST)
+- **Material Theme**: Use Material theme features like:
+  - Admonitions: `!!! note`, `!!! warning`, etc.
+  - Code blocks with syntax highlighting
+  - Tabs for organizing content
+  - Math formulas using `pymdownx.arithmatex`
+
+## Troubleshooting
+
+### Documentation not updating
+
+- Make sure you've saved all files
+- If using `mkdocs serve`, it should auto-reload
+- Check for syntax errors in `mkdocs.yml`
+
+### API links not working
+
+- Ensure class names match exactly (case-sensitive)
+- Check that the module is imported correctly
+- Run `mkdocs build --strict` to check for errors
+
+### Build errors
+
+- Check Python version (requires 3.9+)
+- Ensure all dependencies are installed: `pip install -e ".[docs]"`
+- Check `mkdocs.yml` syntax with `mkdocs build --strict`

docs/contributing/README.md

@@ -0,0 +1,157 @@
+# Contributing to vLLM-Omni
+
+Thank you for your interest in contributing to vLLM-Omni! This document provides guidelines and instructions for contributing.
+
+## Getting Started
+
+vLLM-Omni uses `uv` as the environment manager, to create and manage Python environments. Please follow the documentation to install `uv`. After installing `uv`, you can create a new Python environment using the following commands:
+
+```bash
+uv venv --python 3.12 --seed
+source .venv/bin/activate
+```
+### Development Environment for vLLM
+
+vLLM-Omni is built on top of the stable release 0.11.0 of vLLM. You may install `vllm==0.11.0` directly or install the library from source if you need to check, modify or debug with source code of vLLM.
+
+### Development Environment for vLLM-Omni
+
+Install vLLM-Omni from source with development dependencies:
+
+```bash
+git clone https://github.com/vllm-project/vllm-omni.git
+cd vllm_omni
+uv pip install -e ".[dev]"
+```
+
+!!! tip
+    vLLM-Omni is compatible with Python versions 3.10 to 3.12. However, we recommend developing with Python 3.12 to minimize the chance of your local environment clashing with our CI environment.
+
+### Linting
+
+vLLM-Omni uses `pre-commit` to lint and format the codebase. See [pre-commit documentation](https://pre-commit.com/#usage) if `pre-commit` is new to you. Setting up `pre-commit` is as easy as:
+
+```bash
+uv pip install pre-commit
+pre-commit install
+```
+
+vLLM-Omni's `pre-commit` hooks will now run automatically every time you commit.
+
+!!! tip
+    You can manually run the `pre-commit` hooks using:
+
+    ```bash
+    pre-commit run     # runs on staged files
+    pre-commit run --show-diff-on-failure --color=always --all-files  # runs on all files (short for --all-files)
+    ```
+
+### Documentation
+
+MkDocs is a fast, simple and downright gorgeous static site generator that's geared towards building project documentation. Documentation source files are written in Markdown, and configured with a single YAML configuration file, `mkdocs.yml`.
+
+Get started with:
+
+```bash
+uv pip install -e ".[docs]"
+```
+
+MkDocs comes with a built-in dev-server that lets you preview your documentation as you work on it. From the root of the repository, run:
+
+```bash
+mkdocs serve                           # with API ref (~10 minutes)
+API_AUTONAV_EXCLUDE=vllm_omni mkdocs serve  # API ref off (~15 seconds)
+```
+
+Once you see `Serving on http://127.0.0.1:8000/` in the logs, the live preview is ready! Open <http://127.0.0.1:8000/> in your browser to see it.
+
+For additional features and advanced configurations, refer to the:
+
+- [MkDocs documentation](https://www.mkdocs.org/)
+- [Material for MkDocs documentation](https://squidfunk.github.io/mkdocs-material/) (the MkDocs theme we use)
+
+### Testing
+
+vLLM-Omni uses `pytest` to test the codebase.
+
+```bash
+# Run all tests
+pytest tests/
+
+# Run tests for a single test file with detailed output
+pytest -s -v tests/test_omni_llm.py
+```
+
+!!! warning
+    Currently, not all unit tests pass when run on CPU platforms. If you don't have access to a GPU platform to run unit tests locally, rely on the continuous integration system to run the tests for now.
+
+## Issues
+
+If you encounter a bug or have a feature request, please search existing issues first to see if it has already been reported. If not, please file a new issue, providing as much relevant information as possible.
+
+!!! important
+    If you discover a security vulnerability, please report it by creating a GitHub issue with the `security` label.
+
+## Pull Requests & Code Reviews
+
+Thank you for your contribution to vLLM-Omni! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM-Omni maintain the code quality and improve the efficiency of the review process.
+
+### DCO and Signed-off-by
+
+When contributing changes to this project, you must agree to the [DCO](https://developercertificate.org/). Commits must include a `Signed-off-by:` header which certifies agreement with the terms of the DCO.
+
+Using `-s` with `git commit` will automatically add this header.
+
+!!! tip
+    You can enable automatic sign-off via your IDE:
+
+    - **PyCharm**: Click on the `Show Commit Options` icon to the right of the `Commit and Push...` button in the `Commit` window. It will bring up a `git` window where you can modify the `Author` and enable `Sign-off commit`.
+    - **VSCode**: Open the Settings editor and enable the `Git: Always Sign Off` (`git.alwaysSignOff`) field.
+
+### PR Title and Classification
+
+Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:
+
+- `[Bugfix]` for bug fixes.
+- `[CI/Build]` for build or continuous integration improvements.
+- `[Doc]` for documentation fixes and improvements.
+- `[Model]` for adding a new model or improving an existing model. Model name should appear in the title.
+- `[Frontend]` For changes on the vLLM-Omni frontend (e.g., OpenAI API server, `OmniLLM` class, etc.)
+- `[Kernel]` for changes affecting CUDA kernels or other compute kernels.
+- `[Core]` for changes in the core vLLM-Omni logic (e.g., `OmniProcessor`, `OmniScheduler`, etc.)
+- `[Hardware][Vendor]` for hardware-specific changes. Vendor name should appear in the prefix, such as [Ascend] for Ascend NPUs.
+- `[Misc]` for PRs that do not fit the above categories. Please use this sparingly.
+
+!!! note
+    If the PR spans more than one category, please include all relevant prefixes.
+
+### Code Quality
+
+The PR needs to meet the following code quality standards:
+
+- We adhere to Google Python style guide and Google C++ style guide.
+- Pass all linter checks.
+- The code needs to be well-documented to ensure future contributors can easily understand the code.
+- Include sufficient tests to ensure the project stays correct and robust. This includes both unit tests and integration tests.
+- Please add documentation to `docs/` if the PR modifies the user-facing behaviors of vLLM-Omni. It helps vLLM-Omni users understand and utilize the new features or changes.
+
+### Notes for Large Changes
+
+Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with `rfc-required` and might not go through the PR.
+
+### What to Expect for the Reviews
+
+The goal of the vLLM-Omni team is to be a _transparent reviewing machine_. We would like to make the review process transparent and efficient and make sure no contributor feels confused or frustrated. However, the vLLM-Omni team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:
+
+- After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
+- After the PR is assigned, the reviewer will provide status updates every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM-Omni team.
+- After the review, the reviewer will put an `action-required` label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
+- Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.
+
+## Additional Resources
+
+- [Design Documents](../design/index.md) - Architecture and design documentation
+
+## Thank You
+
+Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM-Omni. All of your contributions help make vLLM-Omni a great tool and community for everyone!

docs/contributing/ci/failures.md

@@ -0,0 +1,4 @@
+# CI Failures
+
+What should I do when a CI job fails on my PR, but I don't think my PR caused
+the failure?

docs/contributing/model/README.md

@@ -0,0 +1,2 @@
+# Adding a New Model
+Stay tuned for guidance on how to add a new model on vLLM-Omni!

docs/design/api_design_doc.md

@@ -0,0 +1,143 @@
+# API Design Documentation
+
+This document contains comprehensive API design documentation for all core modules in vLLM-Omni. These templates provide a standardized structure for designing and implementing the core, engine, executor, and worker modules.
+
+## 📋 Module API
+
+### Core Module API
+**Core module** provides fundamental scheduling, caching, and resource management functionality.
+
+**Key Components:**
+- Request scheduling and prioritization
+- DiT cache management for diffusion models
+- Resource allocation and coordination
+- Inter-module communication
+
+### Engine Module API
+**Engine module** handles model loading, inference execution, and output processing.
+
+**Key Components:**
+- Model loading and initialization
+- Inference execution for AR and diffusion models
+- Input preprocessing for various modalities
+- Output postprocessing and formatting
+
+### Executor Module API
+**Executor module** coordinates and manages request execution across different engines and workers.
+
+**Key Components:**
+- Request routing and load balancing
+- Execution pipeline coordination
+- Worker management and task distribution
+- Error handling and recovery
+
+### Worker Module API
+**Worker module** provides the actual execution environment for model inference.
+
+**Key Components:**
+- Model execution and inference
+- GPU resource management
+- Request batching and processing
+- Performance optimization
+
+
+## 🔧 Implementation Guidelines
+
+### Code Standards
+- Use type hints for all methods
+- Include comprehensive docstrings
+- Follow PEP 8 style guidelines
+- Use async/await for I/O operations
+
+### Error Handling
+- Define specific exception types
+- Provide meaningful error messages
+- Include error recovery strategies
+- Log errors appropriately
+
+### Configuration
+- Use dataclasses for configuration
+- Provide sensible defaults
+- Include validation methods
+- Support environment variables
+
+### Testing
+- Write unit tests for all public methods
+- Include integration tests
+- Test error conditions
+- Validate configuration options
+
+## 🚀 Getting Started
+
+### For Developers
+
+1. **Choose a Module**
+   - Pick the module you want to work on
+   - Read the corresponding API documentation
+   - Understand the responsibilities and interfaces
+
+2. **Set Up Development Environment**
+   ```bash
+   # Clone the repository
+   git clone https://github.com/vllm-project/vllm-omni.git
+   cd vllm-omni
+
+   # Install dependencies
+   pip install -e ".[dev]"
+
+   # Set up pre-commit hooks
+   pre-commit install
+   ```
+
+3. **Start Implementation**
+   - Create the module directory
+   - Implement base classes first
+   - Add concrete implementations
+   - Write tests as you go
+
+### For Contributors
+
+1. **Review API Documentation**
+   - Read through all module APIs
+   - Understand the overall architecture
+   - Identify areas for improvement
+
+2. **Propose Changes**
+   - Create issues for API changes
+   - Discuss in pull requests
+   - Update documentation accordingly
+
+3. **Submit Contributions**
+   - Follow the coding standards
+   - Include tests for new features
+   - Update documentation
+   - Submit pull requests
+
+## 🤝 Contributing
+
+We welcome contributions to improve the API documentation and implementation. Please:
+
+1. **Follow the Template Structure**
+   - Use the [api template](https://github.com/vllm-project/vllm-omni/docs/contributing/design_documents/api_design_template.md) as a guide
+   - Maintain consistency across modules
+   - Include all required sections
+
+2. **Provide Working Examples**
+   - Test all code examples
+   - Include both basic and advanced usage
+   - Show integration patterns
+
+3. **Keep Documentation Updated**
+   - Update docs when APIs change
+   - Version control documentation changes
+   - Maintain backward compatibility notes
+
+## 📝 Notes
+
+- All API documentation should be kept up-to-date with implementation
+- Code examples should be tested and working
+- Configuration options should be validated
+- Error handling should be comprehensive
+- Performance considerations should be documented
+
+For questions or suggestions about the API documentation, please open an issue or start a discussion in the repository.

docs/design/api_design_template.md

@@ -0,0 +1,336 @@
+# API Design Template for vLLM-Omni Modules
+
+This template provides a standardized structure for designing APIs for core, engine, executor, and worker modules in vLLM-Omni. Use this template to ensure consistency and completeness across all modules.
+
+## Template Structure
+
+### 1. Module Overview
+- **Purpose**: What this module does
+- **Responsibilities**: Key responsibilities of the module
+- **Dependencies**: What other modules this depends on
+- **Integration Points**: How it integrates with other modules
+
+### 2. Core Classes/Interfaces
+- **Base Classes**: Abstract base classes and interfaces
+- **Implementation Classes**: Concrete implementations
+- **Data Structures**: Key data structures and models
+
+### 3. Public API Methods
+- **Initialization**: Constructor and setup methods
+- **Core Operations**: Main functionality methods
+- **Configuration**: Configuration and parameter methods
+- **Lifecycle Management**: Start, stop, cleanup methods
+- **Monitoring**: Status, metrics, and debugging methods
+
+### 4. Configuration
+- **Configuration Classes**: Dataclasses or config objects
+- **Required Parameters**: Must-have configuration
+- **Optional Parameters**: Optional configuration with defaults
+- **Validation**: Parameter validation rules
+
+### 5. Error Handling
+- **Custom Exceptions**: Module-specific exceptions
+- **Error Codes**: Standardized error codes
+- **Recovery Strategies**: How to handle and recover from errors
+
+### 6. Examples
+- **Basic Usage**: Simple usage examples
+- **Advanced Usage**: Complex scenarios
+- **Integration Examples**: How to use with other modules
+
+---
+
+## Example: Core Module Template
+
+### 1. Module Overview
+
+**Purpose**: The core module provides fundamental scheduling and caching functionality for vLLM-Omni.
+
+**Responsibilities**:
+- Request scheduling and prioritization
+- DiT cache management
+- Resource allocation and coordination
+- Inter-module communication
+
+**Dependencies**:
+- `vllm_omni.request` - Request handling
+- `vllm_omni.config` - Configuration management
+- `vllm_omni.utils` - Utility functions
+
+**Integration Points**:
+- Receives requests from entrypoints
+- Coordinates with engine modules
+- Manages worker allocation
+- Provides status to monitoring systems
+
+### 2. Core Classes/Interfaces
+
+```python
+from abc import ABC, abstractmethod
+from typing import List, Optional, Dict, Any
+from dataclasses import dataclass
+from enum import Enum
+
+class DiTCacheManager:
+    """Manages DiT cache for diffusion models."""
+
+    def __init__(self, config: DiTCacheConfig):
+        self.config = config
+        self.cache = {}
+        self.cache_stats = {}
+
+    async def get_cache(self, cache_key: str) -> Optional[Any]:
+        """Retrieve cached data."""
+        pass
+
+    async def set_cache(self, cache_key: str, data: Any) -> None:
+        """Store data in cache."""
+        pass
+
+    async def invalidate_cache(self, cache_key: str) -> None:
+        """Invalidate cached data."""
+        pass
+```
+
+### 3. Public API Methods
+
+#### Initialization
+```python
+class OmniScheduler():
+    def __init__(self, config: SchedulerConfig):
+        """Initialize the core scheduler."""
+        self.config = config
+        self.scheduler = self._create_scheduler()
+        self.cache_manager = DiTCacheManager(config.dit_cache_config)
+        self._running = False
+
+    def _create_scheduler(self) -> BaseScheduler:
+        """Factory method to create appropriate scheduler."""
+        if self.config.scheduler_type == SchedulerType.FIFO:
+            return FIFOScheduler(self.config)
+        elif self.config.scheduler_type == SchedulerType.PRIORITY:
+            return PriorityScheduler(self.config)
+        # ... other scheduler types
+```
+
+#### Core Operations
+```python
+    async def schedule(self, request: OmniRequest) -> bool:
+        """
+        Schedule a request for processing.
+
+        Args:
+            request: The request to schedule
+
+        Returns:
+            bool: True if successfully scheduled, False otherwise
+
+        Raises:
+            SchedulerError: If scheduling fails
+            QueueFullError: If queue is at capacity
+        """
+        pass
+```
+
+#### Configuration
+```python
+    def update_config(self, new_config: SchedulerConfig) -> None:
+        """Update scheduler configuration."""
+        pass
+
+    def get_config(self) -> SchedulerConfig:
+        """Get current configuration."""
+        pass
+```
+
+#### Lifecycle Management
+```python
+    async def start(self) -> None:
+        """Start the scheduler."""
+        self._running = True
+        # Start background tasks
+
+    async def stop(self) -> None:
+        """Stop the scheduler gracefully."""
+        self._running = False
+        # Cleanup resources
+
+    async def shutdown(self) -> None:
+        """Force shutdown the scheduler."""
+        # Immediate cleanup
+```
+
+#### Monitoring
+```python
+    def get_status(self) -> Dict[str, Any]:
+        """Get current scheduler status."""
+        return {
+            "running": self._running,
+            "queue_size": self.scheduler.queue_size(),
+            "processed_requests": self.scheduler.processed_count(),
+            "cache_hit_rate": self.cache_manager.get_hit_rate()
+        }
+
+    def get_metrics(self) -> Dict[str, float]:
+        """Get performance metrics."""
+        pass
+```
+
+### 4. Configuration
+
+```python
+@dataclass
+class OmniConfig:
+    """Configuration for the core module."""
+
+    # Scheduler configuration
+    scheduler: SchedulerConfig = field(default_factory=SchedulerConfig)
+
+    # Cache configuration
+    dit_cache: DiTCacheConfig = field(default_factory=DiTCacheConfig)
+
+    # Resource limits
+    max_memory_gb: float = 16.0
+    max_gpu_utilization: float = 0.8
+
+    # Timeouts
+    request_timeout: int = 300
+    worker_timeout: int = 60
+
+    def validate(self) -> None:
+        """Validate configuration parameters."""
+        if self.max_memory_gb <= 0:
+            raise ValueError("max_memory_gb must be positive")
+        if not 0 < self.max_gpu_utilization <= 1:
+            raise ValueError("max_gpu_utilization must be between 0 and 1")
+```
+
+### 5. Error Handling
+
+```python
+class CoreModuleError(Exception):
+    """Base exception for core module errors."""
+    pass
+
+class SchedulerError(CoreModuleError):
+    """Scheduler-related errors."""
+    pass
+
+class QueueFullError(SchedulerError):
+    """Queue is at capacity."""
+    pass
+
+class CacheError(CoreModuleError):
+    """Cache-related errors."""
+    pass
+
+class ResourceError(CoreModuleError):
+    """Resource allocation errors."""
+    pass
+```
+
+### 6. Examples
+
+#### Basic Usage
+```python
+from vllm_omni.core import CoreScheduler, CoreModuleConfig
+from vllm_omni.request import create_text_request
+
+# Create configuration
+config = CoreModuleConfig(
+    scheduler=SchedulerConfig(scheduler_type=SchedulerType.FIFO),
+    max_memory_gb=8.0
+)
+
+# Initialize scheduler
+scheduler = CoreScheduler(config)
+await scheduler.start()
+
+# Create and schedule a request
+request = create_text_request(
+    request_id="req_001",
+    prompt="Hello, world!",
+    sampling_params=sampling_params
+)
+
+success = await scheduler.schedule_request(request)
+if success:
+    result = await scheduler.process_request(request)
+    print(f"Result: {result}")
+```
+
+#### Advanced Usage
+```python
+# Custom scheduler with priority handling
+config = CoreModuleConfig(
+    scheduler=SchedulerConfig(
+        scheduler_type=SchedulerType.PRIORITY,
+        priority_weights={"high": 2.0, "normal": 1.0, "low": 0.5}
+    )
+)
+
+scheduler = CoreScheduler(config)
+await scheduler.start()
+
+# Monitor scheduler status
+status = scheduler.get_status()
+print(f"Queue size: {status['queue_size']}")
+print(f"Cache hit rate: {status['cache_hit_rate']:.2%}")
+```
+
+---
+
+## Module-Specific Guidelines
+
+### Core Module
+- Focus on scheduling, caching, and resource management
+- Provide clear interfaces for other modules
+- Handle concurrency and thread safety
+- Implement comprehensive monitoring
+
+### Engine Module
+- Handle model loading and inference
+- Support both AR and diffusion models
+- Provide unified interface for different model types
+- Implement efficient memory management
+
+### Executor Module
+- Coordinate between different engines
+- Handle request routing and load balancing
+- Manage execution pipelines
+- Provide error recovery mechanisms
+
+### Worker Module
+- Handle actual model execution
+- Manage GPU resources
+- Implement batching strategies
+- Provide performance optimization
+
+---
+
+## Checklist for API Design
+
+- [ ] **Clear Purpose**: Module purpose is well-defined
+- [ ] **Complete Interface**: All public methods are documented
+- [ ] **Error Handling**: Comprehensive error handling strategy
+- [ ] **Configuration**: Flexible configuration system
+- [ ] **Examples**: Basic and advanced usage examples
+- [ ] **Type Hints**: All methods have proper type hints
+- [ ] **Documentation**: Comprehensive docstrings
+- [ ] **Testing**: Unit tests for all public methods
+- [ ] **Integration**: Clear integration points with other modules
+- [ ] **Performance**: Performance considerations documented
+
+---
+
+## Submission Guidelines
+
+1. Create a new file: `docs/api/[module_name]_api.md`
+2. Follow the template structure exactly
+3. Include all required sections
+4. Provide working code examples
+5. Ensure all methods have proper docstrings
+6. Include error handling strategies
+7. Submit for review before implementation
+
+This template ensures consistency and completeness across all vLLM-Omni modules.

docs/design/architecture_overview.md

@@ -0,0 +1,113 @@
+# Architecture Overview
+
+# Introduction
+
+This document outlines the architectural design for vLLM-Omni, a new repository developed under the vLLM project. vLLM-Omni supports multi-modality models with non-autoregressive structures and non-textual outputs, extending vLLM beyond traditional text-based, autoregressive generation.
+
+# Goals
+
+The primary goals for vLLM-Omni are:
+
+* **Support for Non-textual Output:** Enables the integration, efficient processing and output of various data types, including but not limited to, images, audio, and video, alongside text.  
+* **Non-Autoregressive Structure:** Support model structure beyond autoregressive, especially Diffusion Transformer (DiT), which is widely used in visual and audio generation.  
+* **Integration with vLLM Core:** Maintain compatibility and leverage existing vLLM key modules and optimizations where applicable.  
+* **Extensibility:** Design a modular and flexible architecture that can easily accommodate new modalities, model architectures, and output formats.
+
+# Expected Gains
+
+* It breaks through the limitation of vLLM that only supports autoregressive text generation models, and provides architectural support for popular model structures, especially multimodal generation.  
+* By supporting new output formats and new model architectures, developers can easily connect to mature ecosystems such as Higress Gateway and Ray, improving the ease of deployment and efficiency of corresponding models.  
+* Integrating efficient cache management, multiple parallel strategies, disaggregation from vLLM, to improve inference efficiency of corresponding models.
+
+# Non-Autoregressive Structure
+
+* **Diffusion Transformer:** initialized with a random latent noise, the transformer model iteratively predicts latent output for multiple steps, under control of sampling timestamps. Finally, it is decoded by the decoder of Variational AutoEncoder (VAE). It is not supported by vLLM now.  
+* **Autoregressive:** dominant generation paradigm for text, now supported by vLLM. It generates tokens conditioned on previous tokens one by one. The efficient KV Cache management provided by vLLM can accelerate inference effectively.  
+  ![][image1]                       ![][image2]
+
+
+|  | Autoregressive | DiT |
+| :---- | ----- | ----- |
+| Scenarios | Text generation | Multimodal generation |
+| Generation | Token-by-token, left to right | Continuous vector, iteratively refine |
+| Key advantages | Smooth, good local consistency | Global consistency, diversity and controlability |
+| Efficiency | More efficient with KV cache | Need generation repeatedly, slow |
+| Error accumulation | Easy to accumulate | Can fix possibly |
+| Key challenges | Limited global planning, long context generation | Large global computational cost |
+| Sequence length | Varied length | Fixed length |
+| Attention mask | Lower triangular | Full bi-directional matrix |
+| Parallel strategy | TP, DP, PP, .etc | DP, CP, .etc |
+| vLLM support | Supported | Unsupported |
+
+# Targeted models to support
+
+According to analysis for current popular open-source models, most of them have the combination of AR+DiT. Specifically, they can be further categorized into 3 types below:
+
+* **DiT as a main structure, with AR as text encoder (e.g.: Qwen-Image)**  
+  A popular model for image generation and editing. Main visual generation models are similar to the structure, such as Flux.  
+  ![][image3]  
+* **AR as a main structure, with DiT as multi-modal generator (e.g. BAGEL)**  
+  A unified multimodal comprehension and generation model. Visual generation can utilize CoT text generation output. Main unified multimodal models are similar to the structure.  
+      ![][image4]  
+* **Extension, multiple AR+DiT (e.g. Qwen-omni)**  
+  A new model for multimodal input and output. It designs the Thinker-Talker-Codec structure, which is a double-AR+DiT format.  
+  ![][image5]  
+
+
+
+# Architectural Overview
+
+vLLM-Omni will import core modules from existing vLLM framework and introduce several new components and modifications to achieve its goals. The core architecture will be based on a modular design, allowing for independent development and integration of different modality handlers and output generators.  
+![][image6]
+
+## General ideas
+
+* It divides models in vLLM-Omni into multiple stages, which are processed by different Engine Cores. Each stage can choose AR Engine Core for AR model part, and Diffusion Engine Core for diffusion model part.  
+* Designed a specialized diffusion engine core for DiT or dLLM.  
+* Plug-in with external DiT acceleration modules such as xDiT, Cache-DiT.  
+* Output linked with Higress API, Gradio for online demo, ComfyUI for less-code process.
+
+## Key Components
+
+* **Processor:** Input processing, tokenization or encoding  
+* **Scheduler:** for order and queue of incoming requests  
+* **Executor:**  Prepare model input and run  
+* **Multimodal output processor:** support multimodal outputs  
+* **Hidden State Output processor:** support intermediate hidden vectors as request output  
+* **Engine Core:** accelerate AR modules, imported from vLLM main repo  
+* **Diffusion Engine Core:** Wrap and accelerate DiT modules.  
+* **Diffusion model:** Wrap up diffusion models with vLLM format
+
+# Examples data flow of specific models
+
+1. Qwen-omni  
+   ![][image7]![][image8]  
+2. Qwen-Image  
+   ![][image9]      ![][image10]
+
+# Overview of function call dependency
+
+![][image11]
+
+
+[image1]: <data:image/png;base64,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 truncated
+
+[image2]: <data:image/png;base64,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 truncated
+
+[image3]: <data:image/png;base64,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 truncated
+
+[image4]: <data:image/png;base64,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 truncated
+
+[image5]: <data:image/png;base64,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 truncated
+
+[image6]: <data:image/png;base64,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 truncated
+
+[image7]: <data:image/png;base64,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 truncated
+
+[image8]: <data:image/png;base64,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 truncated
+
+[image9]: <data:image/png;base64,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 truncated
+
+[image10]: <data:image/png;base64,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 truncated
+
+[image11]: <data:image/png;base64,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 truncated

docs/design/index.md

@@ -0,0 +1,13 @@
+# Design Documents
+
+This section contains design documents and architecture specifications for vLLM-Omni.
+
+## Architecture Documents
+
+- [Architecture Overview](architecture_overview.md)
+- [vLLM-Omni Software Design Document](vllm_omni_design.md)
+- [Multi-Request Streaming](mrs_design.md)
+
+## API Design Documentation
+
+- [vLLM-Omni API Design Documentation](api_design_doc.md)

docs/design/mrs_design.md

@@ -0,0 +1,188 @@
+## Multi-Request Streaming (MRS) on a Single Machine
+
+### 1. Background & Scope
+- All processing runs on a single physical machine with multi-process, per-stage workers. No proxy or network transport involved.
+- Current alignment with vllm-omni: `OmniLLM` supports multiple stages (`OmniStage`). GPU runners already expose streamable steps (prefill/decoding/diffusion), but the entry layer still collects lists and lacks intra-stage streaming and window scheduling.
+- Goal: implement multi-stage, multi-request streaming (MRS) locally. Each stage outputs segments; downstream stages stitch and trigger compute based on configured windows. Shared memory and zero-copy strategies reduce data movement overhead.
+
+### 2. Key Constraints
+- Multi-process per stage: each stage is an independent process with a while loop; device visibility can be configured (`CUDA_VISIBLE_DEVICES`/`torch.cuda.set_device`).
+- Simple IPC (copy-based): use `multiprocessing.Queue`/Pipe for inter-process communication with CPU copies/serialization; do not rely on CUDA IPC/SHM zero-copy in this version.
+- Cross-stage pipeline: different stages can process different requests concurrently (e.g., stage A handles request 1 while stage B handles request 0).
+
+### 3. Architecture Overview
+- Processes & IPC queues
+  - Each "sub-stage" is an OS process (worker). The loop: take from input_queue → compute → put to output_queue.
+  - Inter-stage connection via IPC: copy-based `multiprocessing.Queue` passing dict payloads; use shared memory for large objects.
+  - Each link is SPSC (single-producer/single-consumer): the upstream is the orchestrator and the downstream is a single stage process; queues are unbounded (maxsize=0) on the orchestrator side.
+- Device visibility
+  - Each stage sets `CUDA_VISIBLE_DEVICES` or calls `torch.cuda.set_device` to bind to GPU sets.
+  - A stage may use multiple GPUs internally (TP/PP/DP) but presents as a single stage unit.
+- Simplified IPC: copy-based queues/pipes for data transfer; zero-copy is future work.
+- Pipeline progression: when a stage finishes a request, it enqueues outputs to the downstream stage; if downstream is idle, it starts immediately.
+- Scheduling
+  - A downstream stage triggers only after the upstream completes the request.
+  - Windowed segmentation/stitched triggering is not implemented; intra-stage streaming is not provided.
+
+### 7. IPC Implementation (simplified: copy-based)
+- Use `multiprocessing.Queue`/Pipe for inter-process communication (control + data).
+- Data is serialized/copied via CPU; no CUDA IPC/SHM zero-copy in this version.
+- Backpressure: queues are unbounded; pressure manifests as compute-rate differences. Optional SHM reduces large-object transfer cost; RX/decoding overhead is recorded for observability.
+
+### 8. Scheduling & Cancellation (simplified)
+- Pipeline: when a stage finishes a request, it enqueues to the next stage; that stage immediately pulls the next request from its input queue, enabling cross-stage concurrency.
+- Cancellation/timeout: explicit cancellation/timeouts are not provided; graceful shutdown uses a `None` sentinel sent to each stage input queue.
+
+#### Short sequence example (req0/req1, stage A→B)
+1) t0: stage A handles req0
+2) t1: req0 completes on A → enters B; A immediately starts req1
+3) t2: B handles req0 while A handles req1 (parallel across stages)
+
+### 9. Integration Points (by file)
+- `vllm_omni/entrypoints/omni_llm.py` (Orchestrator)
+  - Class `OmniLLM` orchestrates multi-process stages; constructs `OmniStage` instances in parallel and spawns per-stage workers.
+  - Spawns stage processes per config (set `CUDA_VISIBLE_DEVICES`/`torch.cuda.set_device`), creates control/data channels, builds simple full-trigger flow.
+  - Stats/logging are disabled by default; per-stage and orchestrator stats are only written when explicitly enabled.
+  - Manages process lifecycle: start/wait for readiness, graceful shutdown; forwards results between stages using copy-based IPC and optional SHM.
+  - Stage readiness: each stage emits `{"type": "stage_ready"}` after initialization; the orchestrator waits for all stages or times out and logs diagnostic suggestions.
+
+### 9.1 Process Device Visibility
+- Device binding: set `CUDA_VISIBLE_DEVICES` before process start, or call `torch.cuda.set_device` early in init; configs can be single/multi-GPU.
+- Cross-device transfer (simplified): via CPU copies; zero-copy paths are out of scope for now.
+
+### 10. Worker Template (pseudo-code)
+```python
+def stage_worker(input_q, output_q, runtime, shm_threshold):
+    # Device binding
+    set_stage_gpu_devices(runtime.devices)
+    engine = OmniStageLLM(model, **engine_args)
+    batch_seq = 0
+    while True:
+        first = input_q.get()
+        if first is None:
+            break
+        # Batch intake (up to runtime.max_batch_size)
+        batch = [first]
+        while len(batch) < int(runtime.max_batch_size or 1) and not input_q.empty():
+            nxt = input_q.get()
+            if nxt is None:
+                input_q.put(None)
+                break
+            batch.append(nxt)
+        # Decode IPC payload and sampling params
+        request_ids = []
+        engine_inputs = []
+        rx_bytes = {}
+        rx_decode_ms = {}
+        for t in batch:
+            ein, rx = maybe_load_from_ipc_with_metrics(t, "engine_inputs", "engine_inputs_shm")
+            request_ids.append(t["request_id"])
+            engine_inputs.extend(ein if isinstance(ein, list) else [ein])
+            rx_bytes[t["request_id"]] = rx["rx_transfer_bytes"]
+            rx_decode_ms[t["request_id"]] = rx["rx_decode_time_ms"]
+        # Generate and dispatch (grouped by request_id)
+        batch_seq += 1
+        outputs = list(engine.generate(engine_inputs, batch[0]["sampling_params"], use_tqdm=False))
+        grouped = group_by_request_id(outputs, request_ids)
+        for rid in request_ids:
+            r_out = grouped.get(rid, [])
+            use_shm, payload = maybe_dump_to_shm(r_out, shm_threshold)
+            msg = {
+                "request_id": rid,
+                "metrics": {
+                    "batch_id": batch_seq,
+                    "rx_transfer_bytes": int(rx_bytes.get(rid, 0)),
+                    "rx_decode_time_ms": float(rx_decode_ms.get(rid, 0.0)),
+                    # omitted: generation latency and token stats
+                },
+            }
+            if use_shm:
+                msg["engine_outputs_shm"] = payload
+            else:
+                msg["engine_outputs"] = payload
+            output_q.put(msg)
+```
+
+### 12. Observability
+- Metrics (as implemented):
+  - per-request (emitted by stages): `num_tokens_out`, `stage_gen_time_ms`, `rx_transfer_bytes`, `rx_decode_time_ms`, `rx_in_flight_time_ms`, `batch_id`
+  - orchestrator aggregates: E2E latency, tokens/s (written only when stats are enabled)
+  - optional per-stage JSONL: `{log_file}.stage{stage_id}.stats.jsonl`
+- Logs/tracing:
+  - optional per-stage log files: `{log_file}.stage{stage_id}.log`
+  - the orchestrator can log readiness, forward size/time, and summary information
+
+### 15. Risks & Mitigations
+— End-to-end latency: full-trigger sacrifices interactivity; start simple/correct, add windowed streaming later.
+— CPU copy overhead: copy-based IPC for maintainability; add SHM/zero-copy later as an optimization path.
+
+### 16. Refactor: Sink LLM init and process into Stage (diagram)
+
+Objective: without changing external call patterns, encapsulate per-stage LLM init, process creation, and worker logic inside `Stage`, so `PipelinedOmniLLM` focuses on orchestration (seeding, polling, forwarding, collecting) while keeping multi-process, shared memory, and device mapping capabilities.
+
+Key changes (vs. current):
+- Before: Orchestrator created processes directly, passing `_stage_worker`; device setup, LLM init, batching, and SHM were implemented there.
+- After: enhanced `OmniStage` owns its subprocess (`init_stage_worker`), with `_stage_worker` as the entry; device setup, LLM init, batching, and SHM live in `OmniStage`.
+
+Classes & responsibilities
+- OmniLLM (orchestrator)
+  - Build `OmniStage` list in parallel (preserve `process_engine_inputs` wiring)
+  - Connect adjacent `Stage` input/output queues; seed requests to stage 0
+  - Poll stage outputs, decode results, call `process_engine_inputs`, then encode and forward to next stage
+  - Termination & cleanup: distribute shutdown signals; join/terminate subprocesses
+- Stage (stage unit)
+  - Members: `stage_config` (with `engine_args` and `runtime`), `in_q/out_q`, subprocess handle, stats
+  - API:
+    - `init_stage_worker(...)`: spawn subprocess
+    - `stop_stage_worker()`: graceful exit
+    - `submit(ipc_payload)`: submit to input queue (may use shared memory)
+    - `try_collect() -> Optional[payload]`: non-blocking get from output queue
+    - `process_engine_inputs(stages, prompts)`: reuse/delegate existing logic
+  - Subprocess entry `worker_main()`:
+    - `set_stage_gpu_devices` (`CUDA_VISIBLE_DEVICES` & logical index mapping)
+    - build `OmniStageLLM(model, **engine_args)`
+    - while-loop:
+      - take first task + up to `runtime.max_batch_size-1` non-blocking; support `max_batch_size>=1`
+      - `maybe_load_from_ipc_with_metrics` per `engine_inputs` in batch
+      - `stage_engine.generate(batched_inputs, sampling_params)` (window `-1`)
+      - write per-request aggregated outputs to output queue after `maybe_dump_to_shm`
+      - log start/finish; report `error` per request on exceptions
+
+Data flow (sketch)
+```
+OmniLLM
+  ├─ Stage[0].start_process(in0, out0)
+  ├─ Stage[1].start_process(in1, out1)
+  ├─ ...
+  └─ Orchestrator loop:
+       - stage0.submit(encode(engine_inputs))
+       - for each stage:
+           res = stage.try_collect()
+           outputs = decode(res)
+           stage.set_engine_outputs(outputs)
+           if !final: next_inputs = next_stage.process_engine_inputs(stages, original_prompt)
+                    next_stage.submit(encode(next_inputs))
+```
+
+Config (preserve & extend)
+- `runtime.devices`:
+  - Comma-separated visibility list (e.g., "2,5,7"): set visible set; default logical index 0 (first GPU)
+  - Number/string: treat as logical index (0/1-based per implementation) mapping to physical GPUs
+- `runtime.max_batch_size`: max requests taken per intake (default 1)
+- `shm_threshold_bytes`: objects larger than this go via shared memory (default 64KB; injected into stages)
+- Stage link and inputs:
+  - `engine_input_source`: list of upstream stage_id(s) from which the downstream stage takes inputs (typically a single source)
+  - `custom_process_input_func`: function to construct downstream engine_inputs from upstream outputs
+- Output flags:
+  - `final_output`, `final_output_type`: marks that a stage produces the final external output and its type
+
+Kept & strengthened
+- Multi-process: each stage in its own process; parallel across stages; serial/batching inside a stage controlled by `max_batch_size`
+- Shared memory: large objects via SHM with strict `memoryview -> close -> unlink`
+- Device mapping: unified `set_stage_gpu_devices`, support comma lists and logical indices
+
+Migration (execution order)
+1) Enhanced `OmniStage` with `init_stage_worker/stop_stage_worker/submit/try_collect/_stage_worker`.
+2) `OmniLLM` holds `OmniStage` instances, replacing the older pipelined class; uses Stage APIs.
+3) Reuse `entrypoints/stage_utils`: device/SHM/encode-decode utilities live here.
+4) Regression: single/multi-request, various `max_batch_size`, SHM threshold, device mapping, multi-stage parallelism, error paths, and shutdown flows.

docs/design/vllm_omni_design.md

@@ -0,0 +1,509 @@
+# Architecture Design Principles
+
+## Overview
+
+vLLM-Omni is a multi-modality extension for vLLM that supports non-autoregressive structures and non-textual outputs. This document outlines the key software abstractions, APIs, and dependencies for the system, designed to maximize reuse of vLLM's proven architecture.
+
+## Architecture Principles
+
+1. **vLLM V1 Compatibility**: Built on vLLM's Engine V1 architecture with AsyncLLM and EngineCore patterns
+2. **Stage-based Processing**: Models are divided into multiple stages, each processed by different Engine Cores
+3. **Multiple Engine Core Support**: Each stage can use either AR Engine Core (reusing vLLM) or Diffusion Engine Core (new DiT support), or other new Cores
+4. **Worker Process Pattern**: Follows vLLM's multiprocess worker architecture for scalability
+5. **Extensibility**: Easy integration of new modalities, model architectures, and output formats
+
+## Key Data Flow
+
+API Server --> OmniLLM/AsyncOmniLLM (New, including multi engines) --> LLMEngine/AsyncLLM --> Engine Core
+ --> Scheduler (New one for DiT) --> Executor (New one for diffusers) --> Worker (New one for DiT)
+ --> ModelRunner (New one for AR hiddenstate, New one for DiT) --> RequestState --> OutputProcessoer (New one for final multimodal output)
+
+## Core Components (Ordered by Data Flow)
+
+### 1. Installation
+```bash
+pip install vllm
+pip install vllm-omni
+```
+
+### 2. Online inference launch (Entry Point)
+
+Keep consistency with the vllm main branch
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8000
+```
+
+#### Design Architecture
+```mermaid
+graph TD
+    A[vllm serve model --omni] --> B[vLLM-Omni CLI Wrapper]
+    B --> C{Detect --omni flag}
+    C -->|Yes| D[Parse OmniConfig]
+    C -->|No| E[Forward to vLLM CLI]
+    D --> F[Initialize AsyncOmniLLM]
+    F --> G[Start omni Server]
+    G --> H[Multi-stage Processing]
+    E --> I[Standard vLLM Pipeline]
+
+    subgraph "vLLM-Omni Components"
+        F
+        G
+        H
+    end
+
+    subgraph "vLLM Components"
+        I
+    end
+```
+
+#### omni Serve Command Implementation
+File: vllm_omni/entrypoints/cli/main.py
+```python
+import sys
+import argparse
+from typing import List, Optional
+from vllm_omni.entrypoints.omni import OmniServeCommand
+
+def main():
+    """Main CLI entry point that intercepts vLLM commands"""
+    # Check if --omni flag is present
+    if "--omni" in sys.argv:
+        # Remove --omni flag and process with vLLM-Omni
+        omni_args = [arg for arg in sys.argv[1:] if arg != "--omni"]
+        omni_serve = OmniServeCommand()
+        omni_serve.run(omni_args)
+    else:
+        # Forward to original vLLM CLI
+        from vllm.entrypoints.cli.main import main as vllm_main
+        vllm_main()
+
+if __name__ == "__main__":
+    main()
+```
+
+#### vLLM Plugin Integration
+File: vllm_omni/pyproject.toml
+```python
+[project.scripts]
+# Override vLLM's CLI entry point
+vllm = "vllm_omni.cli:main"
+
+# Add entry point for vLLM integration
+[project.entry-points."vllm.plugins"]
+omni = "vllm_omni.plugin:OmniPlugin"
+```
+
+### 3. Offline inference launch (Entry Point)
+Design an upper level class to incorporate multi Engines, each engine has a engine call.
+```python
+from vllm.entrypoints.llm import LLM
+from vllm.v1.engine.llm_engine import LLMEngine
+
+class OmniLLM(LLM):
+    """Extended LLM supporting multiple engines and stage-based processing"""
+
+    def __init__(self, stage_configs: List[StageConfig]):
+        super().__init__()
+        self.stage_configs = stage_configs
+        self.engine_list = []  # List of AsyncLLM instances for each stage
+        self.output_processor = MultimodalOutputProcessor()
+        self._initialize_stage_engines()
+
+    def _initialize_stage_engines(self) -> None:
+        """Initialize LLMEngine instances for each stage"""
+        for stage_config in self.stage_configs:
+            stage_llm = LLMEngine.from_vllm_config(
+                    vllm_config=stage_config.vllm_config,
+                    executor_class=self.executor_class,
+                    log_stats=self.log_stats
+                )
+            self.engine_list.append(stage_llm)
+
+    def generate(
+        self,
+        stage_args_list: List[stage_args],
+        use_tqdm: Union[bool, Callable[..., tqdm]] = True,
+        lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
+        priority: Optional[list[int]] = None,
+    ) -> list[RequestOutput]:
+        """Main generation interface - orchestrates multi-stage processing"""
+
+        # Process through each stage sequentially
+        for i, stage_config in enumerate(self.stage_configs):
+            stage_engine = self.engine_list[i]
+
+            # Prepare input for this stage
+            stage_args = stage_args_list[i]
+            prompt_str, engine_request, tokenization_kwargs = self._process_stage_inputs(stage_config, **stage_args)
+
+            # Add inputs to Engine
+            stage_engine.add_request(requesy_id, prompt_str, tokenization_kwargs)
+            # Run Engine
+            stage_output = stage_engine.step()
+
+            # Update stage input and output of each stage for later usage
+            self._update_stage_io(stage_output, stage_config)
+
+
+        # Process final output
+        final_output = self.output_processor.process_output(
+            self.stage_configs[-1].stage_output
+        )
+        return final_output
+
+    def _process_stage_inputs(stage_config, **stage_args) -> Any:
+        """Prepare input for specific stage"""
+        if stage_config.engine_type == "AR":
+            return self._process_ar_inputs(**stage_args)
+        elif stage_config.engine_type == "DiT":
+            return self._process_dit_inputs(**stage_args)
+        else
+            raise NotImplementedError
+
+    def _process_dit_inputs(**stage_args)-> Any:
+        image_latent = self.vae.encode(**stage_args)
+
+        return image_latent
+```
+
+### 4. Stage config for setting of different model stages
+```python
+@dataclass
+class StageConfig:
+    """Configuration for a processing stage"""
+    stage_id: int
+    engine_type: str  # "AR" or "DiT"
+    model_path: str
+    input_modalities: List[str]
+    output_modalities: List[str]
+    vllm_config: Optional[VllmConfig] = None  # For engine config of corresponding stage
+    executor_class: type[Executor]  # For execute class config of corresponding stage
+    dit_config: Optional[DiTConfig] = None  # For diffusion stages
+    cache_config: Optional[DiTCacheConfig] = None
+```
+For AR stage, the setting is:
+```python
+from vllm.v1.core.sched.scheduler import Scheduler
+from vllm.v1.executor.multiproc_executor import MultiprocExecutor
+ar_stage_config = StageConfig()
+ar_stage_config.vllm_config.scheduler_config.scheduler_cls = Scheduler # original vllm scheduler for AR
+ar_stage_config.executor_class = MultiprocExecutor # original vllm executor for AR
+```
+For DiT stage, the setting is:
+```python
+from vllm.v1.executor.multiproc_executor import MultiprocExecutor
+dit_stage_config = StageConfig()
+dit_stage_config.vllm_config.scheduler_config.scheduler_cls = DiffusionScheduler # New scheduler for DiT
+
+# For diffusion models without using diffusers
+dit_stage_config.executor_class = MultiprocExecutor
+
+# For diffusion models using diffusers
+dit_stage_config.executor_class = DiffusersPipelineExecutor
+```
+
+### 5. Online inference main class
+
+Similar to OmniLLM in offline inference, add some asynchronous processing, referring to AsyncLLM
+```python
+from vllm.v1.engine.async_llm import AsyncLLM
+
+
+class AsyncOmniLLM(AsyncLLM):
+    """Extended AsyncLLM supporting multiple engines and stage-based processing"""
+
+    def __init__(self, stage_configs: List[StageConfig]):
+        super().__init__()
+```
+
+### 6. Engine Core
+No need to change. The specific executor, scheduler and output type can be transferred into it with new configs.
+```python
+# class: EngineCore
+# EngineCore.step (simplified)
+class EngineCore:
+    def step(self):
+        scheduler_output = scheduler.schedule()
+        model_output = executor.execute_model(scheduler_output)
+        engine_outputs = scheduler.update_from_output(
+            scheduler_output, model_output
+        )
+        return engine_outputs
+```
+
+### 7. Scheduler
+Create a new Diffusion Scheduler for DiT, which is inherited from original vllm AR scheduler. At first, no complicated strategy.
+Just process the request first in first out. Then create a class of DiT Cache Manager for optimization in the future.
+```python
+from vllm.v1.core.sched.scheduler import Scheduler
+class DiffusionScheduler(Scheduler):
+
+    def __init__(
+        self,
+        vllm_config: VllmConfig,
+        kv_cache_config: KVCacheConfig,
+        dit_cache_config: DiTCacheConfig,
+        structured_output_manager: StructuredOutputManager,
+        mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
+        include_finished_set: bool = False,
+        log_stats: bool = False,
+    ) -> None:
+
+    self.dit_cache_manager = DiTCacheManager(dit_cache_config)
+```
+For DiT Cache Manager (Can refer to xDiT):
+```python
+class DiTCacheManager:
+    """Manages DiT-specific caching"""
+
+    def __init__(self, config: DiTCacheConfig):
+        self.cache_tensors: Dict[str, torch.Tensor] = {}
+        self.cache_groups: List[DiTCacheTensor] = config.dit_cache_tensors
+
+    def allocate_cache(self, request_id: str, size: int) -> torch.Tensor
+    def get_cache(self, request_id: str) -> Optional[torch.Tensor]
+    def release_cache(self, request_id: str) -> None
+    def clear_expired_cache(self) -> None
+```
+
+### 8. Executor for DiT without diffusers
+
+No need to change. Just import original AR executor.
+
+### 9. Diffusers Pipeline Executor (no worker)
+
+A single-process executor that directly runs the Diffusers pipeline without spawning workers or using RPC. Interfaces remain identical to `Executor` so the EngineCore loop is unchanged.
+
+#### Function map（Pipeline Executor）
+
+##### 1) Inherited and overridden
+```python
+from concurrent.futures import Future
+from typing import Optional, Union, Callable, Any
+import torch
+import torch.nn as nn
+from vllm.v1.executor.abstract import Executor
+from vllm.v1.kv_cache_interface import KVCacheConfig, KVCacheSpec
+from vllm.v1.outputs import ModelRunnerOutput
+from vllm.tasks import SupportedTask
+from diffusers import DiffusionPipeline
+
+class DiffusersPipelineExecutor(Executor):
+    supports_pp: bool = False  # Single-process, no TP/PP/DP
+
+    def _init_executor(self) -> None:
+        # Called by ExecutorBase.__init__
+        self._failure_callback: Optional[Callable[[], None]] = None
+        self._device = self._resolve_device()
+        self._dtype = self._resolve_dtype()
+        self._pipeline = self._build_pipeline(device=self._device, dtype=self._dtype)
+        self._profiler = None
+        self._is_failed = False
+        self.is_sleeping = False
+        self.sleeping_tags: set[str] = set()
+
+    # major functions to build/run diffusers pipeline
+    def _build_pipeline(self, device:torch.device, dtype:torch.dtype)->DiffusionPipeline:
+        model_name = "Qwen/Qwen-Image"
+
+        self.pipe = DiffusionPipeline.from_pretrained(model_name, torch_dtype=dtype)
+        self.pipe = pipe.to(device)
+
+    def _run_pipeline(self, scheduler_output) -> ModelRunnerOutput: ...
+        positive_magic = {
+            "en": ", Ultra HD, 4K, cinematic composition.", # for english prompt
+            "zh": ", 超清，4K，电影级构图." # for chinese prompt
+        }
+
+        # Generate image
+        prompt_embeds = self._get_and_process_prompt_embeds(scheduler_output, positive_magic)
+        negative_prompt_embeds = self.pipe.embed_prompt(" ")
+
+        width, height = aspect_ratios["16:9"]
+
+        image = pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            width=width,
+            height=height,
+            num_inference_steps=50,
+            true_cfg_scale=4.0,
+            generator=torch.Generator(device="cuda").manual_seed(42)
+        ).images[0]
+
+        output = self.wrap_image_as_ModelRunnerOutput(image)
+        return output
+
+    # ---- Internal helpers (implementation-specific, not public API) ----
+    def _resolve_device(self): ...
+    def _resolve_dtype(self): ...
+    def _get_model(self) -> nn.Module: ...
+    def _get_and_process_prompt_embeds(self, scheduler_output, positive_magic):
+        ...
+        #append the positive_magic to prompt and embed them to prompt embed tensors
+
+    # Functions related to workers should either raise NotImplementedError
+    # or return empty defaults in this no-worker executor.
+
+    def collective_rpc(self, method, timeout=None, args=(), kwargs=None) -> list[Any]:
+        # No workers in pipeline executor
+        raise NotImplementedError("No workers in DiffusersPipelineExecutor")
+
+    def initialize_from_config(self, kv_cache_configs: list[KVCacheConfig]) -> None:
+        return  # no-op (pipeline already built in _init_executor)
+
+    def register_failure_callback(self, callback):
+        self._failure_callback = callback
+
+    def determine_available_memory(self) -> list[int]:  # bytes
+        # Single device; return [available_bytes]. If CPU-only, return [0].
+        return [self._determine_available_bytes(self._device)]
+
+```
+
+### 10. Worker
+
+#### Inheritance strategy
+Prefer reusing the mature GPU Worker end-to-end. Worker class is selected by configuration (vllm_config). Do not add a new executor-specific worker binding. If customization is needed, override only `init_device` to construct the `DiffusionModelRunner`; all other behaviors (device init details, profiling, sleep/wake, PP/TP comms, execute path) remain from `vllm/v1/worker/gpu_worker.py::Worker`.
+
+
+Inherited and overridden
+```python
+# Optional: only if you need to plug a custom DiffusionModelRunner.
+from vllm.v1.worker.gpu_worker import Worker as GPUWorker
+from vllm.v1.worker.diffusion_model_runner import DiffusionModelRunner
+
+class Worker(GPUWorker):
+    def init_device(self) -> None:
+        #those related to device check and init
+        ...
+
+        self.model_runner = DiffusionModelRunner(self.vllm_config, ...)
+```
+
+### 11. Model Runner
+
+### Function map (Model Runner)
+#### 1) Inherited and overridden
+Those parts relied to the KV Cache will be omitted if we do not register the model to the vllm config. The engine core will view it as do not require KV Cache, and handle it properly
+
+Reuse `vllm/v1/outputs.py::ModelRunnerOutput`：
+- DiffusionModelRunner: Use the `pooler_output=[Tensor,...]` to return multi modal tensors
+- ARModelRunner: Use the `pooler_output=[Tensor,...]` to return hidden states.
+```python
+from typing import Optional, Union
+import torch
+from vllm.v1.worker.gpu_model_runner import GPUModelRunner
+from vllm.v1.outputs import ModelRunnerOutput
+
+
+class DiffusionModelRunner(GPUModelRunner):
+    @torch.inference_mode()
+    def execute_model(
+        self,
+        scheduler_output: "SchedulerOutput",
+        intermediate_tensors: Optional[IntermediateTensors] = None,
+    ) -> Union[ModelRunnerOutput, IntermediateTensors]:
+        ...
+        return ModelRunnerOutput(
+            req_ids=[...],
+            req_id_to_index={...},
+            sampled_token_ids=[],
+            spec_token_ids=None,
+            logprobs=None,
+            prompt_logprobs_dict={},
+            pooler_output=[Tensor,...],         # Return Hidden states
+            kv_connector_output=None,
+            num_nans_in_logits=None,
+        )# return multi modal tensors via pooler_output=[Tensor,...]
+
+
+class ARModelRunner(GPUModelRunner):
+    @torch.inference_mode()
+    def execute_model(
+        self,
+        scheduler_output: "SchedulerOutput",
+        intermediate_tensors: Optional[IntermediateTensors] = None,
+    ) -> Union[ModelRunnerOutput, IntermediateTensors]:
+        ...
+        return ModelRunnerOutput(
+            req_ids=[...],
+            req_id_to_index={...},
+            sampled_token_ids=[],
+            spec_token_ids=None,
+            logprobs=None,
+            prompt_logprobs_dict={},
+            pooler_output=[Tensor,...],         # Return Hidden states
+            kv_connector_output=None,
+            num_nans_in_logits=None,
+        )# return hidden states via pooler_output=[Tensor,...]
+```
+
+### 12. RequestState Role and Usage
+```bash
+# RequestState serves as the per-request state tracker in OutputProcessor:
+# - Maintains request-specific state (tokens, logprobs, detokenizer, etc.)
+# - Converts EngineCoreOutput → RequestOutput/PoolingRequestOutput
+# - Manages request lifecycle from registration to completion
+AsyncLLM.add_request()
+↓
+OutputProcessor.add_request()
+↓
+RequestState.from_new_request() → 创建请求状态
+
+AsyncLLM.__init__() / AsyncLLM.generate() / AsyncLLM.encode() →  创建一个Background loop 持续从EngineCore获取输出
+↓
+OutputProcessor.process_outputs() → 更新状态并处理输出
+↓
+RequestState.make_request_output() → 转换为最终输出,格式为RequestOutput或者PoolingRequestOutput
+↓
+RequestOutputCollector.put() → 推送到队列（AsyncLLM）
+```
+
+Need to add implementation for an existing method
+```python
+class RequestState:
+    def __init__(self, request_id: str, parent_req: Optional[ParentRequest],
+                 request_index: int, lora_name: Optional[str],
+                 output_kind: RequestOutputKind, prompt: Optional[str],
+                 prompt_token_ids: list[int], logprobs_processor: Optional[LogprobsProcessor],
+                 detokenizer: Optional[IncrementalDetokenizer],
+                 max_tokens_param: Optional[int], arrival_time: float,
+                 queue: Optional[RequestOutputCollector], log_stats: bool):
+
+    def _new_pooling_output(self, pooling_output: torch.Tensor) -> PoolingOutput:
+        """Create PoolingOutput for multimodal/pooling requests"""
+        return PoolingOutput(data=pooling_output)
+```
+
+
+### 13. OutputProcessor
+
+For hidden state output, original OutputProcessor already support.
+
+We only need to add new one for final multimodal output.
+```python
+class MultimodalOutputProcessor(OutputProcessor):
+    """Handles multimodal output processing"""
+
+    def __init__(self):
+        self.output_handlers: Dict[str, OutputProcessor] = {}
+
+    def process_outputs(self, engine_core_outputs: list[EngineCoreOutput], ...):
+        for engine_core_output in engine_core_outputs:
+            # Option 1: Use output_type field (if EngineCoreOutput is extended)
+            if engine_core_output.output_type == "image":
+                self._process_image_output(engine_core_output)
+            elif engine_core_output.output_type == "text+image":
+                self._process_text_image_output(engine_core_output)
+            elif engine_core_output.output_type == "latents":
+                self._process_latents_output(engine_core_output)
+            elif engine_core_output.output_type == "text":
+                self._process_text_output(engine_core_output)
+            else:
+                # Fallback: use existing pooling_output logic
+                if engine_core_output.pooling_output is not None:
+                    self._process_pooling_output(engine_core_output)
+                else:
+                    self._process_text_output(engine_core_output)
+```

docs/examples/README.md

@@ -0,0 +1,6 @@
+# Examples
+
+vLLM-Omni's examples are split into two categories:
+
+- If you are using vLLM-Omni from within Python code, see the *Offline Inference* section.
+- If you are using vLLM-Omni from an HTTP application or client, see the *Online Serving* section.

docs/getting_started/installation/.nav.yml

@@ -0,0 +1,4 @@
+nav:
+  - README.md
+  - gpu.md
+  - npu.md

docs/getting_started/installation/README.md

@@ -0,0 +1,7 @@
+# Installation
+
+vLLM supports the following hardware platforms:
+
+- [GPU](gpu.md)
+    - [NVIDIA CUDA](gpu.md)
+- [Ascend NPU](npu.md)

docs/getting_started/installation/gpu.md

@@ -0,0 +1,40 @@
+# GPU
+
+vLLM-Omni is a Python library that supports the following GPU variants. The library itself mainly contains python implementations for framework and models.
+
+## Requirements
+
+- OS: Linux
+- Python: 3.12
+
+!!! note
+    vLLM-Omni is currently not natively supported on Windows.
+
+=== "NVIDIA CUDA"
+
+    --8<-- "docs/getting_started/installation/gpu/cuda.inc.md:requirements"
+
+## Set up using Python
+
+### Create a new Python environment
+
+--8<-- "docs/getting_started/installation/python_env_setup.inc.md"
+
+### Pre-built wheels
+
+=== "NVIDIA CUDA"
+
+    --8<-- "docs/getting_started/installation/gpu/cuda.inc.md:pre-built-wheels"
+
+[](){ #build-from-source }
+
+### Build wheel from source
+
+=== "NVIDIA CUDA"
+
+    --8<-- "docs/getting_started/installation/gpu/cuda.inc.md:build-wheel-from-source"
+
+## Set up using Docker
+
+### Pre-built images
+To be released... Stay tuned!

docs/getting_started/installation/gpu/cuda.inc.md

@@ -0,0 +1,75 @@
+# --8<-- [start:requirements]
+
+- GPU: compute capability 7.0 or higher (e.g., V100, T4, RTX20xx, A100, L4, H100, etc.)
+
+# --8<-- [end:requirements]
+# --8<-- [start:set-up-using-python]
+
+vLLM-Omni depends vLLM. So please follow instructions below mainly for vLLM.
+
+!!! note
+    PyTorch installed via `conda` will statically link `NCCL` library, which can cause issues when vLLM tries to use `NCCL`. See <gh-issue:8420> for more details.
+
+In order to be performant, vLLM has to compile many cuda kernels. The compilation unfortunately introduces binary incompatibility with other CUDA versions and PyTorch versions, even for the same PyTorch version with different building configurations.
+
+Therefore, it is recommended to install vLLM and vLLM-Omni with a **fresh new** environment. If either you have a different CUDA version or you want to use an existing PyTorch installation, you need to build vLLM from source. See [build-from-source-vllm](https://docs.vllm.ai/en/stable/getting_started/installation/gpu/#build-wheel-from-source) for more details.
+
+# --8<-- [start:pre-built-wheels]
+
+#### Installation of vLLM
+
+vLLM-Omni is built based on vLLM v0.11.0. Please install it with command below.
+```bash
+uv pip install vllm==0.11.0 --torch-backend=auto
+```
+
+#### Installation of vLLM-Omni
+
+```bash
+uv pip install vllm-omni
+```
+
+# --8<-- [end:pre-built-wheels]
+
+# --8<-- [start:build-wheel-from-source]
+
+#### Installation of vLLM
+If you do not need to modify source code of vLLM, you can directly install the stable 0.11.0 release version of the library
+
+```bash
+uv pip install vllm==0.11.0 --torch-backend=auto
+```
+
+#### Installation of vLLM-Omni
+Install additional requirements for vLLM-Omni
+```bash
+git clone https://github.com/vllm-project/vllm-omni.git
+cd vllm_omni
+uv pip install -e .
+```
+
+<details><summary>(Optional) Installation of vLLM from source</summary>
+If you want to check, modify or debug with source code of vLLM, install the library from source with the following instructions:
+
+```bash
+git clone https://github.com/vllm-project/vllm.git
+cd vllm
+git checkout v0.11.0
+```
+Set up environment variables to get pre-built wheels. If there are internet problems, just download the whl file manually. And set `VLLM_PRECOMPILED_WHEEL_LOCATION` as your local absolute path of whl file.
+```bash
+export VLLM_PRECOMPILED_WHEEL_LOCATION=https://github.com/vllm-project/vllm/releases/download/v0.11.0/vllm-0.11.0-cp38-abi3-manylinux1_x86_64.whl
+```
+Install vllm with command below (If you have no existing PyTorch).
+```bash
+uv pip install --editable .
+```
+Install vllm with command below (If you already have PyTorch).
+```bash
+python use_existing_torch.py
+uv pip install -r requirements/build.txt
+uv pip install --no-build-isolation --editable .
+```
+</details>
+
+# --8<-- [end:build-wheel-from-source]

docs/getting_started/installation/npu.md

@@ -0,0 +1,3 @@
+# Ascend-NPU
+
+vLLM-Omni is a Python library that supports the following NPU variants. Select your NPU type to see vendor specific instructions:

docs/getting_started/installation/npu/npu.inc.md

@@ -0,0 +1,5 @@
+# --8<-- [start:installation]
+
+vLLM-Omni mainly contains python implementations for framework and models.
+
+# --8<-- [end:installation]

docs/getting_started/installation/python_env_setup.inc.md

@@ -0,0 +1,6 @@
+It's recommended to use [uv](https://docs.astral.sh/uv/), a very fast Python environment manager, to create and manage Python environments. Please follow the [documentation](https://docs.astral.sh/uv/#getting-started) to install `uv`. After installing `uv`, you can create a new Python environment using the following commands:
+
+```bash
+uv venv --python 3.12 --seed
+source .venv/bin/activate
+```

docs/getting_started/quickstart.md

@@ -0,0 +1,23 @@
+# Quickstart
+
+This guide will help you quickly get started with vLLM-Omni to perform:
+
+- Offline batched inference
+- Online serving using OpenAI-compatible server
+
+## Prerequisites
+
+- OS: Linux
+- Python: 3.12
+
+## Installation
+
+Please refer to [installation](installation/README.md)
+
+## Offline Inference
+
+Please refer to [offline inference](../user_guide/examples/offline_inference/qwen2_5_omni.md)
+
+## Online Serving with OpenAI-Completions API
+
+Please refer to [online serving](../user_guide/examples/online_serving/qwen2_5_omni.md)

docs/mkdocs/hooks/generate_api_readme.py

@@ -0,0 +1,268 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+"""
+Hook to automatically generate docs/api/README.md from the codebase.
+
+This script scans the vllm_omni module for public classes and functions,
+categorizes them, and generates a summary README file.
+"""
+
+import ast
+import logging
+from pathlib import Path
+
+logger = logging.getLogger("mkdocs")
+
+ROOT_DIR = Path(__file__).parent.parent.parent.parent
+API_README_PATH = ROOT_DIR / "docs" / "api" / "README.md"
+
+# Category mappings: module prefix -> category name and description
+CATEGORIES = {
+    "entrypoints": {
+        "name": "Entry Points",
+        "description": "Main entry points for vLLM-Omni inference and serving.",
+    },
+    "inputs": {
+        "name": "Inputs",
+        "description": "Input data structures for multi-modal inputs.",
+    },
+    "engine": {
+        "name": "Engine",
+        "description": "Engine classes for offline and online inference.",
+    },
+    "core": {
+        "name": "Core",
+        "description": "Core scheduling and caching components.",
+    },
+    "model_executor": {
+        "name": "Model Executor",
+        "description": "Model execution components.",
+    },
+    "config": {
+        "name": "Configuration",
+        "description": "Configuration classes.",
+    },
+    "worker": {
+        "name": "Workers",
+        "description": "Worker classes and model runners for distributed inference.",
+    },
+}
+
+
+class APIVisitor(ast.NodeVisitor):
+    """AST visitor to extract public classes and module-level functions."""
+
+    def __init__(self, module_path: str):
+        self.module_path = module_path
+        self.classes: list[str] = []
+        self.functions: list[str] = []
+        self._class_stack: list[str] = []  # Track nested class definitions
+
+    def visit_ClassDef(self, node: ast.ClassDef) -> None:
+        """Visit class definitions."""
+        if not node.name.startswith("_"):
+            self.classes.append(f"{self.module_path}.{node.name}")
+        # Track that we're entering a class
+        self._class_stack.append(node.name)
+        self.generic_visit(node)
+        # Remove from stack when done visiting
+        self._class_stack.pop()
+
+    def visit_FunctionDef(self, node: ast.FunctionDef) -> None:
+        """Visit function definitions - only collect module-level functions."""
+        # Only collect if we're not inside a class (stack is empty)
+        if not self._class_stack and not node.name.startswith("_"):
+            self.functions.append(f"{self.module_path}.{node.name}")
+        self.generic_visit(node)
+
+    def visit_AsyncFunctionDef(self, node: ast.AsyncFunctionDef) -> None:
+        """Visit async function definitions - only collect module-level functions."""
+        # Only collect if we're not inside a class (stack is empty)
+        if not self._class_stack and not node.name.startswith("_"):
+            self.functions.append(f"{self.module_path}.{node.name}")
+        self.generic_visit(node)
+
+
+def parse_file_for_symbols(file_path: Path, module_path: str) -> tuple[list[str], list[str]]:
+    """
+    Parse a Python file and extract public classes and functions.
+
+    Returns:
+        Tuple of (classes, functions)
+    """
+    try:
+        # If this is __init__.py, use parent module path
+        if file_path.name == "__init__.py":
+            # Remove .__init__ from module path
+            if module_path.endswith(".__init__"):
+                module_path = module_path[:-9]
+
+        with open(file_path, encoding="utf-8") as f:
+            content = f.read()
+
+        tree = ast.parse(content, filename=str(file_path))
+        visitor = APIVisitor(module_path)
+        visitor.visit(tree)
+
+        return visitor.classes, visitor.functions
+    except Exception as e:
+        logger.debug(f"Could not parse {file_path}: {e}")
+        return [], []
+
+
+def scan_package(package_name: str = "vllm_omni") -> dict[str, list[str]]:
+    """
+    Scan the vllm_omni package and categorize public symbols.
+
+    Returns:
+        Dict mapping category names to lists of symbol full names
+    """
+    categorized: dict[str, list[str]] = {cat["name"]: [] for cat in CATEGORIES.values()}
+
+    try:
+        # Find the package directory
+        package_path = ROOT_DIR / package_name
+        if not package_path.exists():
+            logger.warning(f"Package path not found: {package_path}")
+            return categorized
+
+        # Walk through all Python files
+        for py_file in package_path.rglob("*.py"):
+            # Skip __init__.py and private modules
+            if py_file.name.startswith("_") and py_file.name != "__init__.py":
+                continue
+
+            # Get module path
+            relative_path = py_file.relative_to(ROOT_DIR)
+            module_path = str(relative_path.with_suffix("")).replace("/", ".").replace("\\", ".")
+
+            # Handle __init__.py - use parent module path
+            if py_file.name == "__init__.py":
+                # Remove .__init__ from module path
+                if module_path.endswith(".__init__"):
+                    module_path = module_path[:-9]
+
+            # Determine category from module path
+            category = None
+            for prefix, cat_info in CATEGORIES.items():
+                if prefix in module_path:
+                    category = cat_info["name"]
+                    break
+
+            if not category:
+                continue
+
+            # Parse file for symbols
+            classes, functions = parse_file_for_symbols(py_file, module_path)
+
+            # Filter out internal implementation classes
+            # Skip classes that look like internal components (DiT layers, etc.)
+            internal_patterns = [
+                "Block",
+                "Layer",
+                "Net",
+                "Embedding",
+                "Norm",
+                "Activation",
+                "Solver",
+                "Pooling",
+                "Attention",
+                "MLP",
+                "DecoderLayer",
+                "InputEmbedding",
+                "TimestepEmbedding",
+                "CodecEmbedding",
+                "DownSample",
+                "UpSample",
+                "Res2Net",
+                "SqueezeExcitation",
+                "TimeDelay",
+                "TorchActivation",
+                "SnakeBeta",
+                "SinusPosition",
+                "RungeKutta",
+                "AMPBlock",
+                "AdaLayerNorm",
+            ]
+
+            # Add classes (filter out internal ones)
+            for class_name in classes:
+                class_short_name = class_name.split(".")[-1]
+                # Skip if it matches internal patterns (unless it's a main model class)
+                if any(pattern in class_short_name for pattern in internal_patterns):
+                    # But include main model classes
+                    if not any(
+                        main_class in class_short_name
+                        for main_class in [
+                            "ForConditionalGeneration",
+                            "Model",
+                            "Registry",
+                            "Worker",
+                            "Runner",
+                            "Scheduler",
+                            "Manager",
+                            "Processor",
+                            "Config",
+                        ]
+                    ):
+                        continue
+                categorized[category].append(class_name)
+
+            # Add important functions (parse, preprocess, etc.)
+            for func_name in functions:
+                # Include functions that match certain patterns
+                if any(keyword in func_name.lower() for keyword in ["parse", "preprocess"]):
+                    categorized[category].append(func_name)
+
+        # Sort symbols within each category
+        for category in categorized:
+            categorized[category].sort()
+
+    except Exception as e:
+        logger.error(f"Error scanning package: {e}", exc_info=True)
+
+    return categorized
+
+
+def generate_readme(categorized: dict[str, list[str]]) -> str:
+    """Generate the API README markdown content."""
+    lines = ["# Summary", ""]
+
+    # Generate sections for each category
+    for prefix, cat_info in CATEGORIES.items():
+        category_name = cat_info["name"]
+        description = cat_info["description"]
+        symbols = categorized.get(category_name, [])
+
+        if not symbols:
+            continue
+
+        lines.append(f"## {category_name}")
+        lines.append("")
+        lines.append(description)
+        lines.append("")
+
+        for symbol in symbols:
+            lines.append(f"- [{symbol}][]")
+
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+def on_startup(command, dirty: bool):
+    """MkDocs hook entry point."""
+    logger.info("Generating API README documentation")
+
+    # Scan the package
+    categorized = scan_package()
+
+    # Generate README content
+    content = generate_readme(categorized)
+
+    # Write to file
+    API_README_PATH.parent.mkdir(parents=True, exist_ok=True)
+    with open(API_README_PATH, "w", encoding="utf-8") as f:
+        f.write(content)
+
+    logger.info(f"API README generated: {API_README_PATH.relative_to(ROOT_DIR)}")

docs/mkdocs/hooks/url_schemes.py

@@ -0,0 +1,121 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""
+This is basically a port of MyST parser’s external URL resolution mechanism
+(https://myst-parser.readthedocs.io/en/latest/syntax/cross-referencing.html#customising-external-url-resolution)
+to work with MkDocs.
+
+It allows Markdown authors to use GitHub shorthand links like:
+
+  - [Text](gh-issue:123)
+  - <gh-pr:456>
+  - [File](gh-file:path/to/file.py#L10)
+
+These are automatically rewritten into fully qualified GitHub URLs pointing to
+issues, pull requests, files, directories, or projects in the
+`vllm-project/vllm-omni` repository.
+
+The goal is to simplify cross-referencing common GitHub resources
+in project docs.
+"""
+
+import regex as re
+from mkdocs.config.defaults import MkDocsConfig
+from mkdocs.structure.files import Files
+from mkdocs.structure.pages import Page
+
+
+def on_page_markdown(markdown: str, *, page: Page, config: MkDocsConfig, files: Files) -> str:
+    """
+    Custom MkDocs plugin hook to rewrite special GitHub reference links
+    in Markdown.
+
+    This function scans the given Markdown content for specially formatted
+    GitHub shorthand links, such as:
+      - `[Link text](gh-issue:123)`
+      - `<gh-pr:456>`
+
+    And rewrites them into fully-qualified GitHub URLs with GitHub icons:
+      - `[:octicons-mark-github-16: Link text](https://github.com/vllm-project/vllm/issues/123)`
+      - `[:octicons-mark-github-16: Pull Request #456](https://github.com/vllm-project/vllm/pull/456)`
+
+    Supported shorthand types:
+      - `gh-issue`
+      - `gh-pr`
+      - `gh-project`
+      - `gh-dir`
+      - `gh-file`
+
+    Args:
+        markdown (str): The raw Markdown content of the page.
+        page (Page): The MkDocs page object being processed.
+        config (MkDocsConfig): The MkDocs site configuration.
+        files (Files): The collection of files in the MkDocs build.
+
+    Returns:
+        str: The updated Markdown content with GitHub shorthand links replaced.
+    """
+    gh_icon = ":octicons-mark-github-16:"
+    gh_url = "https://github.com"
+    repo_url = f"{gh_url}/vllm-project/vllm-omni"
+    org_url = f"{gh_url}/orgs/vllm-project"
+
+    # Mapping of shorthand types to their corresponding GitHub base URLs
+    urls = {
+        "issue": f"{repo_url}/issues",
+        "pr": f"{repo_url}/pull",
+        "project": f"{org_url}/projects",
+        "dir": f"{repo_url}/tree/main",
+        "file": f"{repo_url}/blob/main",
+    }
+
+    # Default title prefixes for auto links
+    titles = {
+        "issue": "Issue #",
+        "pr": "Pull Request #",
+        "project": "Project #",
+        "dir": "",
+        "file": "",
+    }
+
+    # Regular expression to match GitHub shorthand links
+    scheme = r"gh-(?P<type>.+?):(?P<path>.+?)(#(?P<fragment>.+?))?"
+    inline_link = re.compile(r"\[(?P<title>[^\[]+?)\]\(" + scheme + r"\)")
+    auto_link = re.compile(f"<{scheme}>")
+
+    def replace_inline_link(match: re.Match) -> str:
+        """
+        Replaces a matched inline-style GitHub shorthand link
+        with a full Markdown link.
+
+        Example:
+            [My issue](gh-issue:123) → [:octicons-mark-github-16: My issue](https://github.com/vllm-project/vllm/issues/123)
+        """
+        url = f"{urls[match.group('type')]}/{match.group('path')}"
+        if fragment := match.group("fragment"):
+            url += f"#{fragment}"
+
+        return f"[{gh_icon} {match.group('title')}]({url})"
+
+    def replace_auto_link(match: re.Match) -> str:
+        """
+        Replaces a matched autolink-style GitHub shorthand
+        with a full Markdown link.
+
+        Example:
+            <gh-pr:456> → [:octicons-mark-github-16: Pull Request #456](https://github.com/vllm-project/vllm/pull/456)
+        """
+        type = match.group("type")
+        path = match.group("path")
+        title = f"{titles[type]}{path}"
+        url = f"{urls[type]}/{path}"
+        if fragment := match.group("fragment"):
+            url += f"#{fragment}"
+
+        return f"[{gh_icon} {title}]({url})"
+
+    # Replace both inline and autolinks
+    markdown = inline_link.sub(replace_inline_link, markdown)
+    markdown = auto_link.sub(replace_auto_link, markdown)
+
+    return markdown

docs/mkdocs/javascript/edit_and_feedback.js

@@ -0,0 +1,47 @@
+/**
+ * edit_and_feedback.js
+ *
+ * Enhances MkDocs Material docs pages by:
+ *
+ * 1. Adding a "Question? Give us feedback" link
+ *    below the "Edit" button.
+ *
+ *    - The link opens a GitHub issue with a template,
+ *      auto-filled with the current page URL and path.
+ *
+ * 2. Ensuring the edit button opens in a new tab
+ *    with target="_blank" and rel="noopener".
+ */
+document.addEventListener("DOMContentLoaded", function () {
+    const url = window.location.href;
+    const page = document.body.dataset.mdUrl || location.pathname;
+
+    const feedbackLink = document.createElement("a");
+    feedbackLink.href = `https://github.com/vllm-project/vllm-omni/issues/new?template=100-documentation.yml&title=${encodeURIComponent(
+      `[Docs] Feedback for \`${page}\``
+    )}&body=${encodeURIComponent(`📄 **Reference:**\n${url}\n\n📝 **Feedback:**\n_Your response_`)}`;
+    feedbackLink.target = "_blank";
+    feedbackLink.rel = "noopener";
+    feedbackLink.title = "Provide feedback";
+    feedbackLink.className = "md-content__button";
+    feedbackLink.innerHTML = `
+    <svg
+      xmlns="http://www.w3.org/2000/svg"
+      height="24px"
+      viewBox="0 -960 960 960"
+      width="24px"
+      fill="currentColor"
+    >
+      <path d="M280-280h280v-80H280v80Zm0-160h400v-80H280v80Zm0-160h400v-80H280v80Zm-80 480q-33 0-56.5-23.5T120-200v-560q0-33 23.5-56.5T200-840h560q33 0 56.5 23.5T840-760v560q0 33-23.5 56.5T760-120H200Zm0-80h560v-560H200v560Zm0-560v560-560Z"/>
+    </svg>
+  `;
+
+    const editButton = document.querySelector('.md-content__button[href*="edit"]');
+
+    if (editButton && editButton.parentNode) {
+      editButton.insertAdjacentElement("beforebegin", feedbackLink);
+
+      editButton.setAttribute("target", "_blank");
+      editButton.setAttribute("rel", "noopener");
+    }
+  });

docs/mkdocs/javascript/mathjax.js

@@ -0,0 +1,20 @@
+// Enables MathJax rendering
+window.MathJax = {
+    tex: {
+      inlineMath: [["\\(", "\\)"]],
+      displayMath: [["\\[", "\\]"]],
+      processEscapes: true,
+      processEnvironments: true
+    },
+    options: {
+      ignoreHtmlClass: ".*|",
+      processHtmlClass: "arithmatex"
+    }
+  };
+
+  document$.subscribe(() => {
+    MathJax.startup.output.clearCache()
+    MathJax.typesetClear()
+    MathJax.texReset()
+    MathJax.typesetPromise()
+  })

docs/mkdocs/javascript/slack_and_forum.js

@@ -0,0 +1,56 @@
+/**
+ * slack_and_forum.js
+ *
+ * Adds a custom Slack and Forum button to the MkDocs Material header.
+ *
+ */
+
+window.addEventListener('DOMContentLoaded', () => {
+    const headerInner = document.querySelector('.md-header__inner');
+
+    if (headerInner) {
+      const slackButton = document.createElement('button');
+      slackButton.className = 'slack-button';
+      slackButton.title = 'Join us on Slack';
+      slackButton.style.border = 'none';
+      slackButton.style.background = 'transparent';
+      slackButton.style.cursor = 'pointer';
+
+      slackButton.innerHTML = `
+        <img src="https://a.slack-edge.com/80588/marketing/img/icons/icon_slack_hash_colored.png"
+             style="height: 1.1rem;"
+             alt="Slack">
+      `;
+
+      slackButton.addEventListener('click', () => {
+        window.open('https://slack.vllm.ai', '_blank', 'noopener');
+      });
+
+      const forumButton = document.createElement('button');
+      forumButton.className = 'forum-button';
+      forumButton.title = 'Join the Forum';
+      forumButton.style.border = 'none';
+      forumButton.style.background = 'transparent';
+      forumButton.style.cursor = 'pointer';
+
+      forumButton.innerHTML = `
+        <svg
+          xmlns="http://www.w3.org/2000/svg"
+          viewBox="0 -960 960 960"
+          fill="currentColor"
+        >
+          <path d="M817.85-198.15 698.46-317.54H320q-24.48 0-41.47-16.99T261.54-376v-11.69h424.61q25.39 0 43.47-18.08 18.07-18.08 18.07-43.46v-268.92h11.69q24.48 0 41.47 16.99 17 16.99 17 41.47v461.54ZM179.08-434.69l66.84-66.85h363.31q10.77 0 17.69-6.92 6.93-6.92 6.93-17.69v-246.77q0-10.77-6.93-17.7-6.92-6.92-17.69-6.92H203.69q-10.77 0-17.69 6.92-6.92 6.93-6.92 17.7v338.23Zm-36.93 89.46v-427.69q0-25.39 18.08-43.46 18.08-18.08 43.46-18.08h405.54q25.39 0 43.46 18.08 18.08 18.07 18.08 43.46v246.77q0 25.38-18.08 43.46-18.07 18.07-43.46 18.07H261.54L142.15-345.23Zm36.93-180.92V-797.54v271.39Z"/>
+        </svg>
+      `;
+
+      forumButton.addEventListener('click', () => {
+        window.open('https://discuss.vllm.ai/', '_blank', 'noopener');
+      });
+
+      const githubSource = document.querySelector('.md-header__source');
+      if (githubSource) {
+        githubSource.parentNode.insertBefore(slackButton, githubSource.nextSibling);
+        githubSource.parentNode.insertBefore(forumButton, slackButton.nextSibling);
+      }
+    }
+  });

docs/mkdocs/overrides/main.html

@@ -0,0 +1 @@
+{% extends "base.html" %}

docs/mkdocs/overrides/partials/toc-item.html

@@ -0,0 +1,21 @@
+<!-- Enables the use of toc_depth in document frontmatter https://github.com/squidfunk/mkdocs-material/issues/4827#issuecomment-1869812019 -->
+<li class="md-nav__item">
+    <a href="{{ toc_item.url }}" class="md-nav__link">
+      <span class="md-ellipsis">
+        {{ toc_item.title }}
+      </span>
+    </a>
+
+    <!-- Table of contents list -->
+    {% if toc_item.children %}
+      <nav class="md-nav" aria-label="{{ toc_item.title | striptags }}">
+        <ul class="md-nav__list">
+          {% for toc_item in toc_item.children %}
+          {% if not page.meta.toc_depth or toc_item.level <= page.meta.toc_depth %}
+            {% include "partials/toc-item.html" %}
+          {% endif %}
+          {% endfor %}
+        </ul>
+      </nav>
+    {% endif %}
+  </li>

docs/mkdocs/stylesheets/extra.css

@@ -0,0 +1,218 @@
+/* Warning for latest docs */
+.md-banner {
+    background-color: var(--md-warning-bg-color);
+    color: var(--md-warning-fg-color);
+}
+
+/* https://christianoliff.com/blog/styling-external-links-with-an-icon-in-css/ */
+a:not(:has(svg)):not(.md-icon):not(.autorefs-external) {
+    align-items: center;
+
+    &[href^="//"]::after,
+    &[href^="http://"]::after,
+    &[href^="https://"]::after {
+        content: "";
+        width: 12px;
+        height: 12px;
+        margin-left: 4px;
+        background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='16' height='16' stroke='gray' viewBox='0 0 16 16'%3E%3Cpath fill-rule='evenodd' d='M8.636 3.5a.5.5 0 0 0-.5-.5H1.5A1.5 1.5 0 0 0 0 4.5v10A1.5 1.5 0 0 0 1.5 16h10a1.5 1.5 0 0 0 1.5-1.5V7.864a.5.5 0 0 0-1 0V14.5a.5.5 0 0 1-.5.5h-10a.5.5 0 0 1-.5-.5v-10a.5.5 0 0 1 .5-.5h6.636a.5.5 0 0 0 .5-.5z'/%3E%3Cpath fill-rule='evenodd' d='M16 .5a.5.5 0 0 0-.5-.5h-5a.5.5 0 0 0 0 1h3.793L6.146 9.146a.5.5 0 1 0 .708.708L15 1.707V5.5a.5.5 0 0 0 1 0v-5z'/%3E%3C/svg%3E");
+        background-position: center;
+        background-repeat: no-repeat;
+        background-size: contain;
+        display: inline-block;
+    }
+}
+
+a[href*="localhost"]::after,
+a[href*="127.0.0.1"]::after,
+a[href*="org.readthedocs.build"]::after,
+a[href*="docs.vllm.ai"]::after {
+    display: none !important;
+}
+
+/* Light mode: darker section titles */
+body[data-md-color-scheme="default"] .md-nav__item--section > label.md-nav__link .md-ellipsis {
+  color: rgba(0, 0, 0, 0.7) !important;
+  font-weight: 700;
+}
+
+/* Dark mode: lighter gray section titles */
+body[data-md-color-scheme="slate"] .md-nav__item--section > label.md-nav__link .md-ellipsis {
+  color: rgba(255, 255, 255, 0.75) !important;
+  font-weight: 700;
+}
+
+/* Custom admonitions */
+:root {
+  --md-admonition-icon--announcement: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" width="16" height="16"><path d="M3.25 9a.75.75 0 0 1 .75.75c0 2.142.456 3.828.733 4.653a.122.122 0 0 0 .05.064.212.212 0 0 0 .117.033h1.31c.085 0 .18-.042.258-.152a.45.45 0 0 0 .075-.366A16.743 16.743 0 0 1 6 9.75a.75.75 0 0 1 1.5 0c0 1.588.25 2.926.494 3.85.293 1.113-.504 2.4-1.783 2.4H4.9c-.686 0-1.35-.41-1.589-1.12A16.4 16.4 0 0 1 2.5 9.75.75.75 0 0 1 3.25 9Z"></path><path d="M0 6a4 4 0 0 1 4-4h2.75a.75.75 0 0 1 .75.75v6.5a.75.75 0 0 1-.75.75H4a4 4 0 0 1-4-4Zm4-2.5a2.5 2.5 0 1 0 0 5h2v-5Z"></path><path d="M15.59.082A.75.75 0 0 1 16 .75v10.5a.75.75 0 0 1-1.189.608l-.002-.001h.001l-.014-.01a5.775 5.775 0 0 0-.422-.25 10.63 10.63 0 0 0-1.469-.64C11.576 10.484 9.536 10 6.75 10a.75.75 0 0 1 0-1.5c2.964 0 5.174.516 6.658 1.043.423.151.787.302 1.092.443V2.014c-.305.14-.669.292-1.092.443C11.924 2.984 9.713 3.5 6.75 3.5a.75.75 0 0 1 0-1.5c2.786 0 4.826-.484 6.155-.957.665-.236 1.154-.47 1.47-.64.144-.077.284-.161.421-.25l.014-.01a.75.75 0 0 1 .78-.061Z"></path></svg>');
+  --md-admonition-icon--important: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" width="16" height="16"><path d="M4.47.22A.749.749 0 0 1 5 0h6c.199 0 .389.079.53.22l4.25 4.25c.141.14.22.331.22.53v6a.749.749 0 0 1-.22.53l-4.25 4.25A.749.749 0 0 1 11 16H5a.749.749 0 0 1-.53-.22L.22 11.53A.749.749 0 0 1 0 11V5c0-.199.079-.389.22-.53Zm.84 1.28L1.5 5.31v5.38l3.81 3.81h5.38l3.81-3.81V5.31L10.69 1.5ZM8 4a.75.75 0 0 1 .75.75v3.5a.75.75 0 0 1-1.5 0v-3.5A.75.75 0 0 1 8 4Zm0 8a1 1 0 1 1 0-2 1 1 0 0 1 0 2Z"></path></svg>');
+  --md-admonition-icon--code: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16"><path d="m11.28 3.22 4.25 4.25a.75.75 0 0 1 0 1.06l-4.25 4.25a.749.749 0 0 1-1.275-.326.75.75 0 0 1 .215-.734L13.94 8l-3.72-3.72a.749.749 0 0 1 .326-1.275.75.75 0 0 1 .734.215m-6.56 0a.75.75 0 0 1 1.042.018.75.75 0 0 1 .018 1.042L2.06 8l3.72 3.72a.749.749 0 0 1-.326 1.275.75.75 0 0 1-.734-.215L.47 8.53a.75.75 0 0 1 0-1.06Z"/></svg>');
+  --md-admonition-icon--console: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16"><path d="M0 2.75C0 1.784.784 1 1.75 1h12.5c.966 0 1.75.784 1.75 1.75v10.5A1.75 1.75 0 0 1 14.25 15H1.75A1.75 1.75 0 0 1 0 13.25Zm1.75-.25a.25.25 0 0 0-.25.25v10.5c0 .138.112.25.25.25h12.5a.25.25 0 0 0 .25-.25V2.75a.25.25 0 0 0-.25-.25ZM7.25 8a.75.75 0 0 1-.22.53l-2.25 2.25a.749.749 0 0 1-1.275-.326.75.75 0 0 1 .215-.734L5.44 8 3.72 6.28a.749.749 0 0 1 .326-1.275.75.75 0 0 1 .734.215l2.25 2.25c.141.14.22.331.22.53m1.5 1.5h3a.75.75 0 0 1 0 1.5h-3a.75.75 0 0 1 0-1.5"/></svg>');
+}
+
+.md-typeset .admonition.announcement,
+.md-typeset details.announcement {
+  border-color: rgb(255, 110, 66);
+}
+.md-typeset .admonition.important,
+.md-typeset details.important {
+  border-color: rgb(239, 85, 82);
+}
+.md-typeset .admonition.code,
+.md-typeset details.code {
+  border-color: #64dd17
+}
+.md-typeset .admonition.console,
+.md-typeset details.console {
+  border-color: #64dd17
+}
+
+.md-typeset .announcement > .admonition-title,
+.md-typeset .announcement > summary {
+  background-color: rgb(255, 110, 66, 0.1);
+}
+.md-typeset .important > .admonition-title,
+.md-typeset .important > summary {
+  background-color: rgb(239, 85, 82, 0.1);
+}
+.md-typeset .code > .admonition-title,
+.md-typeset .code > summary {
+  background-color: #64dd171a;
+}
+.md-typeset .console > .admonition-title,
+.md-typeset .console > summary {
+  background-color: #64dd171a;
+}
+
+.md-typeset .announcement > .admonition-title::before,
+.md-typeset .announcement > summary::before {
+  background-color: rgb(239, 85, 82);
+  -webkit-mask-image: var(--md-admonition-icon--announcement);
+          mask-image: var(--md-admonition-icon--announcement);
+}
+.md-typeset .important > .admonition-title::before,
+.md-typeset .important > summary::before {
+  background-color: rgb(239, 85, 82);
+  -webkit-mask-image: var(--md-admonition-icon--important);
+          mask-image: var(--md-admonition-icon--important);
+}
+.md-typeset .code > .admonition-title::before,
+.md-typeset .code > summary::before {
+  background-color: #64dd17;
+  -webkit-mask-image: var(--md-admonition-icon--code);
+          mask-image: var(--md-admonition-icon--code);
+}
+.md-typeset .console > .admonition-title::before,
+.md-typeset .console > summary::before {
+  background-color: #64dd17;
+  -webkit-mask-image: var(--md-admonition-icon--console);
+          mask-image: var(--md-admonition-icon--console);
+}
+
+/* Make label fully visible on hover */
+.md-content__button[href*="edit"]:hover::after {
+  opacity: 1;
+}
+
+/* Hide edit button on generated docs/examples pages */
+@media (min-width: 960px) {
+  .md-content__button[href*="docs/examples/"] {
+    display: none !important;
+  }
+}
+
+.md-content__button-wrapper {
+  position: absolute;
+  top: 0.6rem;
+  right: 0.8rem;
+  display: flex;
+  flex-direction: row;
+  align-items: center;
+  gap: 0.4rem;
+  z-index: 1;
+}
+
+.md-content__button-wrapper a {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  height: 24px;
+  width: 24px;
+  color: var(--md-default-fg-color);
+  text-decoration: none;
+}
+
+.md-content__button-wrapper a:hover {
+  color: var(--md-accent-fg-color);
+}
+
+/* Slack and Forum css */
+.slack-button,
+.forum-button {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  margin-left: 0.4rem;
+  height: 24px;
+}
+
+.slack-button img {
+  height: 18px;
+  filter: none !important;
+}
+
+.slack-button:hover,
+.forum-button:hover {
+  opacity: 0.7;
+}
+
+.forum-button svg {
+  height: 28px;
+  opacity: 0.9;
+  transform: translateY(2px);
+}
+
+/* For logo css */
+[data-md-color-scheme="default"] .logo-dark {
+  display: none;
+}
+
+[data-md-color-scheme="slate"] .logo-light {
+  display: none;
+}
+
+/* Outline for content tabs */
+.md-typeset .tabbed-set {
+  border: 0.075rem solid var(--md-default-fg-color);
+  border-radius: 0.2rem;
+}
+
+.md-typeset .tabbed-content {
+  padding: 0 0.6em;
+}
+
+/* Hide link icon in header logo */
+.md-header__button.md-logo :is(img, svg) {
+  pointer-events: none;
+}
+
+.md-header__button.md-logo::after {
+  display: none !important;
+}
+
+/* Hide link icons in content tabs (tabbed content) */
+.md-typeset .tabbed-set > label::after,
+.md-typeset .tabbed-set > input:checked + label::after {
+  display: none !important;
+}
+
+/* Hide link icons in navigation tabs */
+.md-nav__link[href]::after,
+.md-nav__item--nested > .md-nav__link::after {
+  display: none !important;
+}
+
+/* Hide link icons in top navigation tabs */
+.md-tabs__link::after {
+  display: none !important;
+}

docs/models/supported_models.md

@@ -0,0 +1,35 @@
+# Supported Models
+
+vLLM-Omni supports unified multimodal comprehension and generation models across various tasks.
+
+## Model Implementation
+
+If vLLM-Omni natively supports a model, its implementation can be found in <gh-file:vllm-omni/model_executor/models> and <gh-file:vllm_omni/diffusion/models>.
+
+## List of Supported Models for Nvidia GPU
+
+<style>
+th {
+  white-space: nowrap;
+  min-width: 0 !important;
+}
+</style>
+
+| Architecture | Models | Example HF Models |
+|--------------|--------|-------------------|
+| `Qwen2_5OmniForConditionalGeneration` | Qwen2.5-Omni | `Qwen/Qwen2.5-Omni-7B`, `Qwen/Qwen2.5-Omni-3B` |
+| `QwenImagePipeline` | Qwen-Image | `Qwen/Qwen-Image` |
+
+
+## List of Supported Models for Ascend NPU
+
+<style>
+th {
+  white-space: nowrap;
+  min-width: 0 !important;
+}
+</style>
+
+| Architecture | Models | Example HF Models |
+|--------------|--------|-------------------|
+| `Qwen2_5OmniForConditionalGeneration` | Qwen2.5-Omni | `Qwen/Qwen2.5-Omni-7B`, `Qwen/Qwen2.5-Omni-3B`|

docs/usage/faq.md

@@ -0,0 +1,13 @@
+# Frequently Asked Questions
+
+> Q: How many chips do I need to infer a model in vLLM-Omni?
+
+A: Now, we support natively disaggregated deployment for different model stages within a model. There is a restriction that one chip can only has one AutoRegressive model stage. This is because the unified KV cache management of vLLM. Stages of other types can coexist with in a chip. The restriction will be resolved in later version.
+
+> Q: When trying to run examples, I encounter error about backend of librosa or soundfile. How to solve it?
+
+A: If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```

docs/user_guide/examples/offline_inference/qwen2_5_omni.md

@@ -0,0 +1,60 @@
+# Offline Inference Example of vLLM-Omni for Qwen2.5-Omni
+
+Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen2_5_omni>.
+
+
+## 🛠️ Installation
+
+Please refer to [installation](../../../getting_started/installation/README.md).
+
+## Run Offline inference with Qwen2.5-Omni.
+First, navigate to the example folder
+```bash
+cd examples/offline_inference/qwen2_5_omni
+```
+Inside the directory, `end2end.py` is a comprehensive demo suite for initializing a model instance of `Qwen/Qwen2.5-Omni-7B` and use it for various offline inference tasks.
+??? abstract "end2end.py"
+    ``````py
+    --8<-- "examples/offline_inference/qwen2_5_omni/end2end.py"
+    ``````
+
+Below we also provide simple bash scripts to execute this file.
+### Single Prompt
+```bash
+bash run_single_prompt.sh
+```
+??? abstract "run_single_prompt.sh"
+    ``````sh
+    --8<-- "examples/offline_inference/qwen2_5_omni/run_single_prompt.sh"
+    ``````
+
+### Multiple Prompts
+Download dataset from [seed_tts](https://drive.google.com/file/d/1GlSjVfSHkW3-leKKBlfrjuuTGqQ_xaLP/edit) and extract the prompts with `extract_prompts.py`
+??? abstract "extract_prompts.py"
+    ``````py
+    --8<-- "examples/offline_inference/qwen2_5_omni/extract_prompts.py"
+    ``````
+```bash
+tar -xf <Your Download Path>/seedtts_testset.tar
+cp seedtts_testset/en/meta.lst examples/offline_inference/qwen2_5_omni/meta.lst
+python3 examples/offline_inference/qwen2_5_omni/extract_prompts.py \
+  --input examples/offline_inference/qwen2_5_omni/meta.lst \
+  --output examples/offline_inference/qwen2_5_omni/top100.txt \
+  --topk 10
+```
+Then run the command below.
+```bash
+bash run_multiple_prompts.sh
+```
+??? abstract "run_multiple_prompts.sh"
+    ``````sh
+    --8<-- "examples/offline_inference/qwen2_5_omni/run_multiple_prompts.sh"
+    ``````
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```

docs/user_guide/examples/online_serving/gradio_demo.md

@@ -0,0 +1,7 @@
+# Gradio Demo
+
+Source <https://github.com/vllm-project/vllm-omni/blob/main/examples/online_serving/gradio_demo.py>.
+
+``````py
+--8<-- "examples/online_serving/gradio_demo.py"
+``````

docs/user_guide/examples/online_serving/openai_chat_completion_client_for_multimodal_generation.md

@@ -0,0 +1,7 @@
+# OpenAI Chat Completion Client For Multimodal Generation
+
+Source <https://github.com/vllm-project/vllm-omni/blob/main/examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py>.
+
+``````py
+--8<-- "examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py"
+``````

docs/user_guide/examples/online_serving/qwen2_5_omni.md

@@ -0,0 +1,80 @@
+# Online Serving Example of vLLM-Omni for Qwen2.5-Omni
+
+Source <https://github.com/vllm-project/vllm-omni/blob/main/examples/online_serving/README.md>.
+
+
+## 🛠️ Installation
+
+Please refer to [installation](../../../getting_started/installation/README.md).
+
+## Deploy Qwen/Qwen2.5-Omni-7B
+
+First launch the OpenAI-compatible inference server
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091
+```
+If you have custom stage configs file, launch the server with command below
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
+```
+## Query the model
+Navigate to the example folder
+```bash
+cd examples/online_serving
+```
+Query the model server with OpenAI Python API client:
+```bash
+python openai_chat_completion_client_for_multimodal_generation.py --query-type mixed_modalities
+```
+??? abstract "openai_chat_completion_client_for_multimodal_generation.py"
+    ``````py
+    --8<-- "examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py"
+    ``````
+You can also query the model with `curl` command:
+```bash
+bash run_curl_multimodal_generation.sh mixed_modalities
+```
+??? abstract "run_curl_multimodal_generation.sh"
+    ``````py
+    --8<-- "examples/online_serving/run_curl_multimodal_generation.sh"
+    ``````
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```
+
+## Run Local Web UI Demo
+
+You can also deploy a Gradio Web UI that allows users to interact with the model through a web browser. Below is an example on how to do so with `Qwen/Qwen2.5-Omni-7B`.
+
+### Running Gradio Demo
+
+Install gradio with `uv pip install "gradio>=5.49.1,<6.0.0"`, then you can launch the web service built on AsyncOmni by
+
+```bash
+python gradio_demo.py  --model Qwen/Qwen2.5-Omni-7B --port 7861
+```
+Now you can interact with model via the web UI at `http://localhost:7861/` on your local browser.
+
+
+### Options
+
+You can customize its basic launch parameters:
+
+```bash
+python gradio_demo.py \
+    --model Qwen/Qwen2.5-Omni-7B \
+    --ip 127.0.0.1 \
+    --port 7861 \
+    --stage-configs-path /path/to/stage_configs.yaml
+```
+
+- `--model`: Local model checkpoint to load (default `Qwen/Qwen2.5-Omni-7B`).
+- `--ip`: Host/IP for the Gradio server (default `127.0.0.1`).
+- `--port`: Port for the Gradio server (default `7861`).
+- `--stage-configs-path`: Optional path to custom stage configs YAML.
+- `--share`: Set to expose a temporary public link via Gradio.

examples/offline_inference/qwen2_5_omni/README.md

@@ -0,0 +1,42 @@
+# Offline Example of vLLM-Omni for Qwen2.5-omni
+
+## 🛠️ Installation
+
+Please refer to [README.md](../../../README.md)
+
+## Run examples (Qwen2.5-omni)
+### Multiple Prompts
+Download dataset from [seed_tts](https://drive.google.com/file/d/1GlSjVfSHkW3-leKKBlfrjuuTGqQ_xaLP/edit). To get the prompt, you can:
+```bash
+tar -xf <Your Download Path>/seedtts_testset.tar
+cp seedtts_testset/en/meta.lst examples/offline_inference/qwen2_5_omni/meta.lst
+python3 examples/offline_inference/qwen2_5_omni/extract_prompts.py \
+  --input examples/offline_inference/qwen2_5_omni/meta.lst \
+  --output examples/offline_inference/qwen2_5_omni/top100.txt \
+  --topk 100
+```
+Get into the example folder
+```bash
+cd examples/offline_inference/qwen2_5_omni
+```
+Then run the command below.
+```bash
+bash run_multiple_prompts.sh
+```
+### Single Prompts
+Get into the example folder
+```bash
+cd examples/offline_inference/qwen2_5_omni
+```
+Then run the command below.
+```bash
+bash run_single_prompt.sh
+```
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```

examples/offline_inference/qwen2_5_omni/end2end.py

@@ -0,0 +1,292 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""
+This example shows how to use vLLM-Omni for running offline inference
+with the correct prompt format on Qwen2.5-Omni
+"""
+
+import os
+from typing import NamedTuple
+
+import soundfile as sf
+from vllm.assets.audio import AudioAsset
+from vllm.assets.image import ImageAsset
+from vllm.assets.video import VideoAsset
+from vllm.multimodal.image import convert_image_mode
+from vllm.sampling_params import SamplingParams
+from vllm.utils import FlexibleArgumentParser
+
+from vllm_omni.entrypoints.omni_llm import OmniLLM
+
+SEED = 42
+
+
+class QueryResult(NamedTuple):
+    inputs: dict
+    limit_mm_per_prompt: dict[str, int]
+
+
+# NOTE: The default `max_num_seqs` and `max_model_len` may result in OOM on
+# lower-end GPUs.
+# Unless specified, these settings have been tested to work on a single L4.
+
+default_system = (
+    "You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
+    "Group, capable of perceiving auditory and visual inputs, as well as "
+    "generating text and speech."
+)
+
+
+def get_text_query(question: str = None) -> QueryResult:
+    if question is None:
+        question = "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+        },
+        limit_mm_per_prompt={},
+    )
+
+
+def get_mixed_modalities_query() -> QueryResult:
+    question = "What is recited in the audio? What is the content of this image? Why is this video funny?"
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|audio_bos|><|AUDIO|><|audio_eos|>"
+        "<|vision_bos|><|IMAGE|><|vision_eos|>"
+        "<|vision_bos|><|VIDEO|><|vision_eos|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "audio": AudioAsset("mary_had_lamb").audio_and_sample_rate,
+                "image": convert_image_mode(ImageAsset("cherry_blossom").pil_image, "RGB"),
+                "video": VideoAsset(name="baby_reading", num_frames=16).np_ndarrays,
+            },
+        },
+        limit_mm_per_prompt={"audio": 1, "image": 1, "video": 1},
+    )
+
+
+def get_use_audio_in_video_query() -> QueryResult:
+    question = "Describe the content of the video, then convert what the baby say into text."
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|vision_bos|><|VIDEO|><|vision_eos|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    asset = VideoAsset(name="baby_reading", num_frames=16)
+    audio = asset.get_audio(sampling_rate=16000)
+
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "video": asset.np_ndarrays,
+                "audio": audio,
+            },
+            "mm_processor_kwargs": {
+                "use_audio_in_video": True,
+            },
+        },
+        limit_mm_per_prompt={"audio": 1, "video": 1},
+    )
+
+
+def get_multi_audios_query() -> QueryResult:
+    question = "Are these two audio clips the same?"
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|audio_bos|><|AUDIO|><|audio_eos|>"
+        "<|audio_bos|><|AUDIO|><|audio_eos|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "audio": [
+                    AudioAsset("winning_call").audio_and_sample_rate,
+                    AudioAsset("mary_had_lamb").audio_and_sample_rate,
+                ],
+            },
+        },
+        limit_mm_per_prompt={
+            "audio": 2,
+        },
+    )
+
+
+query_map = {
+    "mixed_modalities": get_mixed_modalities_query,
+    "use_audio_in_video": get_use_audio_in_video_query,
+    "multi_audios": get_multi_audios_query,
+    "text": get_text_query,
+}
+
+
+def main(args):
+    model_name = "Qwen/Qwen2.5-Omni-7B"
+    query_result = query_map[args.query_type]()
+
+    omni_llm = OmniLLM(
+        model=model_name,
+        log_stats=args.enable_stats,
+        log_file=("omni_llm_pipeline.log" if args.enable_stats else None),
+        init_sleep_seconds=args.init_sleep_seconds,
+        batch_timeout=args.batch_timeout,
+        init_timeout=args.init_timeout,
+        shm_threshold_bytes=args.shm_threshold_bytes,
+    )
+    thinker_sampling_params = SamplingParams(
+        temperature=0.0,  # Deterministic - no randomness
+        top_p=1.0,  # Disable nucleus sampling
+        top_k=-1,  # Disable top-k sampling
+        max_tokens=2048,
+        seed=SEED,  # Fixed seed for sampling
+        detokenize=True,
+        repetition_penalty=1.1,
+    )
+    talker_sampling_params = SamplingParams(
+        temperature=0.9,
+        top_p=0.8,
+        top_k=40,
+        max_tokens=2048,
+        seed=SEED,  # Fixed seed for sampling
+        detokenize=True,
+        repetition_penalty=1.05,
+        stop_token_ids=[8294],
+    )
+    code2wav_sampling_params = SamplingParams(
+        temperature=0.0,  # Deterministic - no randomness
+        top_p=1.0,  # Disable nucleus sampling
+        top_k=-1,  # Disable top-k sampling
+        max_tokens=2048,
+        seed=SEED,  # Fixed seed for sampling
+        detokenize=True,
+        repetition_penalty=1.1,
+    )
+
+    sampling_params_list = [
+        thinker_sampling_params,
+        talker_sampling_params,
+        code2wav_sampling_params,
+    ]
+
+    if args.txt_prompts is None:
+        prompts = [query_result.inputs for _ in range(args.num_prompts)]
+    else:
+        assert args.query_type == "text", "txt-prompts is only supported for text query type"
+        with open(args.txt_prompts, encoding="utf-8") as f:
+            lines = [ln.strip() for ln in f.readlines()]
+            prompts = [get_text_query(ln).inputs for ln in lines if ln != ""]
+            print(f"[Info] Loaded {len(prompts)} prompts from {args.txt_prompts}")
+
+    omni_outputs = omni_llm.generate(prompts, sampling_params_list)
+
+    # Determine output directory: prefer --output-dir; fallback to --output-wav
+    output_dir = args.output_dir if getattr(args, "output_dir", None) else args.output_wav
+    os.makedirs(output_dir, exist_ok=True)
+    for stage_outputs in omni_outputs:
+        if stage_outputs.final_output_type == "text":
+            for output in stage_outputs.request_output:
+                request_id = int(output.request_id)
+                text_output = output.outputs[0].text
+                # Save aligned text file per request
+                prompt_text = prompts[request_id]["prompt"]
+                out_txt = os.path.join(output_dir, f"{request_id:05d}.txt")
+                lines = []
+                lines.append("Prompt:\n")
+                lines.append(str(prompt_text) + "\n")
+                lines.append("vllm_text_output:\n")
+                lines.append(str(text_output).strip() + "\n")
+                try:
+                    with open(out_txt, "w", encoding="utf-8") as f:
+                        f.writelines(lines)
+                except Exception as e:
+                    print(f"[Warn] Failed writing text file {out_txt}: {e}")
+                print(f"Request ID: {request_id}, Text saved to {out_txt}")
+        elif stage_outputs.final_output_type == "audio":
+            for output in stage_outputs.request_output:
+                request_id = int(output.request_id)
+                audio_tensor = output.multimodal_output["audio"]
+                output_wav = os.path.join(output_dir, f"output_{output.request_id}.wav")
+                sf.write(output_wav, audio_tensor.detach().cpu().numpy(), samplerate=24000)
+                print(f"Request ID: {request_id}, Saved audio to {output_wav}")
+
+
+def parse_args():
+    parser = FlexibleArgumentParser(description="Demo on using vLLM for offline inference with audio language models")
+    parser.add_argument(
+        "--query-type",
+        "-q",
+        type=str,
+        default="mixed_modalities",
+        choices=query_map.keys(),
+        help="Query type.",
+    )
+    parser.add_argument(
+        "--enable-stats",
+        action="store_true",
+        default=False,
+        help="Enable writing detailed statistics (default: disabled)",
+    )
+    parser.add_argument(
+        "--init-sleep-seconds",
+        type=int,
+        default=20,
+        help="Sleep seconds after starting each stage process to allow initialization (default: 20)",
+    )
+    parser.add_argument(
+        "--batch-timeout",
+        type=int,
+        default=5,
+        help="Timeout for batching in seconds (default: 5)",
+    )
+    parser.add_argument(
+        "--init-timeout",
+        type=int,
+        default=300,
+        help="Timeout for initializing stages in seconds (default: 300)",
+    )
+    parser.add_argument(
+        "--shm-threshold-bytes",
+        type=int,
+        default=65536,
+        help="Threshold for using shared memory in bytes (default: 65536)",
+    )
+    parser.add_argument(
+        "--output-wav",
+        default="output_audio",
+        help="[Deprecated] Output wav directory (use --output-dir).",
+    )
+    parser.add_argument(
+        "--num-prompts",
+        type=int,
+        default=1,
+        help="Number of prompts to generate.",
+    )
+    parser.add_argument(
+        "--txt-prompts",
+        type=str,
+        default=None,
+        help="Path to a .txt file with one prompt per line (preferred).",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

examples/offline_inference/qwen2_5_omni/extract_prompts.py

@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+import argparse
+from typing import Optional
+
+
+def extract_prompt(line: str) -> Optional[str]:
+    # 提取第一个 '|' 与第二个 '|' 之间的内容
+    i = line.find("|")
+    if i == -1:
+        return None
+    j = line.find("|", i + 1)
+    if j == -1:
+        return None
+    return line[i + 1 : j].strip()
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--input", "-i", required=True, help="Input .lst file path")
+    parser.add_argument("--output", "-o", required=True, help="Output file path")
+    parser.add_argument(
+        "--topk",
+        "-k",
+        type=int,
+        default=100,
+        help="Extract the top K prompts (default: 100)",
+    )
+    args = parser.parse_args()
+
+    prompts = []
+    with open(args.input, encoding="utf-8", errors="ignore") as f:
+        for line in f:
+            if len(prompts) >= args.topk:
+                break
+            p = extract_prompt(line.rstrip("\n"))
+            if p:
+                prompts.append(p)
+
+    with open(args.output, "w", encoding="utf-8") as f:
+        for p in prompts:
+            f.write(p + "\n")
+
+
+if __name__ == "__main__":
+    main()

examples/offline_inference/qwen2_5_omni/run_multiple_prompts.sh

@@ -0,0 +1,3 @@
+python end2end.py --output-wav output_audio \
+                  --query-type text \
+                  --txt-prompts top10.txt

examples/offline_inference/qwen2_5_omni/run_single_prompt.sh

@@ -0,0 +1,2 @@
+python end2end.py --output-wav output_audio \
+                  --query-type use_audio_in_video

examples/offline_inference/qwen3_omni/README.md

@@ -0,0 +1,34 @@
+# Offline Example of vLLM-Omni for Qwen2.5-omni
+
+## 🛠️ Installation
+
+Please refer to [README.md](../../../README.md)
+
+## Run examples (Qwen3-omni)
+### Multiple Prompts
+Download dataset from [seed_tts](https://drive.google.com/file/d/1GlSjVfSHkW3-leKKBlfrjuuTGqQ_xaLP/edit). For processing dataset please refer to [Qwen2.5-omni README.md](../qwen2_5_omni/README.md)
+Get into the example folder
+```bash
+cd examples/offline_inference/qwen3_omni
+```
+Then run the command below.
+```bash
+bash run_multiple_prompts.sh
+```
+### Single Prompts
+Get into the example folder
+```bash
+cd examples/offline_inference/qwen3_omni
+```
+Then run the command below.
+```bash
+bash run_single_prompt.sh
+```
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```

examples/offline_inference/qwen3_omni/end2end.py

@@ -0,0 +1,289 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""
+This example shows how to use vLLM for running offline inference
+with the correct prompt format on Qwen3-Omni (thinker only).
+"""
+
+import os
+from typing import NamedTuple
+
+import soundfile as sf
+from vllm import SamplingParams
+from vllm.assets.audio import AudioAsset
+from vllm.assets.image import ImageAsset
+from vllm.assets.video import VideoAsset
+from vllm.multimodal.image import convert_image_mode
+from vllm.utils import FlexibleArgumentParser
+
+from vllm_omni.entrypoints.omni_llm import OmniLLM
+
+SEED = 42
+
+
+class QueryResult(NamedTuple):
+    inputs: dict
+    limit_mm_per_prompt: dict[str, int]
+
+
+# NOTE: The default `max_num_seqs` and `max_model_len` may result in OOM on
+# lower-end GPUs.
+# Unless specified, these settings have been tested to work on a single L4.
+
+default_system = (
+    "You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
+    "Group, capable of perceiving auditory and visual inputs, as well as "
+    "generating text and speech."
+)
+
+
+def get_text_query(question: str = None) -> QueryResult:
+    if question is None:
+        question = "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+        },
+        limit_mm_per_prompt={},
+    )
+
+
+def get_video_query(question: str = None) -> QueryResult:
+    if question is None:
+        question = "Why is this video funny?"
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|vision_start|><|video_pad|><|vision_end|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "video": VideoAsset(name="baby_reading", num_frames=16).np_ndarrays,
+            },
+        },
+        limit_mm_per_prompt={"video": 1},
+    )
+
+
+def get_image_query(question: str = None) -> QueryResult:
+    if question is None:
+        question = "What is the content of this image?"
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "image": convert_image_mode(ImageAsset("cherry_blossom").pil_image, "RGB"),
+            },
+        },
+        limit_mm_per_prompt={"image": 1},
+    )
+
+
+def get_audio_query(question: str = None) -> QueryResult:
+    if question is None:
+        question = "What is the content of this audio?"
+    prompt = (
+        f"<|im_start|>system\n{default_system}<|im_end|>\n"
+        "<|im_start|>user\n<|audio_start|><|audio_pad|><|audio_end|>"
+        f"{question}<|im_end|>\n"
+        f"<|im_start|>assistant\n"
+    )
+    return QueryResult(
+        inputs={
+            "prompt": prompt,
+            "multi_modal_data": {
+                "audio": AudioAsset("mary_had_lamb").audio_and_sample_rate,
+            },
+        },
+        limit_mm_per_prompt={"audio": 1},
+    )
+
+
+query_map = {
+    "text": get_text_query,
+    "use_audio": get_audio_query,
+    "use_image": get_image_query,
+    "use_video": get_video_query,
+}
+
+
+def main(args):
+    model_name = "Qwen/Qwen3-Omni-30B-A3B-Instruct"
+    query_result = query_map[args.query_type]()
+
+    omni_llm = OmniLLM(
+        model=model_name,
+    )
+
+    thinker_sampling_params = SamplingParams(
+        temperature=0.4,
+        top_p=0.9,
+        top_k=-1,
+        max_tokens=1200,
+        repetition_penalty=1.05,
+        logit_bias={},
+        seed=SEED,
+    )
+
+    talker_sampling_params = SamplingParams(
+        temperature=0.9,
+        top_k=50,
+        max_tokens=4096,
+        seed=SEED,
+        detokenize=False,
+        repetition_penalty=1.05,
+        stop_token_ids=[2150],  # TALKER_CODEC_EOS_TOKEN_ID
+    )
+
+    # Sampling parameters for Code2Wav stage (audio generation)
+    code2wav_sampling_params = SamplingParams(
+        temperature=0.0,
+        top_p=1.0,
+        top_k=-1,
+        max_tokens=4096 * 16,
+        seed=SEED,
+        detokenize=True,
+        repetition_penalty=1.1,
+    )
+
+    sampling_params_list = [
+        thinker_sampling_params,
+        talker_sampling_params,  # code predictor is integrated into talker for Qwen3 Omni
+        code2wav_sampling_params,
+    ]
+
+    if args.txt_prompts is None:
+        prompts = [query_result.inputs for _ in range(args.num_prompts)]
+    else:
+        assert args.query_type == "text", "txt-prompts is only supported for text query type"
+        with open(args.txt_prompts, encoding="utf-8") as f:
+            lines = [ln.strip() for ln in f.readlines()]
+            prompts = [get_text_query(ln).inputs for ln in lines if ln != ""]
+            print(f"[Info] Loaded {len(prompts)} prompts from {args.txt_prompts}")
+
+    omni_outputs = omni_llm.generate(prompts, sampling_params_list)
+    # Determine output directory: prefer --output-dir; fallback to --output-wav
+    output_dir = args.output_dir if getattr(args, "output_dir", None) else args.output_wav
+    os.makedirs(output_dir, exist_ok=True)
+
+    for stage_outputs in omni_outputs:
+        if stage_outputs.final_output_type == "text":
+            for output in stage_outputs.request_output:
+                request_id = int(output.request_id)
+                text_output = output.outputs[0].text
+                # Save aligned text file per request
+                prompt_text = prompts[request_id]["prompt"]
+                out_txt = os.path.join(output_dir, f"{request_id:05d}.txt")
+                lines = []
+                lines.append("Prompt:\n")
+                lines.append(str(prompt_text) + "\n")
+                lines.append("vllm_text_output:\n")
+                lines.append(str(text_output).strip() + "\n")
+                try:
+                    with open(out_txt, "w", encoding="utf-8") as f:
+                        f.writelines(lines)
+                except Exception as e:
+                    print(f"[Warn] Failed writing text file {out_txt}: {e}")
+                print(f"Request ID: {request_id}, Text saved to {out_txt}")
+        elif stage_outputs.final_output_type == "audio":
+            for output in stage_outputs.request_output:
+                request_id = int(output.request_id)
+                audio_tensor = output.multimodal_output["audio"]
+                output_wav = os.path.join(output_dir, f"output_{output.request_id}.wav")
+
+                # Convert to numpy array and ensure correct format
+                audio_numpy = audio_tensor.float().detach().cpu().numpy()
+
+                # Ensure audio is 1D (flatten if needed)
+                if audio_numpy.ndim > 1:
+                    audio_numpy = audio_numpy.flatten()
+
+                # Save audio file with explicit WAV format
+                sf.write(output_wav, audio_numpy, samplerate=24000, format="WAV")
+                print(f"Request ID: {request_id}, Saved audio to {output_wav}")
+
+
+def parse_args():
+    parser = FlexibleArgumentParser(description="Demo on using vLLM for offline inference with audio language models")
+    parser.add_argument(
+        "--query-type",
+        "-q",
+        type=str,
+        default="mixed_modalities",
+        choices=query_map.keys(),
+        help="Query type.",
+    )
+    parser.add_argument(
+        "--enable-stats",
+        action="store_true",
+        default=False,
+        help="Enable writing detailed statistics (default: disabled)",
+    )
+    parser.add_argument(
+        "--init-sleep-seconds",
+        type=int,
+        default=20,
+        help="Sleep seconds after starting each stage process to allow initialization (default: 20)",
+    )
+    parser.add_argument(
+        "--batch-timeout",
+        type=int,
+        default=5,
+        help="Timeout for batching in seconds (default: 5)",
+    )
+    parser.add_argument(
+        "--init-timeout",
+        type=int,
+        default=300,
+        help="Timeout for initializing stages in seconds (default: 300)",
+    )
+    parser.add_argument(
+        "--shm-threshold-bytes",
+        type=int,
+        default=65536,
+        help="Threshold for using shared memory in bytes (default: 65536)",
+    )
+    parser.add_argument(
+        "--output-wav",
+        default="output_audio",
+        help="[Deprecated] Output wav directory (use --output-dir).",
+    )
+    parser.add_argument(
+        "--num-prompts",
+        type=int,
+        default=1,
+        help="Number of prompts to generate.",
+    )
+    parser.add_argument(
+        "--txt-prompts",
+        type=str,
+        default=None,
+        help="Path to a .txt file with one prompt per line (preferred).",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)
+
+    # use examples:
+    # python end2end.py --model Qwen/Qwen3-Omni-30B-A3B-Instruct --query-type text
+    # python end2end.py --model /custom_path/Qwen3-Omni-30B-A3B-Instruct --query-type use_video_only

examples/offline_inference/qwen3_omni/run_multiple_prompts.sh

@@ -0,0 +1,3 @@
+python end2end.py --output-wav output_audio \
+                  --query-type text \
+                  --txt-prompts top10.txt

examples/offline_inference/qwen3_omni/run_single_prompt.sh

@@ -0,0 +1,2 @@
+python end2end.py --output-wav output_audio \
+                  --query-type use_audio

examples/offline_inference/qwen_image/README.md

@@ -0,0 +1,43 @@
+# Qwen-Image Offline Inference
+
+This folder provides two simple entrypoints for experimenting with `Qwen/Qwen-Image` using vLLM-Omni:
+
+- `text_to_image.py`: command-line script for single image generation.
+- `web_demo.py`: lightweight Gradio UI for interactive prompt/seed/CFG exploration.
+
+
+## Local CLI Usage
+
+```bash
+python text_to_image.py \
+  --prompt "a cup of coffee on the table" \
+  --seed 42 \
+  --cfg_scale 4.0 \
+  --num_images_per_prompt 1 \
+  --num_inference_steps 50 \
+  --height 1024 \
+  --width 1024 \
+  --output outputs/coffee.png
+```
+
+Key arguments:
+
+- `--prompt`: text description (string).
+- `--seed`: integer seed for deterministic sampling.
+- `--cfg_scale`: true CFG scale (model-specific guidance strength).
+- `--num_images_per_prompt`: number of images to generate per prompt (saves as `output`, `output_1`, ...).
+- `--num_inference_steps`: diffusion sampling steps (more steps = higher quality, slower).
+- `--height/--width`: output resolution (defaults 1024x1024).
+- `--output`: path to save the generated PNG.
+
+> ℹ️ Qwen-Image currently publishes best-effort presets at `1328x1328`, `1664x928`, `928x1664`, `1472x1140`, `1140x1472`, `1584x1056`, and `1056x1584`. Adjust `--height/--width` accordingly for the most reliable outcomes.
+
+## Web UI Demo
+
+Launch the gradio demo:
+
+```bash
+python gradio_demo.py --port 7862
+```
+
+Then open `http://localhost:7862/` on your local browser to interact with the web UI.

examples/offline_inference/qwen_image/gradio_demo.py

@@ -0,0 +1,214 @@
+import argparse
+from functools import lru_cache
+
+import gradio as gr
+import torch
+
+from vllm_omni.entrypoints.omni import Omni
+
+ASPECT_RATIOS: dict[str, tuple[int, int]] = {
+    "1:1": (1328, 1328),
+    "16:9": (1664, 928),
+    "9:16": (928, 1664),
+    "4:3": (1472, 1140),
+    "3:4": (1140, 1472),
+    "3:2": (1584, 1056),
+    "2:3": (1056, 1584),
+}
+ASPECT_RATIO_CHOICES = [f"{ratio} ({w}x{h})" for ratio, (w, h) in ASPECT_RATIOS.items()]
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Gradio demo for Qwen-Image offline inference.")
+    parser.add_argument("--model", default="Qwen/Qwen-Image", help="Diffusion model name or local path.")
+    parser.add_argument(
+        "--height",
+        type=int,
+        default=1328,
+        help="Default image height (must match one of the supported presets).",
+    )
+    parser.add_argument(
+        "--width",
+        type=int,
+        default=1328,
+        help="Default image width (must match one of the supported presets).",
+    )
+    parser.add_argument("--default-prompt", default="a cup of coffee on the table", help="Initial prompt shown in UI.")
+    parser.add_argument("--default-seed", type=int, default=42, help="Initial seed shown in UI.")
+    parser.add_argument("--default-cfg-scale", type=float, default=4.0, help="Initial CFG scale shown in UI.")
+    parser.add_argument(
+        "--num_inference_steps",
+        type=int,
+        default=50,
+        help="Default number of denoising steps shown in the UI.",
+    )
+    parser.add_argument("--ip", default="127.0.0.1", help="Host/IP for Gradio `launch`.")
+    parser.add_argument("--port", type=int, default=7862, help="Port for Gradio `launch`.")
+    parser.add_argument("--share", action="store_true", help="Share the Gradio demo publicly.")
+    args = parser.parse_args()
+    args.aspect_ratio_label = next(
+        (ratio for ratio, dims in ASPECT_RATIOS.items() if dims == (args.width, args.height)),
+        None,
+    )
+    if args.aspect_ratio_label is None:
+        supported = ", ".join(f"{ratio} ({w}x{h})" for ratio, (w, h) in ASPECT_RATIOS.items())
+        parser.error(f"Unsupported resolution {args.width}x{args.height}. Please pick one of: {supported}.")
+    return args
+
+
+@lru_cache(maxsize=1)
+def get_omni(model_name: str) -> Omni:
+    return Omni(model=model_name)
+
+
+def build_demo(args: argparse.Namespace) -> gr.Blocks:
+    device = "cuda"
+    omni = get_omni(args.model)
+
+    def run_inference(
+        prompt: str,
+        seed_value: float,
+        cfg_scale_value: float,
+        resolution_choice: str,
+        num_steps_value: float,
+        num_images_choice: float,
+    ):
+        if not prompt or not prompt.strip():
+            raise gr.Error("Please enter a non-empty prompt.")
+        ratio_label = resolution_choice.split(" ", 1)[0]
+        if ratio_label not in ASPECT_RATIOS:
+            raise gr.Error(f"Unsupported aspect ratio: {ratio_label}")
+        width, height = ASPECT_RATIOS[ratio_label]
+        try:
+            seed = int(seed_value)
+            num_steps = int(num_steps_value)
+            num_images = int(num_images_choice)
+        except (TypeError, ValueError) as exc:
+            raise gr.Error("Seed, inference steps, and number of images must be valid integers.") from exc
+        if num_steps <= 0:
+            raise gr.Error("Inference steps must be a positive integer.")
+        if num_images not in {1, 2, 3, 4}:
+            raise gr.Error("Number of images must be 1, 2, 3, or 4.")
+        generator = torch.Generator(device=device).manual_seed(seed)
+        images = omni.generate(
+            prompt.strip(),
+            height=height,
+            width=width,
+            generator=generator,
+            true_cfg_scale=float(cfg_scale_value),
+            num_inference_steps=num_steps,
+            num_images_per_prompt=num_images,
+            num_outputs_per_prompt=num_images,
+        )
+        return [img for img in images[:num_images]]
+
+    with gr.Blocks(
+        title="vLLM-Omni Web Serving Demo",
+        css="""
+        /* Left column button width */
+        .left-column button {
+            width: 100%;
+        }
+        /* Right preview area: fixed height, hide unnecessary buttons */
+        .fixed-image {
+            height: 660px;
+            display: flex;
+            flex-direction: column;
+            justify-content: center;
+            align-items: center;
+        }
+        .fixed-image .duplicate-button,
+        .fixed-image .svelte-drgfj2 {
+            display: none !important;
+        }
+        /* Gallery container: fill available space and center content */
+        #image-gallery {
+            width: 100%;
+            height: 100%;
+            display: flex;
+            align-items: center;
+            justify-content: center;
+        }
+        /* Gallery grid: center horizontally and vertically, set gap */
+        #image-gallery .grid {
+            display: flex;
+            flex-wrap: wrap;
+            justify-content: center;
+            align-items: center;
+            align-content: center;
+            gap: 16px;
+            width: 100%;
+            height: 100%;
+        }
+        /* Gallery grid items: center content */
+        #image-gallery .grid > div {
+            display: flex;
+            align-items: center;
+            justify-content: center;
+        }
+        /* Gallery images: limit max height, maintain aspect ratio */
+        .fixed-image img {
+            max-height: 660px !important;
+            width: auto !important;
+            object-fit: contain;
+        }
+        """,
+    ) as demo:
+        gr.Markdown("# vLLM-Omni Web Serving Demo")
+        gr.Markdown(f"**Model:** {args.model}")
+
+        with gr.Row():
+            with gr.Column(scale=1, elem_classes="left-column"):
+                prompt_input = gr.Textbox(
+                    label="Prompt",
+                    value=args.default_prompt,
+                    placeholder="Describe the image you want to generate...",
+                    lines=5,
+                )
+                seed_input = gr.Number(label="Seed", value=args.default_seed, precision=0)
+                cfg_input = gr.Number(label="CFG Scale", value=args.default_cfg_scale)
+                steps_input = gr.Number(
+                    label="Inference Steps",
+                    value=args.num_inference_steps,
+                    precision=0,
+                    minimum=1,
+                )
+                aspect_dropdown = gr.Dropdown(
+                    label="Aspect Ratio (W:H)",
+                    choices=ASPECT_RATIO_CHOICES,
+                    value=f"{args.aspect_ratio_label} ({ASPECT_RATIOS[args.aspect_ratio_label][0]}x{ASPECT_RATIOS[args.aspect_ratio_label][1]})",
+                )
+                num_images = gr.Dropdown(
+                    label="Number of images",
+                    choices=["1", "2", "3", "4"],
+                    value="1",
+                )
+                generate_btn = gr.Button("Generate", variant="primary")
+            with gr.Column(scale=2, elem_classes="fixed-image"):
+                gallery = gr.Gallery(
+                    label="Preview",
+                    columns=2,
+                    rows=2,
+                    height=660,
+                    allow_preview=True,
+                    show_label=True,
+                    elem_id="image-gallery",
+                )
+
+        generate_btn.click(
+            fn=run_inference,
+            inputs=[prompt_input, seed_input, cfg_input, aspect_dropdown, steps_input, num_images],
+            outputs=gallery,
+        )
+
+    return demo
+
+
+def main():
+    args = parse_args()
+    demo = build_demo(args)
+    demo.launch(server_name=args.ip, server_port=args.port, share=args.share)
+
+
+if __name__ == "__main__":
+    main()

examples/offline_inference/qwen_image/text_to_image.py

@@ -0,0 +1,78 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import argparse
+from pathlib import Path
+
+import torch
+
+from vllm_omni.entrypoints.omni import Omni
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Generate an image with Qwen-Image.")
+    parser.add_argument("--model", default="Qwen/Qwen-Image", help="Diffusion model name or local path.")
+    parser.add_argument("--prompt", default="a cup of coffee on the table", help="Text prompt for image generation.")
+    parser.add_argument("--seed", type=int, default=42, help="Random seed for deterministic results.")
+    parser.add_argument(
+        "--cfg_scale",
+        type=float,
+        default=4.0,
+        help="True classifier-free guidance scale specific to Qwen-Image.",
+    )
+    parser.add_argument("--height", type=int, default=1024, help="Height of generated image.")
+    parser.add_argument("--width", type=int, default=1024, help="Width of generated image.")
+    parser.add_argument(
+        "--output",
+        type=str,
+        default="qwen_image_output.png",
+        help="Path to save the generated image (PNG).",
+    )
+    parser.add_argument(
+        "--num_images_per_prompt",
+        type=int,
+        default=1,
+        help="Number of images to generate for the given prompt.",
+    )
+    parser.add_argument(
+        "--num_inference_steps",
+        type=int,
+        default=50,
+        help="Number of denoising steps for the diffusion sampler.",
+    )
+    return parser.parse_args()
+
+
+def main():
+    args = parse_args()
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    generator = torch.Generator(device=device).manual_seed(args.seed)
+
+    omni = Omni(model=args.model)
+    images = omni.generate(
+        args.prompt,
+        height=args.height,
+        width=args.width,
+        generator=generator,
+        true_cfg_scale=args.cfg_scale,
+        num_inference_steps=args.num_inference_steps,
+        num_images_per_prompt=args.num_images_per_prompt,
+        num_outputs_per_prompt=args.num_images_per_prompt,
+    )
+
+    output_path = Path(args.output)
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    suffix = output_path.suffix or ".png"
+    stem = output_path.stem or "qwen_image_output"
+    if args.num_images_per_prompt <= 1:
+        images[0].save(output_path)
+        print(f"Saved generated image to {output_path}")
+    else:
+        for idx, img in enumerate(images):
+            save_path = output_path.parent / f"{stem}_{idx}{suffix}"
+            img.save(save_path)
+            print(f"Saved generated image to {save_path}")
+
+
+if __name__ == "__main__":
+    main()

examples/online_serving/qwen2_5_omni/README.md

@@ -0,0 +1,73 @@
+# Online serving Example of vLLM-Omni for Qwen2.5-omni
+
+## 🛠️ Installation
+
+Please refer to [README.md](../../README.md)
+
+## Run examples (Qwen2.5-omni)
+
+Launch the server
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091
+```
+
+If you have custom stage configs file, launch the server with command below
+```bash
+vllm serve Qwen/Qwen2.5-Omni-7B --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
+```
+
+Get into the example folder
+```bash
+cd examples/online_serving
+```
+
+Send request via python
+```bash
+python openai_chat_completion_client_for_multimodal_generation.py --query-type mixed_modalities
+```
+
+Send request via curl
+```bash
+bash run_curl_multimodal_generation.sh mixed_modalities
+```
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```
+
+## Run Local Web UI Demo
+
+This Web UI demo allows users to interact with the model through a web browser.
+
+### Running Gradio Demo
+
+Once vllm and vllm-omni are installed, you can launch the web service built on AsyncOmniLLM by
+
+```bash
+python gradio_demo.py  --model Qwen/Qwen2.5-Omni-7B --port 7861
+```
+
+Then open `http://localhost:7861/` on your local browser to interact with the web UI.
+
+
+### Options
+
+You can customize its basic launch parameters:
+
+```bash
+python gradio_demo.py \
+    --model Qwen/Qwen2.5-Omni-7B \
+    --ip 127.0.0.1 \
+    --port 7861 \
+    --stage-configs-path /path/to/stage_configs.yaml
+```
+
+- `--model`: Local model checkpoint to load (default `Qwen/Qwen2.5-Omni-7B`).
+- `--ip`: Host/IP for the Gradio server (default `127.0.0.1`).
+- `--port`: Port for the Gradio server (default `7861`).
+- `--stage-configs-path`: Optional path to custom stage configs YAML.
+- `--share`: Set to expose a temporary public link via Gradio.

examples/online_serving/qwen2_5_omni/gradio_demo.py

@@ -0,0 +1,535 @@
+import argparse
+import os
+import random
+import signal
+import sys
+from pathlib import Path
+from types import SimpleNamespace
+from typing import Any, Optional
+
+import gradio as gr
+import numpy as np
+import soundfile as sf
+import torch
+from PIL import Image
+from vllm.assets.video import video_get_metadata, video_to_ndarrays
+from vllm.sampling_params import SamplingParams
+
+from vllm_omni.entrypoints.async_omni_llm import AsyncOmniLLM
+
+# Import utils from offline inference example
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../offline_inference/qwen2_5_omni"))
+try:
+    from utils import make_omni_prompt
+except ImportError:
+    # Fallback if utils doesn't exist - we'll build prompts directly
+    make_omni_prompt = None
+
+SEED = 42
+ASYNC_INIT_TIMEOUT = 600
+
+SUPPORTED_MODELS: dict[str, dict[str, Any]] = {
+    "Qwen/Qwen2.5-Omni-7B": {
+        "sampling_params": {
+            "thinker": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": -1,
+                "max_tokens": 2048,
+                "detokenize": True,
+                "repetition_penalty": 1.1,
+            },
+            "talker": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": -1,
+                "max_tokens": 2048,
+                "detokenize": True,
+                "repetition_penalty": 1.1,
+                "stop_token_ids": [8294],
+            },
+            "code2wav": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": -1,
+                "max_tokens": 2048,
+                "detokenize": True,
+                "repetition_penalty": 1.1,
+            },
+        },
+    },
+}
+# Ensure deterministic behavior across runs.
+random.seed(SEED)
+np.random.seed(SEED)
+torch.manual_seed(SEED)
+torch.cuda.manual_seed(SEED)
+torch.cuda.manual_seed_all(SEED)
+torch.backends.cudnn.deterministic = True
+torch.backends.cudnn.benchmark = False
+os.environ["PYTHONHASHSEED"] = str(SEED)
+os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Gradio demo for Qwen2.5-Omni online inference.")
+    parser.add_argument(
+        "--model",
+        default="Qwen/Qwen2.5-Omni-7B",
+        help="Path to local model checkpoint.",
+    )
+    parser.add_argument(
+        "--ip",
+        default="127.0.0.1",
+        help="Host/IP for gradio `launch`.",
+    )
+    parser.add_argument("--port", type=int, default=7861, help="Port for gradio `launch`.")
+    parser.add_argument("--share", action="store_true", help="Share the Gradio demo publicly.")
+    parser.add_argument(
+        "--stage-configs-path",
+        type=str,
+        default=None,
+        help="Path to custom stage configs YAML file (optional).",
+    )
+    return parser.parse_args()
+
+
+def build_sampling_params(seed: int, model_key: str) -> list[SamplingParams]:
+    """Build SamplingParams objects by reusing the dict definitions."""
+    return [SamplingParams(**params_dict) for params_dict in build_sampling_params_dict(seed, model_key)]
+
+
+def build_sampling_params_dict(seed: int, model_key: str) -> list[dict]:
+    """Build sampling params as dict for HTTP API mode."""
+    model_conf = SUPPORTED_MODELS.get(model_key)
+    if model_conf is None:
+        raise ValueError(f"Unsupported model '{model_key}'")
+
+    sampling_templates: dict[str, dict[str, Any]] = model_conf["sampling_params"]
+    sampling_params: list[dict] = []
+    for stage_name, template in sampling_templates.items():
+        params = dict(template)
+        params["seed"] = seed
+        sampling_params.append(params)
+    return sampling_params
+
+
+def create_prompt_args(base_args: argparse.Namespace) -> SimpleNamespace:
+    # The prompt builder expects a minimal namespace with these attributes.
+    return SimpleNamespace(
+        model=base_args.model,
+        prompt_type="text",
+        tokenize=True,
+        use_torchvision=True,
+        legacy_omni_video=False,
+    )
+
+
+def process_audio_file(
+    audio_file: Optional[Any],
+) -> Optional[tuple[np.ndarray, int]]:
+    """Normalize Gradio audio input to (np.ndarray, sample_rate)."""
+    if audio_file is None:
+        return None
+
+    sample_rate: Optional[int] = None
+    audio_np: Optional[np.ndarray] = None
+
+    def _load_from_path(path_str: str) -> Optional[tuple[np.ndarray, int]]:
+        if not path_str:
+            return None
+        path = Path(path_str)
+        if not path.exists():
+            return None
+        data, sr = sf.read(path)
+        if data.ndim > 1:
+            data = data[:, 0]
+        return data.astype(np.float32), int(sr)
+
+    if isinstance(audio_file, tuple):
+        if len(audio_file) == 2:
+            first, second = audio_file
+            # Case 1: (sample_rate, np.ndarray)
+            if isinstance(first, (int, float)) and isinstance(second, np.ndarray):
+                sample_rate = int(first)
+                audio_np = second
+            # Case 2: (filepath, (sample_rate, np.ndarray or list))
+            elif isinstance(first, str):
+                if isinstance(second, tuple) and len(second) == 2:
+                    sr_candidate, data_candidate = second
+                    if isinstance(sr_candidate, (int, float)) and isinstance(data_candidate, np.ndarray):
+                        sample_rate = int(sr_candidate)
+                        audio_np = data_candidate
+                if audio_np is None:
+                    loaded = _load_from_path(first)
+                    if loaded is not None:
+                        audio_np, sample_rate = loaded
+            # Case 3: (None, (sample_rate, np.ndarray))
+            elif first is None and isinstance(second, tuple) and len(second) == 2:
+                sr_candidate, data_candidate = second
+                if isinstance(sr_candidate, (int, float)) and isinstance(data_candidate, np.ndarray):
+                    sample_rate = int(sr_candidate)
+                    audio_np = data_candidate
+        elif len(audio_file) == 1 and isinstance(audio_file[0], str):
+            loaded = _load_from_path(audio_file[0])
+            if loaded is not None:
+                audio_np, sample_rate = loaded
+    elif isinstance(audio_file, str):
+        loaded = _load_from_path(audio_file)
+        if loaded is not None:
+            audio_np, sample_rate = loaded
+
+    if audio_np is None or sample_rate is None:
+        return None
+
+    if audio_np.ndim > 1:
+        audio_np = audio_np[:, 0]
+
+    return audio_np.astype(np.float32), sample_rate
+
+
+def process_image_file(image_file: Optional[Image.Image]) -> Optional[Image.Image]:
+    """Process image file from Gradio input.
+
+    Returns:
+        PIL Image in RGB mode or None if no image provided.
+    """
+    if image_file is None:
+        return None
+    # Convert to RGB if needed
+    if image_file.mode != "RGB":
+        image_file = image_file.convert("RGB")
+    return image_file
+
+
+def process_video_file(
+    video_file: Optional[str],
+    enable_audio_in_video: bool = False,
+    max_frames: int = 32,
+) -> Optional[tuple[np.ndarray, dict[str, Any], Optional[tuple[np.ndarray, int]]]]:
+    """Process video file and optionally extract audio track."""
+    if video_file is None:
+        return None
+
+    video_path = Path(video_file)
+    if not video_path.exists():
+        print(f"Video file not found: {video_path}")
+        return None
+
+    try:
+        frames = video_to_ndarrays(str(video_path), num_frames=max_frames)
+        metadata = video_get_metadata(str(video_path), num_frames=frames.shape[0])
+    except Exception as exc:  # pylint: disable=broad-except
+        print(f"Failed to decode video {video_path}: {exc}")
+        return None
+
+    audio_tuple: Optional[tuple[np.ndarray, int]] = None
+    if enable_audio_in_video:
+        try:
+            import librosa  # type: ignore import
+
+            audio_signal, sampling_rate = librosa.load(str(video_path), sr=16000)
+            audio_tuple = (audio_signal.astype(np.float32), sampling_rate)
+        except Exception as exc:  # pylint: disable=broad-except
+            print(f"Failed to extract audio from video {video_path}: {exc}")
+
+    return frames, metadata, audio_tuple
+
+
+async def run_inference_async_omni_llm(
+    omni_llm: AsyncOmniLLM,
+    sampling_params: list[SamplingParams],
+    prompt_args_template: SimpleNamespace,
+    user_prompt: str,
+    audio_file: Optional[tuple[str, tuple[int, np.ndarray]]] = None,
+    image_file: Optional[Image.Image] = None,
+    video_file: Optional[str] = None,
+    use_audio_in_video: bool = False,
+):
+    """Run inference using AsyncOmniLLM directly with multimodal support."""
+    if not user_prompt.strip() and not audio_file and not image_file and not video_file:
+        return "Please provide at least a text prompt or multimodal input.", None
+
+    try:
+        # Build prompt with multimodal data
+        prompt_args = SimpleNamespace(**prompt_args_template.__dict__)
+
+        # Process multimodal inputs
+        multi_modal_data = {}
+        mm_processor_kwargs = {}
+
+        # Process audio
+        audio_data = process_audio_file(audio_file)
+        if audio_data is not None:
+            multi_modal_data["audio"] = audio_data
+
+        # Process image
+        image_data = process_image_file(image_file)
+        if image_data is not None:
+            multi_modal_data["image"] = image_data
+
+        # Process video
+        video_payload = process_video_file(video_file, enable_audio_in_video=use_audio_in_video)
+        if video_payload is not None:
+            video_frames, video_metadata, extracted_audio = video_payload
+            video_entry: Any
+            if video_metadata:
+                video_entry = (video_frames, video_metadata)
+            else:
+                video_entry = video_frames
+            multi_modal_data["video"] = video_entry
+            if use_audio_in_video and extracted_audio is not None and "audio" not in multi_modal_data:
+                multi_modal_data["audio"] = extracted_audio
+                mm_processor_kwargs["use_audio_in_video"] = True
+
+        # Build the prompt input
+        if make_omni_prompt is not None:
+            omni_prompt = make_omni_prompt(prompt_args, user_prompt)
+            # Add multimodal data if present
+            if multi_modal_data:
+                if isinstance(omni_prompt, dict):
+                    omni_prompt["multi_modal_data"] = multi_modal_data
+                    if mm_processor_kwargs:
+                        omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+                else:
+                    # If make_omni_prompt returns a string, we need to wrap it
+                    omni_prompt = {
+                        "prompt": omni_prompt,
+                        "multi_modal_data": multi_modal_data,
+                    }
+                    if mm_processor_kwargs:
+                        omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+        else:
+            # Fallback: build prompt directly
+            default_system = (
+                "You are Qwen, a virtual human developed by the Qwen Team, "
+                "Alibaba Group, capable of perceiving auditory and visual inputs, "
+                "as well as generating text and speech."
+            )
+            prompt = f"<|im_start|>system\n{default_system}<|im_end|>\n<|im_start|>user\n"
+            if audio_data:
+                prompt += "<|audio_bos|><|AUDIO|><|audio_eos|>"
+            if image_data:
+                prompt += "<|vision_bos|><|IMAGE|><|vision_eos|>"
+            if video_payload is not None:
+                prompt += "<|vision_bos|><|VIDEO|><|vision_eos|>"
+            if user_prompt.strip():
+                prompt += f"{user_prompt}"
+            prompt += "<|im_end|>\n<|im_start|>assistant\n"
+
+            omni_prompt = {
+                "prompt": prompt,
+                "multi_modal_data": multi_modal_data,
+            }
+            if mm_processor_kwargs:
+                omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+
+        request_id = "0"
+        text_outputs: list[str] = []
+        audio_output = None
+
+        async for stage_outputs in omni_llm.generate(
+            prompt=omni_prompt,
+            request_id=request_id,
+            sampling_params_list=sampling_params,
+        ):
+            # stage_outputs.request_output is a RequestOutput object, not a list
+            request_output = stage_outputs.request_output
+            if stage_outputs.final_output_type == "text":
+                if request_output.outputs:
+                    for output in request_output.outputs:
+                        if output.text:
+                            text_outputs.append(output.text)
+            elif stage_outputs.final_output_type == "audio":
+                # multimodal_output is on the RequestOutput object
+                # See vllm_omni/entrypoints/openai/serving_chat.py:680 for reference
+                if hasattr(request_output, "multimodal_output") and request_output.multimodal_output:
+                    if "audio" in request_output.multimodal_output:
+                        audio_tensor = request_output.multimodal_output["audio"]
+                        # Ensure audio is 1D (flatten if needed)
+                        if hasattr(audio_tensor, "ndim") and audio_tensor.ndim > 1:
+                            audio_tensor = audio_tensor.flatten()
+                        audio_np = audio_tensor.detach().cpu().numpy()
+                        audio_output = (
+                            24000,  # sampling rate in Hz
+                            audio_np,
+                        )
+
+        text_response = "\n\n".join(text_outputs) if text_outputs else "No text output."
+        return text_response, audio_output
+    except Exception as exc:  # pylint: disable=broad-except
+        return f"Inference failed: {exc}", None
+
+
+def build_interface(
+    omni_llm: AsyncOmniLLM,
+    sampling_params: list[SamplingParams],
+    prompt_args_template: SimpleNamespace,
+    model: str,
+):
+    """Build Gradio interface for AsyncOmniLLM mode."""
+
+    async def run_inference(
+        user_prompt: str,
+        audio_file: Optional[tuple[str, tuple[int, np.ndarray]]],
+        image_file: Optional[Image.Image],
+        video_file: Optional[str],
+        use_audio_in_video: bool,
+    ):
+        return await run_inference_async_omni_llm(
+            omni_llm,
+            sampling_params,
+            prompt_args_template,
+            user_prompt,
+            audio_file,
+            image_file,
+            video_file,
+            use_audio_in_video,
+        )
+
+    css = """
+    .media-input-container {
+        display: flex;
+        gap: 10px;
+    }
+    .media-input-container > div {
+        flex: 1;
+    }
+    .media-input-container .image-input,
+    .media-input-container .audio-input {
+        height: 300px;
+    }
+    .media-input-container .video-column {
+        height: 300px;
+        display: flex;
+        flex-direction: column;
+    }
+    .media-input-container .video-input {
+        flex: 1;
+        min-height: 0;
+    }
+    #generate-btn button {
+        width: 100%;
+    }
+    """
+
+    with gr.Blocks(css=css) as demo:
+        gr.Markdown("# vLLM-Omni Online Serving Demo")
+        gr.Markdown(f"**Model:** {model} \n\n")
+
+        with gr.Column():
+            with gr.Row():
+                input_box = gr.Textbox(
+                    label="Text Prompt",
+                    placeholder="For example: Describe what happens in the media inputs.",
+                    lines=4,
+                    scale=1,
+                )
+            with gr.Row(elem_classes="media-input-container"):
+                image_input = gr.Image(
+                    label="Image Input (optional)",
+                    type="pil",
+                    sources=["upload"],
+                    scale=1,
+                    elem_classes="image-input",
+                )
+                with gr.Column(scale=1, elem_classes="video-column"):
+                    video_input = gr.Video(
+                        label="Video Input (optional)",
+                        sources=["upload"],
+                        elem_classes="video-input",
+                    )
+                    use_audio_in_video_checkbox = gr.Checkbox(
+                        label="Use audio from video",
+                        value=False,
+                        info="Extract the video's audio track when provided.",
+                    )
+                audio_input = gr.Audio(
+                    label="Audio Input (optional)",
+                    type="numpy",
+                    sources=["upload", "microphone"],
+                    scale=1,
+                    elem_classes="audio-input",
+                )
+
+        with gr.Row():
+            generate_btn = gr.Button(
+                "Generate",
+                variant="primary",
+                size="lg",
+                elem_id="generate-btn",
+            )
+
+        with gr.Row():
+            text_output = gr.Textbox(label="Text Output", lines=10, scale=2)
+            audio_output = gr.Audio(label="Audio Output", interactive=False, scale=1)
+
+        generate_btn.click(
+            fn=run_inference,
+            inputs=[input_box, audio_input, image_input, video_input, use_audio_in_video_checkbox],
+            outputs=[text_output, audio_output],
+        )
+        demo.queue()
+    return demo
+
+
+def main():
+    args = parse_args()
+    omni_llm = None
+
+    model_name = "/".join(args.model.split("/")[-2:])
+    assert model_name in SUPPORTED_MODELS, (
+        f"Unsupported model '{model_name}'. Supported models: {SUPPORTED_MODELS.keys()}"
+    )
+
+    # Register signal handlers for graceful shutdown
+    def signal_handler(sig, frame):
+        print("\nReceived interrupt signal, shutting down...")
+        if omni_llm is not None:
+            try:
+                omni_llm.shutdown()
+            except Exception as e:
+                print(f"Error during shutdown: {e}")
+        sys.exit(0)
+
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    print(f"Initializing AsyncOmniLLM with model: {args.model}")
+    if args.stage_configs_path:
+        print(f"Using custom stage configs: {args.stage_configs_path}")
+
+    sampling_params = build_sampling_params(SEED, model_name)
+    omni_llm = AsyncOmniLLM(
+        model=args.model,
+        stage_configs_path=args.stage_configs_path,
+        init_timeout=ASYNC_INIT_TIMEOUT,
+    )
+    print("✓ AsyncOmniLLM initialized successfully")
+    prompt_args_template = create_prompt_args(args)
+
+    demo = build_interface(
+        omni_llm,
+        sampling_params,
+        prompt_args_template,
+        args.model,
+    )
+    try:
+        demo.launch(
+            server_name=args.ip,
+            server_port=args.port,
+            share=args.share,
+        )
+    except KeyboardInterrupt:
+        print("\nShutting down...")
+    finally:
+        # Cleanup
+        if omni_llm is not None:
+            try:
+                omni_llm.shutdown()
+            except Exception as e:
+                print(f"Error during cleanup: {e}")
+
+
+if __name__ == "__main__":
+    main()

examples/online_serving/qwen2_5_omni/openai_chat_completion_client_for_multimodal_generation.py

@@ -0,0 +1,228 @@
+import base64
+
+import requests
+from openai import OpenAI
+from vllm.assets.audio import AudioAsset
+from vllm.utils import FlexibleArgumentParser
+
+# Modify OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://localhost:8091/v1"
+
+client = OpenAI(
+    # defaults to os.environ.get("OPENAI_API_KEY")
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+SEED = 42
+
+
+def encode_base64_content_from_url(content_url: str) -> str:
+    """Encode a content retrieved from a remote url to base64 format."""
+
+    with requests.get(content_url) as response:
+        response.raise_for_status()
+        result = base64.b64encode(response.content).decode("utf-8")
+
+    return result
+
+
+def get_system_prompt():
+    return {
+        "role": "system",
+        "content": [
+            {
+                "type": "text",
+                "text": (
+                    "You are Qwen, a virtual human developed by the Qwen Team, "
+                    "Alibaba Group, capable of perceiving auditory and visual inputs, "
+                    "as well as generating text and speech."
+                ),
+            }
+        ],
+    }
+
+
+def get_text_query():
+    question = "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "text",
+                "text": f"{question}",
+            }
+        ],
+    }
+    return prompt
+
+
+def get_mixed_modalities_query():
+    question = "What is recited in the audio? What is the content of this image? Why is this video funny?"
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "audio_url",
+                "audio_url": {"url": AudioAsset("mary_had_lamb").url},
+            },
+            {
+                "type": "image_url",
+                "image_url": {
+                    "url": "https://vllm-public-assets.s3.us-west-2.amazonaws.com/vision_model_images/cherry_blossom.jpg"
+                },
+            },
+            {
+                "type": "video_url",
+                "video_url": {
+                    "url": "https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/sample_demo_1.mp4"
+                },
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+
+    return prompt
+
+
+def get_use_audio_in_video_query():
+    question = "Describe the content of the video, then convert what the baby say into text."
+
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "video_url",
+                "video_url": {
+                    "url": "https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/sample_demo_1.mp4",
+                    "num_frames": 16,
+                },
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+
+    return prompt
+
+
+def get_multi_audios_query():
+    question = "Are these two audio clips the same?"
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "audio_url",
+                "audio_url": {"url": AudioAsset("mary_had_lamb").url},
+            },
+            {
+                "type": "audio_url",
+                "audio_url": {"url": AudioAsset("winning_call").url},
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+    return prompt
+
+
+query_map = {
+    "mixed_modalities": get_mixed_modalities_query,
+    "use_audio_in_video": get_use_audio_in_video_query,
+    "multi_audios": get_multi_audios_query,
+    "text": get_text_query,
+}
+
+
+def run_multimodal_generation(args) -> None:
+    model_name = "Qwen/Qwen2.5-Omni-7B"
+    thinker_sampling_params = {
+        "temperature": 0.0,  # Deterministic - no randomness
+        "top_p": 1.0,  # Disable nucleus sampling
+        "top_k": -1,  # Disable top-k sampling
+        "max_tokens": 2048,
+        "seed": SEED,  # Fixed seed for sampling
+        "detokenize": True,
+        "repetition_penalty": 1.1,
+    }
+    talker_sampling_params = {
+        "temperature": 0.9,
+        "top_p": 0.8,
+        "top_k": 40,
+        "max_tokens": 2048,
+        "seed": SEED,  # Fixed seed for sampling
+        "detokenize": True,
+        "repetition_penalty": 1.05,
+        "stop_token_ids": [8294],
+    }
+    code2wav_sampling_params = {
+        "temperature": 0.0,  # Deterministic - no randomness
+        "top_p": 1.0,  # Disable nucleus sampling
+        "top_k": -1,  # Disable top-k sampling
+        "max_tokens": 2048,
+        "seed": SEED,  # Fixed seed for sampling
+        "detokenize": True,
+        "repetition_penalty": 1.1,
+    }
+
+    sampling_params_list = [
+        thinker_sampling_params,
+        talker_sampling_params,
+        code2wav_sampling_params,
+    ]
+
+    prompt = query_map[args.query_type]()
+    extra_body = {
+        "sampling_params_list": sampling_params_list  # Optional, it has a default setting in stage_configs of the corresponding model.
+    }
+
+    if args.query_type == "use_audio_in_video":
+        extra_body["mm_processor_kwargs"] = {"use_audio_in_video": True}
+
+    chat_completion = client.chat.completions.create(
+        messages=[
+            get_system_prompt(),
+            prompt,
+        ],
+        model=model_name,
+        extra_body=extra_body,
+    )
+
+    count = 0
+    for choice in chat_completion.choices:
+        if choice.message.audio:
+            audio_data = base64.b64decode(choice.message.audio.data)
+            audio_file_path = f"audio_{count}.wav"
+            with open(audio_file_path, "wb") as f:
+                f.write(audio_data)
+            print(f"Audio saved to {audio_file_path}")
+            count += 1
+        elif choice.message.content:
+            print("Chat completion output from text:", choice.message.content)
+
+
+def parse_args():
+    parser = FlexibleArgumentParser(description="Demo on using vLLM for offline inference with audio language models")
+    parser.add_argument(
+        "--query-type",
+        "-q",
+        type=str,
+        default="mixed_modalities",
+        choices=query_map.keys(),
+        help="Query type.",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    run_multimodal_generation(args)

examples/online_serving/qwen2_5_omni/run_curl_multimodal_generation.sh

@@ -0,0 +1,187 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Default query type
+QUERY_TYPE="${1:-mixed_modalities}"
+
+# Validate query type
+if [[ ! "$QUERY_TYPE" =~ ^(mixed_modalities|use_audio_in_video|multi_audios|text)$ ]]; then
+    echo "Error: Invalid query type '$QUERY_TYPE'"
+    echo "Usage: $0 [mixed_modalities|use_audio_in_video|multi_audios|text]"
+    echo "  mixed_modalities: Audio + Image + Video + Text query"
+    echo "  use_audio_in_video: Video + Text query (with audio extraction from video)"
+    echo "  multi_audios: Two audio clips + Text query"
+    echo "  text: Text query"
+    exit 1
+fi
+
+SEED=42
+
+thinker_sampling_params='{
+  "temperature": 0.0,
+  "top_p": 1.0,
+  "top_k": -1,
+  "max_tokens": 2048,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.1
+}'
+
+talker_sampling_params='{
+  "temperature": 0.9,
+  "top_p": 0.8,
+  "top_k": 40,
+  "max_tokens": 2048,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.05,
+  "stop_token_ids": [8294]
+}'
+
+code2wav_sampling_params='{
+  "temperature": 0.0,
+  "top_p": 1.0,
+  "top_k": -1,
+  "max_tokens": 2048,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.1
+}'
+# Above is optional, it has a default setting in stage_configs of the corresponding model.
+
+# Define URLs for assets
+MARY_HAD_LAMB_AUDIO_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/mary_had_lamb.ogg"
+WINNING_CALL_AUDIO_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/winning_call.ogg"
+CHERRY_BLOSSOM_IMAGE_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/vision_model_images/cherry_blossom.jpg"
+SAMPLE_VIDEO_URL="https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/sample_demo_1.mp4"
+
+# Build user content and extra fields based on query type
+case "$QUERY_TYPE" in
+  text)
+    user_content='[
+      {
+        "type": "text",
+        "text": "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+      }
+    ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+  mixed_modalities)
+    user_content='[
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$MARY_HAD_LAMB_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "image_url",
+          "image_url": {
+            "url": "'"$CHERRY_BLOSSOM_IMAGE_URL"'"
+          }
+        },
+        {
+          "type": "video_url",
+          "video_url": {
+            "url": "'"$SAMPLE_VIDEO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "What is recited in the audio? What is the content of this image? Why is this video funny?"
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+  use_audio_in_video)
+    user_content='[
+        {
+          "type": "video_url",
+          "video_url": {
+            "url": "'"$SAMPLE_VIDEO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Describe the content of the video, then convert what the baby say into text."
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs='{
+      "use_audio_in_video": true
+    }'
+    ;;
+  multi_audios)
+    user_content='[
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$MARY_HAD_LAMB_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$WINNING_CALL_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Are these two audio clips the same?"
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+esac
+
+echo "Running query type: $QUERY_TYPE"
+echo ""
+
+
+output=$(curl -sS -X POST http://localhost:8091/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d @- <<EOF
+{
+  "model": "Qwen/Qwen2.5-Omni-7B",
+  "sampling_params_list": $sampling_params_list,
+  "mm_processor_kwargs": $mm_processor_kwargs,
+  "messages": [
+    {
+      "role": "system",
+      "content": [
+        {
+          "type": "text",
+          "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."
+        }
+      ]
+    },
+    {
+      "role": "user",
+      "content": $user_content
+    }
+  ]
+}
+EOF
+  )
+
+# Here it only shows the text content of the first choice. Audio content has many binaries, so it's not displayed here.
+echo "Output of request: $(echo "$output" | jq '.choices[0].message.content')"

examples/online_serving/qwen3_omni/README.md

@@ -0,0 +1,77 @@
+# Online serving Example of vLLM-omni for Qwen2.5-omni
+
+## 🛠️ Installation
+
+Please refer to [README.md](../../README.md)
+
+## Run examples (Qwen3-Omni)
+
+Launch the server
+```bash
+vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091
+```
+
+If you have custom stage configs file, launch the server with command below
+```bash
+vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
+```
+
+Get into the example folder
+```bash
+cd examples/online_serving
+```
+
+Send request via python
+```bash
+python openai_chat_completion_client_for_multimodal_generation.py --query-type mixed_modalities
+```
+
+Send request via curl
+```bash
+bash run_curl_multimodal_generation.sh mixed_modalities
+```
+
+### FAQ
+
+If you encounter error about backend of librosa, try to install ffmpeg with command below.
+```
+sudo apt update
+sudo apt install ffmpeg
+```
+
+## Run Local Web UI Demo
+
+This Web UI demo allows users to interact with the model through a web browser.
+
+### Running Gradio Demo
+
+Once vllm and vllm-omni are installed, you can launch the web service built on AsyncOmniLLM by
+
+```bash
+python gradio_demo.py  --model Qwen/Qwen3-Omni-30B-A3B-Instruct --port 7861
+```
+
+Then open `http://localhost:7861/` on your local browser to interact with the web UI.
+
+
+### Options
+
+The gradio demo also supports running with an existing API server and can be customized with the following arguments.
+
+
+```bash
+python gradio_demo.py \
+    --model Qwen/Qwen3-Omni-30B-A3B-Instruct \
+    --use-api-server \
+    --api-base http://localhost:8091/v1 \
+    --ip 127.0.0.1 \
+    --port 7861
+```
+
+- `--model`: Model name
+- `--use-api-server`: If set, connect to an existing vLLM HTTP API server instead of running AsyncOmniLLM locally.
+- `--api-base`: Base URL for vllm serve (only used when `use-api-server` is set, default: http://localhost:8091/v1)
+- `--ip`: Host/IP for Gradio server (default: 127.0.0.1)
+- `--port`: Port for Gradio server (default: 7861)
+- `--stage-configs-path`: Path to custom stage configs YAML file (optional)
+- `--share`: Share the Gradio demo publicly (creates a public link)

examples/online_serving/qwen3_omni/gradio_demo.py

@@ -0,0 +1,537 @@
+import argparse
+import os
+import random
+import signal
+import sys
+from pathlib import Path
+from types import SimpleNamespace
+from typing import Any, Optional
+
+import gradio as gr
+import numpy as np
+import soundfile as sf
+import torch
+from PIL import Image
+from vllm.assets.video import video_get_metadata, video_to_ndarrays
+from vllm.sampling_params import SamplingParams
+
+from vllm_omni.entrypoints.async_omni_llm import AsyncOmniLLM
+
+# Import utils from offline inference example
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../offline_inference/qwen3_omni"))
+try:
+    from utils import make_omni_prompt
+except ImportError:
+    # Fallback if utils doesn't exist - we'll build prompts directly
+    make_omni_prompt = None
+
+SEED = 42
+ASYNC_INIT_TIMEOUT = 600
+
+SUPPORTED_MODELS: dict[str, dict[str, Any]] = {
+    "Qwen/Qwen3-Omni-30B-A3B-Instruct": {
+        "sampling_params": {
+            "thinker": {
+                "temperature": 0.4,
+                "top_p": 0.9,
+                "top_k": 1,
+                "max_tokens": 16384,
+                "detokenize": True,
+                "repetition_penalty": 1.05,
+                "stop_token_ids": [151645],
+                "seed": SEED,
+            },
+            "talker": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": -1,
+                "max_tokens": 4096,
+                "detokenize": True,
+                "repetition_penalty": 1.1,
+                "stop_token_ids": [2150],
+            },
+            "code2wav": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": -1,
+                "max_tokens": 2048,
+                "detokenize": True,
+                "repetition_penalty": 1.1,
+            },
+        },
+    },
+}
+# Ensure deterministic behavior across runs.
+random.seed(SEED)
+np.random.seed(SEED)
+torch.manual_seed(SEED)
+torch.cuda.manual_seed(SEED)
+torch.cuda.manual_seed_all(SEED)
+torch.backends.cudnn.deterministic = True
+torch.backends.cudnn.benchmark = False
+os.environ["PYTHONHASHSEED"] = str(SEED)
+os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Gradio demo for Qwen3-Omni online inference.")
+    parser.add_argument(
+        "--model",
+        default="Qwen/Qwen3-Omni-30B-A3B-Instruct",
+        help="Path to local model checkpoint.",
+    )
+    parser.add_argument(
+        "--ip",
+        default="127.0.0.1",
+        help="Host/IP for gradio `launch`.",
+    )
+    parser.add_argument("--port", type=int, default=7861, help="Port for gradio `launch`.")
+    parser.add_argument("--share", action="store_true", help="Share the Gradio demo publicly.")
+    parser.add_argument(
+        "--stage-configs-path",
+        type=str,
+        default=None,
+        help="Path to custom stage configs YAML file (optional).",
+    )
+    return parser.parse_args()
+
+
+def build_sampling_params(seed: int, model_key: str) -> list[SamplingParams]:
+    """Build SamplingParams objects by reusing the dict definitions."""
+    return [SamplingParams(**params_dict) for params_dict in build_sampling_params_dict(seed, model_key)]
+
+
+def build_sampling_params_dict(seed: int, model_key: str) -> list[dict]:
+    """Build sampling params as dict for HTTP API mode."""
+    model_conf = SUPPORTED_MODELS.get(model_key)
+    if model_conf is None:
+        raise ValueError(f"Unsupported model '{model_key}'")
+
+    sampling_templates: dict[str, dict[str, Any]] = model_conf["sampling_params"]
+    sampling_params: list[dict] = []
+    for stage_name, template in sampling_templates.items():
+        params = dict(template)
+        params["seed"] = seed
+        sampling_params.append(params)
+    return sampling_params
+
+
+def create_prompt_args(base_args: argparse.Namespace) -> SimpleNamespace:
+    # The prompt builder expects a minimal namespace with these attributes.
+    return SimpleNamespace(
+        model=base_args.model,
+        prompt_type="text",
+        tokenize=True,
+        use_torchvision=True,
+        legacy_omni_video=False,
+    )
+
+
+def process_audio_file(
+    audio_file: Optional[Any],
+) -> Optional[tuple[np.ndarray, int]]:
+    """Normalize Gradio audio input to (np.ndarray, sample_rate)."""
+    if audio_file is None:
+        return None
+
+    sample_rate: Optional[int] = None
+    audio_np: Optional[np.ndarray] = None
+
+    def _load_from_path(path_str: str) -> Optional[tuple[np.ndarray, int]]:
+        if not path_str:
+            return None
+        path = Path(path_str)
+        if not path.exists():
+            return None
+        data, sr = sf.read(path)
+        if data.ndim > 1:
+            data = data[:, 0]
+        return data.astype(np.float32), int(sr)
+
+    if isinstance(audio_file, tuple):
+        if len(audio_file) == 2:
+            first, second = audio_file
+            # Case 1: (sample_rate, np.ndarray)
+            if isinstance(first, (int, float)) and isinstance(second, np.ndarray):
+                sample_rate = int(first)
+                audio_np = second
+            # Case 2: (filepath, (sample_rate, np.ndarray or list))
+            elif isinstance(first, str):
+                if isinstance(second, tuple) and len(second) == 2:
+                    sr_candidate, data_candidate = second
+                    if isinstance(sr_candidate, (int, float)) and isinstance(data_candidate, np.ndarray):
+                        sample_rate = int(sr_candidate)
+                        audio_np = data_candidate
+                if audio_np is None:
+                    loaded = _load_from_path(first)
+                    if loaded is not None:
+                        audio_np, sample_rate = loaded
+            # Case 3: (None, (sample_rate, np.ndarray))
+            elif first is None and isinstance(second, tuple) and len(second) == 2:
+                sr_candidate, data_candidate = second
+                if isinstance(sr_candidate, (int, float)) and isinstance(data_candidate, np.ndarray):
+                    sample_rate = int(sr_candidate)
+                    audio_np = data_candidate
+        elif len(audio_file) == 1 and isinstance(audio_file[0], str):
+            loaded = _load_from_path(audio_file[0])
+            if loaded is not None:
+                audio_np, sample_rate = loaded
+    elif isinstance(audio_file, str):
+        loaded = _load_from_path(audio_file)
+        if loaded is not None:
+            audio_np, sample_rate = loaded
+
+    if audio_np is None or sample_rate is None:
+        return None
+
+    if audio_np.ndim > 1:
+        audio_np = audio_np[:, 0]
+
+    return audio_np.astype(np.float32), sample_rate
+
+
+def process_image_file(image_file: Optional[Image.Image]) -> Optional[Image.Image]:
+    """Process image file from Gradio input.
+
+    Returns:
+        PIL Image in RGB mode or None if no image provided.
+    """
+    if image_file is None:
+        return None
+    # Convert to RGB if needed
+    if image_file.mode != "RGB":
+        image_file = image_file.convert("RGB")
+    return image_file
+
+
+def process_video_file(
+    video_file: Optional[str],
+    enable_audio_in_video: bool = False,
+    max_frames: int = 32,
+) -> Optional[tuple[np.ndarray, dict[str, Any], Optional[tuple[np.ndarray, int]]]]:
+    """Process video file and optionally extract audio track."""
+    if video_file is None:
+        return None
+
+    video_path = Path(video_file)
+    if not video_path.exists():
+        print(f"Video file not found: {video_path}")
+        return None
+
+    try:
+        frames = video_to_ndarrays(str(video_path), num_frames=max_frames)
+        metadata = video_get_metadata(str(video_path), num_frames=frames.shape[0])
+    except Exception as exc:  # pylint: disable=broad-except
+        print(f"Failed to decode video {video_path}: {exc}")
+        return None
+
+    audio_tuple: Optional[tuple[np.ndarray, int]] = None
+    if enable_audio_in_video:
+        try:
+            import librosa  # type: ignore import
+
+            audio_signal, sampling_rate = librosa.load(str(video_path), sr=16000)
+            audio_tuple = (audio_signal.astype(np.float32), sampling_rate)
+        except Exception as exc:  # pylint: disable=broad-except
+            print(f"Failed to extract audio from video {video_path}: {exc}")
+
+    return frames, metadata, audio_tuple
+
+
+async def run_inference_async_omni_llm(
+    omni_llm: AsyncOmniLLM,
+    sampling_params: list[SamplingParams],
+    prompt_args_template: SimpleNamespace,
+    user_prompt: str,
+    audio_file: Optional[tuple[str, tuple[int, np.ndarray]]] = None,
+    image_file: Optional[Image.Image] = None,
+    video_file: Optional[str] = None,
+    use_audio_in_video: bool = False,
+):
+    """Run inference using AsyncOmniLLM directly with multimodal support."""
+    if not user_prompt.strip() and not audio_file and not image_file and not video_file:
+        return "Please provide at least a text prompt or multimodal input.", None
+
+    try:
+        # Build prompt with multimodal data
+        prompt_args = SimpleNamespace(**prompt_args_template.__dict__)
+
+        # Process multimodal inputs
+        multi_modal_data = {}
+        mm_processor_kwargs = {}
+
+        # Process audio
+        audio_data = process_audio_file(audio_file)
+        if audio_data is not None:
+            multi_modal_data["audio"] = audio_data
+
+        # Process image
+        image_data = process_image_file(image_file)
+        if image_data is not None:
+            multi_modal_data["image"] = image_data
+
+        # Process video
+        video_payload = process_video_file(video_file, enable_audio_in_video=use_audio_in_video)
+        if video_payload is not None:
+            video_frames, video_metadata, extracted_audio = video_payload
+            video_entry: Any
+            if video_metadata:
+                video_entry = (video_frames, video_metadata)
+            else:
+                video_entry = video_frames
+            multi_modal_data["video"] = video_entry
+            if use_audio_in_video and extracted_audio is not None and "audio" not in multi_modal_data:
+                multi_modal_data["audio"] = extracted_audio
+                mm_processor_kwargs["use_audio_in_video"] = True
+
+        # Build the prompt input
+        if make_omni_prompt is not None:
+            omni_prompt = make_omni_prompt(prompt_args, user_prompt)
+            # Add multimodal data if present
+            if multi_modal_data:
+                if isinstance(omni_prompt, dict):
+                    omni_prompt["multi_modal_data"] = multi_modal_data
+                    if mm_processor_kwargs:
+                        omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+                else:
+                    # If make_omni_prompt returns a string, we need to wrap it
+                    omni_prompt = {
+                        "prompt": omni_prompt,
+                        "multi_modal_data": multi_modal_data,
+                    }
+                    if mm_processor_kwargs:
+                        omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+        else:
+            # Fallback: build prompt directly
+            default_system = (
+                "You are Qwen, a virtual human developed by the Qwen Team, "
+                "Alibaba Group, capable of perceiving auditory and visual inputs, "
+                "as well as generating text and speech."
+            )
+            prompt = f"<|im_start|>system\n{default_system}<|im_end|>\n<|im_start|>user\n"
+            if audio_data:
+                prompt += "<|audio_bos|><|AUDIO|><|audio_eos|>"
+            if image_data:
+                prompt += "<|vision_bos|><|IMAGE|><|vision_eos|>"
+            if video_payload is not None:
+                prompt += "<|vision_bos|><|VIDEO|><|vision_eos|>"
+            if user_prompt.strip():
+                prompt += f"{user_prompt}"
+            prompt += "<|im_end|>\n<|im_start|>assistant\n"
+
+            omni_prompt = {
+                "prompt": prompt,
+                "multi_modal_data": multi_modal_data,
+            }
+            if mm_processor_kwargs:
+                omni_prompt["mm_processor_kwargs"] = mm_processor_kwargs
+
+        request_id = "0"
+        text_outputs: list[str] = []
+        audio_output = None
+
+        async for stage_outputs in omni_llm.generate(
+            prompt=omni_prompt,
+            request_id=request_id,
+            sampling_params_list=sampling_params,
+        ):
+            # stage_outputs.request_output is a RequestOutput object, not a list
+            request_output = stage_outputs.request_output
+            if stage_outputs.final_output_type == "text":
+                if request_output.outputs:
+                    for output in request_output.outputs:
+                        if output.text:
+                            text_outputs.append(output.text)
+            elif stage_outputs.final_output_type == "audio":
+                # multimodal_output is on the RequestOutput object
+                # See vllm_omni/entrypoints/openai/serving_chat.py:680 for reference
+                if hasattr(request_output, "multimodal_output") and request_output.multimodal_output:
+                    if "audio" in request_output.multimodal_output:
+                        audio_tensor = request_output.multimodal_output["audio"]
+                        # Ensure audio is 1D (flatten if needed)
+                        if hasattr(audio_tensor, "ndim") and audio_tensor.ndim > 1:
+                            audio_tensor = audio_tensor.flatten()
+                        audio_np = audio_tensor.detach().cpu().numpy()
+                        audio_output = (
+                            24000,  # sampling rate in Hz
+                            audio_np,
+                        )
+
+        text_response = "\n\n".join(text_outputs) if text_outputs else "No text output."
+        return text_response, audio_output
+    except Exception as exc:  # pylint: disable=broad-except
+        return f"Inference failed: {exc}", None
+
+
+def build_interface(
+    omni_llm: AsyncOmniLLM,
+    sampling_params: list[SamplingParams],
+    prompt_args_template: SimpleNamespace,
+    model: str,
+):
+    """Build Gradio interface for AsyncOmniLLM mode."""
+
+    async def run_inference(
+        user_prompt: str,
+        audio_file: Optional[tuple[str, tuple[int, np.ndarray]]],
+        image_file: Optional[Image.Image],
+        video_file: Optional[str],
+        use_audio_in_video: bool,
+    ):
+        return await run_inference_async_omni_llm(
+            omni_llm,
+            sampling_params,
+            prompt_args_template,
+            user_prompt,
+            audio_file,
+            image_file,
+            video_file,
+            use_audio_in_video,
+        )
+
+    css = """
+    .media-input-container {
+        display: flex;
+        gap: 10px;
+    }
+    .media-input-container > div {
+        flex: 1;
+    }
+    .media-input-container .image-input,
+    .media-input-container .audio-input {
+        height: 300px;
+    }
+    .media-input-container .video-column {
+        height: 300px;
+        display: flex;
+        flex-direction: column;
+    }
+    .media-input-container .video-input {
+        flex: 1;
+        min-height: 0;
+    }
+    #generate-btn button {
+        width: 100%;
+    }
+    """
+
+    with gr.Blocks(css=css) as demo:
+        gr.Markdown("# vLLM-Omni Online Serving Demo")
+        gr.Markdown(f"**Model:** {model} \n\n")
+
+        with gr.Column():
+            with gr.Row():
+                input_box = gr.Textbox(
+                    label="Text Prompt",
+                    placeholder="For example: Describe what happens in the media inputs.",
+                    lines=4,
+                    scale=1,
+                )
+            with gr.Row(elem_classes="media-input-container"):
+                image_input = gr.Image(
+                    label="Image Input (optional)",
+                    type="pil",
+                    sources=["upload"],
+                    scale=1,
+                    elem_classes="image-input",
+                )
+                with gr.Column(scale=1, elem_classes="video-column"):
+                    video_input = gr.Video(
+                        label="Video Input (optional)",
+                        sources=["upload"],
+                        elem_classes="video-input",
+                    )
+                    use_audio_in_video_checkbox = gr.Checkbox(
+                        label="Use audio from video",
+                        value=False,
+                        info="Extract the video's audio track when provided.",
+                    )
+                audio_input = gr.Audio(
+                    label="Audio Input (optional)",
+                    type="numpy",
+                    sources=["upload", "microphone"],
+                    scale=1,
+                    elem_classes="audio-input",
+                )
+
+        with gr.Row():
+            generate_btn = gr.Button(
+                "Generate",
+                variant="primary",
+                size="lg",
+                elem_id="generate-btn",
+            )
+
+        with gr.Row():
+            text_output = gr.Textbox(label="Text Output", lines=10, scale=2)
+            audio_output = gr.Audio(label="Audio Output", interactive=False, scale=1)
+
+        generate_btn.click(
+            fn=run_inference,
+            inputs=[input_box, audio_input, image_input, video_input, use_audio_in_video_checkbox],
+            outputs=[text_output, audio_output],
+        )
+        demo.queue()
+    return demo
+
+
+def main():
+    args = parse_args()
+    omni_llm = None
+
+    model_name = "/".join(args.model.split("/")[-2:])
+    assert model_name in SUPPORTED_MODELS, (
+        f"Unsupported model '{model_name}'. Supported models: {SUPPORTED_MODELS.keys()}"
+    )
+
+    # Register signal handlers for graceful shutdown
+    def signal_handler(sig, frame):
+        print("\nReceived interrupt signal, shutting down...")
+        if omni_llm is not None:
+            try:
+                omni_llm.shutdown()
+            except Exception as e:
+                print(f"Error during shutdown: {e}")
+        sys.exit(0)
+
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    print(f"Initializing AsyncOmniLLM with model: {args.model}")
+    if args.stage_configs_path:
+        print(f"Using custom stage configs: {args.stage_configs_path}")
+
+    sampling_params = build_sampling_params(SEED, model_name)
+    omni_llm = AsyncOmniLLM(
+        model=args.model,
+        stage_configs_path=args.stage_configs_path,
+        init_timeout=ASYNC_INIT_TIMEOUT,
+    )
+    print("✓ AsyncOmniLLM initialized successfully")
+    prompt_args_template = create_prompt_args(args)
+
+    demo = build_interface(
+        omni_llm,
+        sampling_params,
+        prompt_args_template,
+        args.model,
+    )
+    try:
+        demo.launch(
+            server_name=args.ip,
+            server_port=args.port,
+            share=args.share,
+        )
+    except KeyboardInterrupt:
+        print("\nShutting down...")
+    finally:
+        # Cleanup
+        if omni_llm is not None:
+            try:
+                omni_llm.shutdown()
+            except Exception as e:
+                print(f"Error during cleanup: {e}")
+
+
+if __name__ == "__main__":
+    main()

examples/online_serving/qwen3_omni/openai_chat_completion_client_for_multimodal_generation.py

@@ -0,0 +1,235 @@
+import base64
+from typing import NamedTuple
+
+import requests
+from openai import OpenAI
+from vllm.assets.audio import AudioAsset
+from vllm.utils import FlexibleArgumentParser
+
+# Modify OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://localhost:8091/v1"
+
+client = OpenAI(
+    # defaults to os.environ.get("OPENAI_API_KEY")
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+SEED = 42
+
+
+class QueryResult(NamedTuple):
+    inputs: dict
+    limit_mm_per_prompt: dict[str, int]
+
+
+def encode_base64_content_from_url(content_url: str) -> str:
+    """Encode a content retrieved from a remote url to base64 format."""
+
+    with requests.get(content_url) as response:
+        response.raise_for_status()
+        result = base64.b64encode(response.content).decode("utf-8")
+
+    return result
+
+
+def get_system_prompt():
+    return {
+        "role": "system",
+        "content": [
+            {
+                "type": "text",
+                "text": (
+                    "You are Qwen, a virtual human developed by the Qwen Team, "
+                    "Alibaba Group, capable of perceiving auditory and visual inputs, "
+                    "as well as generating text and speech."
+                ),
+            }
+        ],
+    }
+
+
+def get_text_query():
+    question = "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "text",
+                "text": f"{question}",
+            }
+        ],
+    }
+    return prompt
+
+
+default_system = (
+    "You are Qwen, a virtual human developed by the Qwen Team, Alibaba "
+    "Group, capable of perceiving auditory and visual inputs, as well as "
+    "generating text and speech."
+)
+
+
+def get_video_query() -> QueryResult:
+    question = "Why is this video funny?"
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "video_url",
+                "video_url": {
+                    "url": "https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/sample_demo_1.mp4"
+                },
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+    return prompt
+
+
+def get_image_query() -> QueryResult:
+    question = "What is the content of this image?"
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "image_url",
+                "image_url": {
+                    "url": "https://vllm-public-assets.s3.us-west-2.amazonaws.com/vision_model_images/cherry_blossom.jpg"
+                },
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+    return prompt
+
+
+def get_audio_query() -> QueryResult:
+    question = "What is the content of this audio?"
+    prompt = {
+        "role": "user",
+        "content": [
+            {
+                "type": "audio_url",
+                "audio_url": {"url": AudioAsset("mary_had_lamb").url},
+            },
+            {
+                "type": "text",
+                "text": f"{question}",
+            },
+        ],
+    }
+    return prompt
+
+
+query_map = {
+    "text": get_text_query,
+    "use_audio": get_audio_query,
+    "use_image": get_image_query,
+    "use_video": get_video_query,
+}
+
+
+def run_multimodal_generation(args) -> None:
+    model_name = args.model
+    thinker_sampling_params = {
+        "temperature": 0.4,  # Deterministic
+        "top_p": 0.9,
+        "top_k": 1,
+        "max_tokens": 16384,
+        "repetition_penalty": 1.05,
+        "stop_token_ids": [151645],  # Qwen EOS token <|im_end|>
+        "seed": SEED,
+    }
+
+    # Sampling parameters for Talker stage (codec generation)
+    # Stop at codec EOS token
+    talker_sampling_params = {
+        "temperature": 0.9,
+        "top_k": 50,
+        "max_tokens": 4096,
+        "seed": SEED,
+        "detokenize": False,
+        "repetition_penalty": 1.05,
+        "stop_token_ids": [2150],  # TALKER_CODEC_EOS_TOKEN_ID
+    }
+
+    # # Sampling parameters for Code2Wav stage (audio generation)
+    code2wav_sampling_params = {
+        "temperature": 0.0,
+        "top_p": 1.0,
+        "top_k": -1,
+        "max_tokens": 4096 * 16,
+        "seed": SEED,
+        "detokenize": True,
+        "repetition_penalty": 1.1,
+    }
+
+    sampling_params_list = [
+        thinker_sampling_params,
+        talker_sampling_params,
+        code2wav_sampling_params,
+    ]
+
+    prompt = query_map[args.query_type]()
+    extra_body = {
+        "sampling_params_list": sampling_params_list  # Optional, it has a default setting in stage_configs of the corresponding model.
+    }
+
+    if args.query_type == "use_audio_in_video":
+        extra_body["mm_processor_kwargs"] = {"use_audio_in_video": True}
+
+    chat_completion = client.chat.completions.create(
+        messages=[
+            get_system_prompt(),
+            prompt,
+        ],
+        model=model_name,
+        extra_body=extra_body,
+    )
+
+    count = 0
+    for choice in chat_completion.choices:
+        print(f"choice: {choice}")
+        if choice.message.audio:
+            audio_data = base64.b64decode(choice.message.audio.data)
+            audio_file_path = f"audio_{count}.wav"
+            with open(audio_file_path, "wb") as f:
+                f.write(audio_data)
+            print(f"Audio saved to {audio_file_path}")
+            count += 1
+        elif choice.message.content:
+            print("Chat completion output from text:", choice.message.content)
+
+
+def parse_args():
+    parser = FlexibleArgumentParser(description="Demo on using vLLM for offline inference with audio language models")
+    parser.add_argument(
+        "--query-type",
+        "-q",
+        type=str,
+        default="use_video",
+        choices=query_map.keys(),
+        help="Query type.",
+    )
+    parser.add_argument(
+        "--model",
+        "-m",
+        type=str,
+        default="Qwen/Qwen3-Omni-30B-A3B-Instruct",
+        help="Model Name / Path",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    run_multimodal_generation(args)

examples/online_serving/qwen3_omni/run_curl_multimodal_generation.sh

@@ -0,0 +1,187 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Default query type
+QUERY_TYPE="${1:-mixed_modalities}"
+
+# Validate query type
+if [[ ! "$QUERY_TYPE" =~ ^(mixed_modalities|use_audio_in_video|multi_audios|text)$ ]]; then
+    echo "Error: Invalid query type '$QUERY_TYPE'"
+    echo "Usage: $0 [mixed_modalities|use_audio_in_video|multi_audios|text]"
+    echo "  mixed_modalities: Audio + Image + Video + Text query"
+    echo "  use_audio_in_video: Video + Text query (with audio extraction from video)"
+    echo "  multi_audios: Two audio clips + Text query"
+    echo "  text: Text query"
+    exit 1
+fi
+
+SEED=42
+
+thinker_sampling_params='{
+  "temperature": 0.4,
+  "top_p": 0.9,
+  "top_k": 1,
+  "max_tokens": 16384,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.05,
+  "stop_token_ids": [151645]
+}'
+
+talker_sampling_params='{
+  "temperature": 0.9,
+  "top_k": 50,
+  "max_tokens": 4096,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.05,
+  "stop_token_ids": [2150]
+}'
+
+code2wav_sampling_params='{
+  "temperature": 0.0,
+  "top_p": 1.0,
+  "top_k": -1,
+  "max_tokens": 4096 * 16,
+  "seed": 42,
+  "detokenize": true,
+  "repetition_penalty": 1.1
+}'
+# Above is optional, it has a default setting in stage_configs of the corresponding model.
+
+# Define URLs for assets
+MARY_HAD_LAMB_AUDIO_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/mary_had_lamb.ogg"
+WINNING_CALL_AUDIO_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/winning_call.ogg"
+CHERRY_BLOSSOM_IMAGE_URL="https://vllm-public-assets.s3.us-west-2.amazonaws.com/vision_model_images/cherry_blossom.jpg"
+SAMPLE_VIDEO_URL="https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/sample_demo_1.mp4"
+
+# Build user content and extra fields based on query type
+case "$QUERY_TYPE" in
+  text)
+    user_content='[
+      {
+        "type": "text",
+        "text": "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+      }
+    ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+  mixed_modalities)
+    user_content='[
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$MARY_HAD_LAMB_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "image_url",
+          "image_url": {
+            "url": "'"$CHERRY_BLOSSOM_IMAGE_URL"'"
+          }
+        },
+        {
+          "type": "video_url",
+          "video_url": {
+            "url": "'"$SAMPLE_VIDEO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "What is recited in the audio? What is the content of this image? Why is this video funny?"
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+  use_audio_in_video)
+    user_content='[
+        {
+          "type": "video_url",
+          "video_url": {
+            "url": "'"$SAMPLE_VIDEO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Describe the content of the video, then convert what the baby say into text."
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs='{
+      "use_audio_in_video": true
+    }'
+    ;;
+  multi_audios)
+    user_content='[
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$MARY_HAD_LAMB_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "'"$WINNING_CALL_AUDIO_URL"'"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Are these two audio clips the same?"
+        }
+      ]'
+    sampling_params_list='[
+      '"$thinker_sampling_params"',
+      '"$talker_sampling_params"',
+      '"$code2wav_sampling_params"'
+    ]'
+    mm_processor_kwargs="{}"
+    ;;
+esac
+
+echo "Running query type: $QUERY_TYPE"
+echo ""
+
+
+output=$(curl -sS -X POST http://localhost:8091/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d @- <<EOF
+{
+  "model": "Qwen/Qwen2.5-Omni-7B",
+  "sampling_params_list": $sampling_params_list,
+  "mm_processor_kwargs": $mm_processor_kwargs,
+  "messages": [
+    {
+      "role": "system",
+      "content": [
+        {
+          "type": "text",
+          "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."
+        }
+      ]
+    },
+    {
+      "role": "user",
+      "content": $user_content
+    }
+  ]
+}
+EOF
+  )
+
+# Here it only shows the text content of the first choice. Audio content has many binaries, so it's not displayed here.
+echo "Output of request: $(echo "$output" | jq '.choices[0].message.content')"

mkdocs.yml

@@ -0,0 +1,151 @@
+site_name: vLLM-Omni
+site_description: Multi-modality models inference and serving with non-autoregressive structures
+site_author: vLLM-Omni Team
+site_url: https://vllm-project.github.io/vllm-omni/
+
+repo_name: vllm-project/vllm-omni
+repo_url: https://github.com/vllm-project/vllm-omni
+edit_uri: edit/main/docs/
+
+# Copyright
+copyright: Copyright © 2025 vLLM-Omni Team
+
+# Theme
+theme:
+  name: material
+  logo: source/logos/vllm-logo-only-light.ico
+  favicon: source/logos/vllm-logo-only-light.ico
+  palette:
+    # Palette toggle for automatic mode
+    - media: "(prefers-color-scheme)"
+      toggle:
+        icon: material/brightness-auto
+        name: Switch to light mode
+    # Palette toggle for light mode
+    - media: "(prefers-color-scheme: light)"
+      scheme: default
+      primary: white
+      toggle:
+        icon: material/brightness-7
+        name: Switch to dark mode
+    # Palette toggle for dark mode
+    - media: "(prefers-color-scheme: dark)"
+      scheme: slate
+      primary: black
+      toggle:
+        icon: material/brightness-2
+        name: Switch to system preference
+  features:
+    - content.action.edit
+    - content.code.copy
+    - navigation.instant
+    - navigation.instant.progress
+    - navigation.tracking
+    - navigation.tabs
+    - navigation.tabs.sticky
+    - navigation.sections
+    - navigation.indexes
+    - navigation.top
+    - search.suggest
+    - search.highlight
+    - search.share
+    - content.code.annotate
+    - content.tabs
+    - content.tooltips
+    - toc.follow
+  custom_dir: docs/mkdocs/overrides
+
+hooks:
+  - docs/mkdocs/hooks/generate_api_readme.py
+  - docs/mkdocs/hooks/url_schemes.py
+
+# Plugins
+plugins:
+  - meta
+  - search
+  - autorefs
+  - awesome-nav
+  - glightbox
+  - git-revision-date-localized:
+      # exclude files
+      exclude:
+        - api/*
+        - user_guide/examples/**
+        - contributing/design_documents/api_design_template.md
+        - DOCS_GUIDE.md
+  - minify:
+      minify_html: true
+      minify_js: true
+      minify_css: true
+      cache_safe: true
+      js_files: [docs/mkdocs/javascript/*.js]
+      css_files: [docs/mkdocs/stylesheets/*.css]
+  - api-autonav:
+      modules: ["vllm_omni"]
+      api_root_uri: "api"
+      nav_item_prefix: ""  # No prefix in navigation tree (clean names)
+      show_full_namespace: false  # Show only module name, not full path
+      exclude:
+        - "re:vllm_omni\\._.*"  # Internal modules
+  - mkdocstrings:
+      handlers:
+        python:
+          options:
+            show_symbol_type_heading: true
+            show_symbol_type_toc: true
+            filters:
+              - "!^_"  # Exclude private members (methods/classes starting with underscore)
+            summary:
+              modules: true
+            show_if_no_docstring: true
+            show_signature_annotations: true
+            separate_signature: true
+            show_overloads: true
+            signature_crossrefs: true
+          inventories:
+          - https://docs.python.org/3/objects.inv
+          - https://typing-extensions.readthedocs.io/en/latest/objects.inv
+          - https://docs.aiohttp.org/en/stable/objects.inv
+          - https://pillow.readthedocs.io/en/stable/objects.inv
+          - https://numpy.org/doc/stable/objects.inv
+          - https://pytorch.org/docs/stable/objects.inv
+          - https://psutil.readthedocs.io/en/stable/objects.inv
+
+markdown_extensions:
+  - attr_list
+  - md_in_html
+  - admonition
+  - pymdownx.details
+  # For content tabs
+  - pymdownx.superfences
+  - pymdownx.tabbed:
+      slugify: !!python/object/apply:pymdownx.slugs.slugify
+        kwds:
+          case: lower
+      alternate_style: true
+  # For code highlighting
+  - pymdownx.highlight:
+      anchor_linenums: true
+      line_spans: __span
+      pygments_lang_class: true
+  - pymdownx.inlinehilite
+  - pymdownx.snippets
+  # For emoji and icons
+  - pymdownx.emoji:
+      emoji_index: !!python/name:material.extensions.emoji.twemoji
+      emoji_generator: !!python/name:material.extensions.emoji.to_svg
+  # For in page [TOC] (not sidebar)
+  - toc:
+      permalink: true
+  # For math rendering
+  - pymdownx.arithmatex:
+      generic: true
+
+extra_css:
+  - mkdocs/stylesheets/extra.css
+
+extra_javascript:
+  - mkdocs/javascript/mathjax.js
+  - https://unpkg.com/mathjax@3.2.2/es5/tex-mml-chtml.js
+  - mkdocs/javascript/edit_and_feedback.js
+  - mkdocs/javascript/slack_and_forum.js

pyproject.toml

@@ -0,0 +1,170 @@
+[build-system]
+requires = ["setuptools>=61.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "vllm-omni"
+version = "0.11.0rc1"
+description = "vLLM-Omni: Multi-modality models inference and serving with non-autoregressive structures"
+readme = "README.md"
+requires-python = ">=3.9,<3.14"  # Align with vLLM v0.11 supported Python versions
+license = {text = "Apache-2.0"}
+authors = [
+    {name = "vLLM-Omni Team", email = "vllm-project@gmail.com"}
+]
+keywords = ["vllm", "multimodal", "diffusion", "transformer", "inference", "serving"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "Intended Audience :: Science/Research",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+]
+
+dependencies = [
+    # Core runtime dependencies (required for actual usage)
+    "omegaconf>=2.3.0",
+    # "decord>=0.6.0", # install manually
+    "librosa>=0.11.0",
+    "resampy>=0.4.3",
+    "diffusers==0.35.2",
+    "accelerate==1.12.0",
+    "gradio==5.50",
+    "soundfile>=0.13.1"
+    # "vllm==0.11.0",  # TODO: fix the entrypoints overwrite problem
+]
+
+[project.optional-dependencies]
+
+dev = [
+    "pytest>=7.0.0",
+    "pytest-asyncio>=0.21.0",
+    "pytest-cov>=4.0.0",
+    "black>=23.0.0",
+    "isort>=5.12.0",
+    "mypy>=1.0.0",
+    "pre-commit>=3.0.0",
+]
+
+docs = [
+    "mkdocs>=1.5.0",
+    "mkdocs-api-autonav",
+    "mkdocs-material",
+    "mkdocstrings-python",
+    "mkdocs-gen-files",
+    "mkdocs-awesome-nav",
+    "mkdocs-glightbox",
+    "mkdocs-git-revision-date-localized-plugin",
+    "mkdocs-minify-plugin",
+    "regex",
+    "pydantic",
+    "black>=23.0.0",
+]
+
+
+[project.urls]
+Homepage = "https://github.com/vllm-project/vllm-omni"
+Repository = "https://github.com/vllm-project/vllm-omni"
+Documentation = "https://vllm-omni.readthedocs.io"
+"Bug Tracker" = "https://github.com/vllm-project/vllm-omni/issues"
+
+[project.scripts]
+vllm = "vllm_omni.entrypoints.cli.main:main"
+vllm-omni = "vllm_omni.entrypoints.cli.main:main"
+
+
+[tool.setuptools.packages.find]
+where = ["."]
+include = ["vllm_omni*"]
+
+[tool.setuptools.package-data]
+"vllm_omni.model_executor.stage_configs" = ["*.yaml"]
+
+[tool.ruff]
+line-length = 120
+exclude = [
+    ".eggs",
+    ".git",
+    ".hg",
+    ".mypy_cache",
+    ".tox",
+    ".venv",
+    "build",
+    "dist",
+    "vllm_omni.egg-info",
+]
+
+[tool.ruff.lint]
+select = [
+    "E",  # pycodestyle errors
+    "W",  # pycodestyle warnings
+    "F",  # pyflakes
+    "I",  # isort (handled separately, but included for compatibility)
+    "N",  # pep8-naming
+    "UP", # pyupgrade
+]
+ignore = [
+    "E203",  # whitespace before ':' (conflicts with black)
+    # W503 is not needed in ruff as it's compatible with black by default
+    "N801",  # class names should use CapWords convention
+    "N802",  # function name should follow snake_case
+    "N806",  # variable in function should follow snake_case
+    "N812",  # lowercase imported as non-lowercase: functional as F
+]
+
+[tool.ruff.lint.per-file-ignores]
+"examples/**" = ["E501"]  # Allow long lines in examples
+"tests/**" = ["E501"]  # Allow long lines in tests
+
+[tool.mypy]
+python_version = "3.12, 3.13"
+warn_return_any = true
+warn_unused_configs = true
+disallow_untyped_defs = true
+disallow_incomplete_defs = true
+check_untyped_defs = true
+disallow_untyped_decorators = true
+no_implicit_optional = true
+warn_redundant_casts = true
+warn_unused_ignores = true
+warn_no_return = true
+warn_unreachable = true
+strict_equality = true
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+python_files = ["test_*.py", "*_test.py"]
+python_classes = ["Test*"]
+python_functions = ["test_*"]
+addopts = [
+    "--strict-markers",
+    "--strict-config",
+    "--cov=vllm_omni",
+    "--cov-report=term-missing",
+    "--cov-report=html",
+    "--cov-report=xml",
+]
+markers = [
+    "unit: Unit tests",
+    "integration: Integration tests",
+    "benchmark: Benchmark tests",
+    "slow: Slow tests",
+]
+
+[tool.typos.default]
+extend-ignore-identifiers-re = [
+    ".*_thw",
+    ".*thw",
+    "ein",
+    ".*arange",
+    ".*MoBA",
+    ".*temperal_downsample",
+    ".*nothink.*",
+    ".*NOTHINK.*"
+]

scripts/build_wheel.sh

@@ -0,0 +1,139 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd -- "${SCRIPT_DIR}/.." && pwd)"
+PROJECT_NAME="vllm-omni"
+RUN_QUALITY=false
+SKIP_CLEAN=false
+CREATE_VENV=false
+VENV_DIR=".venv-build"
+PYTHON_BIN="python"
+UV_BIN="uv"
+
+log() {
+  local level="$1"
+  shift
+  printf '[%s] %s\n' "${level}" "$*"
+}
+
+abort() {
+  log "ERROR" "$*"
+  exit 1
+}
+
+usage() {
+  cat <<EOF
+Usage: $(basename "$0") [options]
+
+Options:
+  --run-quality    Run pre-commit, install dev deps, and pytest before building
+  --skip-clean     Skip removing previous build artifacts
+  --create-venv    Build inside a fresh virtual environment (default path: .venv-build)
+  --venv-dir PATH  Custom directory for the virtual environment (implies --create-venv)
+  --python PATH    Python executable to use (default: python)
+  -h, --help       Show this help message
+EOF
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --run-quality)
+      RUN_QUALITY=true
+      ;;
+    --skip-clean)
+      SKIP_CLEAN=true
+      ;;
+    --create-venv)
+      CREATE_VENV=true
+      ;;
+    --venv-dir)
+      CREATE_VENV=true
+      shift
+      [[ $# -gt 0 ]] || abort "--venv-dir requires a path"
+      VENV_DIR="$1"
+      ;;
+    --python)
+      shift
+      [[ $# -gt 0 ]] || abort "--python requires a path"
+      PYTHON_BIN="$1"
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      usage
+      abort "Unknown option: $1"
+      ;;
+  esac
+  shift
+done
+
+HOST_PYTHON="${PYTHON_BIN}"
+
+log "INFO" "Switching to repository root: ${REPO_ROOT}"
+cd "${REPO_ROOT}" || abort "Cannot enter repository root"
+
+[[ -f pyproject.toml ]] || abort "pyproject.toml not found, please ensure correct script location"
+
+ensure_uv() {
+  if ! command -v "${UV_BIN}" >/dev/null 2>&1; then
+    log "INFO" "uv not found, installing via ${HOST_PYTHON}"
+    "${HOST_PYTHON}" -m pip install --upgrade pip
+    "${HOST_PYTHON}" -m pip install uv
+  fi
+}
+
+ensure_uv
+
+if [[ "${CREATE_VENV}" == "true" ]]; then
+  log "INFO" "Creating fresh virtual environment at ${VENV_DIR} via uv"
+  "${UV_BIN}" venv --python "${HOST_PYTHON}" --seed "${VENV_DIR}"
+  PYTHON_BIN="${VENV_DIR}/bin/python"
+  [[ -x "${PYTHON_BIN}" ]] || abort "Failed to locate python inside ${VENV_DIR}"
+  log "INFO" "Installing build module inside virtual environment"
+  "${UV_BIN}" pip install --python "${PYTHON_BIN}" build
+else
+  log "INFO" "Ensuring build module is available via uv pip"
+  "${UV_BIN}" pip install --python "${PYTHON_BIN}" build
+fi
+
+log "INFO" "Checking build module"
+if ! "${PYTHON_BIN}" -m build --version >/dev/null 2>&1; then
+  abort "${PYTHON_BIN} -m build is not available, install build first"
+fi
+
+run_quality_steps() {
+  log "INFO" "Running quality checks"
+  "${UV_BIN}" pip install --python "${PYTHON_BIN}" -e ".[dev]"
+  "${PYTHON_BIN}" -m pre_commit run --all-files
+  "${PYTHON_BIN}" -m pytest tests/ -v -m "not slow"
+}
+
+if [[ "${RUN_QUALITY}" == "true" ]]; then
+  run_quality_steps
+else
+  log "INFO" "Quality steps available via --run-quality"
+  log "INFO" "  - pre-commit run --all-files"
+  log "INFO" "  - pip install -e '.[dev]'"
+  log "INFO" "  - pytest tests/ -v -m \"not slow\""
+fi
+
+cleanup_artifacts() {
+  log "INFO" "Cleaning previous build artifacts"
+  rm -rf build dist "${PROJECT_NAME}.egg-info" "${PROJECT_NAME//-/_}.egg-info"
+}
+
+if [[ "${SKIP_CLEAN}" == "true" ]]; then
+  log "INFO" "Skipping cleanup as requested"
+else
+  cleanup_artifacts
+fi
+
+log "INFO" "Building source and wheel distributions"
+"${PYTHON_BIN}" -m build
+
+log "INFO" "Build finished, artifacts:"
+ls -lh dist

tests/__init__.py

@@ -0,0 +1,6 @@
+"""
+Test suite for vLLM-omni.
+
+This package contains unit tests, integration tests, and benchmarks
+for vLLM-omni.
+"""

vllm_omni/__init__.py

@@ -0,0 +1,42 @@
+"""
+vLLM-Omni: Multi-modality models inference and serving with
+non-autoregressive structures.
+
+This package extends vLLM beyond traditional text-based, autoregressive
+generation to support multi-modality models with non-autoregressive
+structures and non-textual outputs.
+
+Architecture:
+- 🟡 Modified: vLLM components modified for multimodal support
+- 🔴 Added: New components for multimodal and non-autoregressive
+  processing
+"""
+
+try:
+    from . import patch  # noqa: F401
+except ModuleNotFoundError as exc:  # pragma: no cover - optional dependency
+    if exc.name != "vllm":
+        raise
+    # Allow importing vllm_omni without vllm (e.g., documentation builds)
+    patch = None  # type: ignore
+
+from .config import OmniModelConfig
+from .entrypoints.async_omni_llm import AsyncOmniLLM
+
+# Main entry points
+# from .entrypoints.omni import Omni
+
+from .version import __version__, __version_tuple__  # isort:skip
+
+
+__all__ = [
+    "__version__",
+    "__version_tuple__",
+    # Main components
+    # "Omni",
+    "AsyncOmniLLM",
+    # Configuration
+    "OmniModelConfig",
+    # All other components are available through their respective modules
+    # processors.*, schedulers.*, executors.*, etc.
+]

vllm_omni/config/__init__.py

@@ -0,0 +1,9 @@
+"""
+Configuration module for vLLM-Omni.
+"""
+
+from vllm_omni.config.model import OmniModelConfig
+
+__all__ = [
+    "OmniModelConfig",
+]

vllm_omni/config/model.py

@@ -0,0 +1,387 @@
+import json
+import warnings
+from importlib.util import find_spec
+from typing import Any, Literal, Optional
+
+import torch
+import vllm.envs as envs
+from omegaconf import DictConfig, OmegaConf
+from pydantic import ConfigDict
+from pydantic.dataclasses import dataclass
+from vllm.config import ModelConfig, config
+from vllm.config.model import (
+    _RUNNER_CONVERTS,
+    _RUNNER_TASKS,
+    ConvertType,
+    RunnerOption,
+    TaskOption,
+    _get_and_verify_dtype,
+    get_served_model_name,
+)
+from vllm.config.multimodal import MMCacheType, MMEncoderTPMode, MultiModalConfig
+from vllm.config.pooler import PoolerConfig
+from vllm.logger import init_logger
+from vllm.platforms import current_platform
+from vllm.transformers_utils.config import (
+    get_config,
+    get_hf_image_processor_config,
+    get_hf_text_config,
+    get_pooling_config,
+    is_interleaved,
+)
+from vllm.transformers_utils.utils import maybe_model_redirect
+
+import vllm_omni.model_executor.models as me_models
+
+logger = init_logger(__name__)
+
+
+@config
+@dataclass(config=ConfigDict(arbitrary_types_allowed=True))
+class OmniModelConfig(ModelConfig):
+    """Configuration for Omni models, extending the base ModelConfig.
+
+    This configuration class extends the base vLLM ModelConfig with
+    omni-specific fields for multi-stage pipeline processing.
+
+    Attributes:
+        stage_id: Identifier for the stage in a multi-stage pipeline (default: 0)
+        model_stage: Stage type identifier, e.g., "thinker" or "talker"
+            (default: "thinker")
+        model_arch: Model architecture name
+            (default: "Qwen2_5OmniForConditionalGeneration")
+        engine_output_type: Optional output type specification for the engine.
+            Used to route outputs to appropriate processors (e.g., "image",
+            "audio", "latents"). If None, output type is inferred.
+
+    Example:
+        >>> config = OmniModelConfig(
+        ...     stage_id=0,
+        ...     model_stage="thinker",
+        ...     model_arch="Qwen2_5OmniForConditionalGeneration"
+        ... )
+    """
+
+    stage_id: int = 0
+    model_stage: str = "thinker"
+    model_arch: str = "Qwen2_5OmniForConditionalGeneration"
+    engine_output_type: Optional[str] = None
+    hf_config_name: Optional[str] = None
+
+    @property
+    def registry(self):
+        return me_models.OmniModelRegistry
+
+    @property
+    def architectures(self) -> list[str]:
+        return [self.model_arch]
+
+    def draw_hf_text_config(self):
+        # transformers' get_text_config method is used to get the text config from thinker_config.
+        # to handle the case that each model stage has their own text config,
+        # we need to draw the text config from the corresponding model stage.
+        if self.hf_config_name is None:
+            return get_hf_text_config(self.hf_config)
+        return getattr(self.hf_config, self.hf_config_name).get_text_config()
+
+    def __post_init__(
+        self,
+        # Multimodal config init vars
+        limit_mm_per_prompt: Optional[dict[str, int]],
+        media_io_kwargs: Optional[dict[str, dict[str, Any]]],
+        mm_processor_kwargs: Optional[dict[str, Any]],
+        mm_processor_cache_gb: Optional[float],
+        mm_processor_cache_type: Optional[MMCacheType],
+        mm_shm_cache_max_object_size_mb: Optional[int],
+        mm_encoder_tp_mode: Optional[MMEncoderTPMode],
+        interleave_mm_strings: Optional[bool],
+        skip_mm_profiling: Optional[bool],
+        video_pruning_rate: Optional[float],
+    ) -> None:
+        # Set the default seed to 0 in V1.
+        if envs.VLLM_USE_V1 and self.seed is None:
+            self.seed = 0
+            if not envs.VLLM_ENABLE_V1_MULTIPROCESSING:
+                logger.warning(
+                    "The global random seed is set to %d. Since "
+                    "VLLM_ENABLE_V1_MULTIPROCESSING is set to False, this may "
+                    "affect the random state of the Python process that "
+                    "launched vLLM.",
+                    self.seed,
+                )
+
+        # Keep set served_model_name before maybe_model_redirect(self.model)
+        self.served_model_name = get_served_model_name(self.model, self.served_model_name)
+        self.model = maybe_model_redirect(self.model)
+        # The tokenizer is consistent with the model by default.
+        if self.tokenizer is None:
+            self.tokenizer = self.model
+        if self.tokenizer_revision is None:
+            self.tokenizer_revision = self.revision
+        self.tokenizer = maybe_model_redirect(self.tokenizer)
+
+        if isinstance(self.hf_config_path, str):
+            self.hf_config_path = maybe_model_redirect(self.hf_config_path)
+
+        if callable(self.hf_overrides):
+            hf_overrides_kw: dict[str, Any] = {}
+            hf_overrides_fn = self.hf_overrides
+            dict_overrides: dict[str, Any] = {}
+        else:
+            # Separate dict overrides from flat ones; convert OmegaConf to dict
+            hf_overrides_kw = {}
+            dict_overrides = {}
+            for key, value in self.hf_overrides.items():
+                if isinstance(value, DictConfig):
+                    value = OmegaConf.to_container(value, resolve=True, structured_config_mode=None)
+                if isinstance(value, dict):
+                    dict_overrides[key] = value
+                else:
+                    hf_overrides_kw[key] = value
+            hf_overrides_fn = None
+
+        if self.rope_scaling:
+            hf_override: dict[str, Any] = {"rope_scaling": self.rope_scaling}
+            hf_overrides_kw.update(hf_override)
+            hf_overrides_str = json.dumps(hf_overrides_kw)
+            msg = (
+                "`--rope-scaling` will be removed in a future release. "
+                f"'Please instead use `--hf-overrides '{hf_overrides_str}'`"
+            )
+            warnings.warn(DeprecationWarning(msg), stacklevel=2)
+        if self.rope_theta is not None:
+            hf_override = {"rope_theta": self.rope_theta}
+            hf_overrides_kw.update(hf_override)
+            hf_overrides_str = json.dumps(hf_overrides_kw)
+            msg = (
+                "`--rope-theta` will be removed in a future release. "
+                f"'Please instead use `--hf-overrides '{hf_overrides_str}'`"
+            )
+            warnings.warn(DeprecationWarning(msg), stacklevel=2)
+
+        self.maybe_pull_model_tokenizer_for_runai(self.model, self.tokenizer)
+
+        if (backend := envs.VLLM_ATTENTION_BACKEND) and backend == "FLASHINFER" and find_spec("flashinfer") is None:
+            raise ValueError(
+                "VLLM_ATTENTION_BACKEND is set to FLASHINFER, but flashinfer "
+                "module was not found. See "
+                "https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile "
+                "for instructions on how to install it."
+            )
+
+        if self.override_attention_dtype is not None and not current_platform.is_rocm():
+            warnings.warn("override-attention-dtype is set but not using ROCm platform", stacklevel=2)
+
+        if self.enable_sleep_mode and not current_platform.is_sleep_mode_available():
+            raise ValueError("Sleep mode is not supported on current platform.")
+
+        hf_config = get_config(
+            self.hf_config_path or self.model,
+            self.trust_remote_code,
+            self.revision,
+            self.code_revision,
+            self.config_format,
+            hf_overrides_kw=hf_overrides_kw,
+            hf_overrides_fn=hf_overrides_fn,
+        )
+
+        if dict_overrides:
+            self._apply_dict_overrides(hf_config, dict_overrides)
+
+        self.hf_config = hf_config
+        self.hf_text_config = self.draw_hf_text_config()
+        self.attention_chunk_size = getattr(self.hf_text_config, "attention_chunk_size", None)
+        self.encoder_config = self._get_encoder_config()
+        self.hf_image_processor_config = get_hf_image_processor_config(
+            self.model, hf_token=self.hf_token, revision=self.revision
+        )
+
+        architectures = self.architectures
+        registry = self.registry
+        is_generative_model = registry.is_text_generation_model(architectures, self)
+        is_pooling_model = registry.is_pooling_model(architectures, self)
+
+        def _task_to_convert(task: TaskOption) -> ConvertType:
+            if task == "embedding" or task == "embed":
+                return "embed"
+            if task == "classify":
+                return "classify"
+            if task == "reward":
+                return "reward"
+            if task == "score":
+                new_task = self._get_default_pooling_task(architectures)
+                return "classify" if new_task == "classify" else "embed"
+
+            return "none"
+
+        if self.task is not None:
+            runner: RunnerOption = "auto"
+            convert: ConvertType | Literal["auto"] = "auto"  # type: ignore[name-defined]
+            msg_prefix = (
+                "The 'task' option has been deprecated and will be removed in v0.13.0 or v1.0, whichever comes first."
+            )
+            msg_hint = "Please remove this option."
+
+            is_generative_task = self.task in _RUNNER_TASKS["generate"]
+            is_pooling_task = self.task in _RUNNER_TASKS["pooling"]
+
+            if is_generative_model and is_pooling_model:
+                if is_generative_task:
+                    runner = "generate"
+                    convert = "auto"  # type: ignore[assignment]
+                    msg_hint = (
+                        "Please replace this option with `--runner "
+                        "generate` to continue using this model "
+                        "as a generative model."
+                    )
+                elif is_pooling_task:
+                    runner = "pooling"
+                    convert = "auto"  # type: ignore[assignment]
+                    msg_hint = (
+                        "Please replace this option with `--runner "
+                        "pooling` to continue using this model "
+                        "as a pooling model."
+                    )
+                else:  # task == "auto"
+                    pass
+            elif is_generative_model or is_pooling_model:
+                if is_generative_task:
+                    runner = "generate"
+                    convert = "auto"  # type: ignore[assignment]
+                    msg_hint = "Please remove this option"
+                elif is_pooling_task:
+                    runner = "pooling"
+                    convert = _task_to_convert(self.task)
+                    msg_hint = (
+                        "Please replace this option with `--convert "
+                        f"{convert}` to continue using this model "
+                        "as a pooling model."
+                    )
+                else:  # task == "auto"
+                    pass
+            else:
+                raise AssertionError(
+                    f"The model should be a generative or pooling model when task is set to {self.task!r}."
+                )
+
+            self.runner = runner
+            self.convert = convert  # type: ignore[assignment]
+
+            msg = f"{msg_prefix} {msg_hint}"
+            warnings.warn(msg, DeprecationWarning, stacklevel=2)
+
+        self.runner_type = self._get_runner_type(architectures, self.runner)
+        self.convert_type = self._get_convert_type(architectures, self.runner_type, self.convert)
+
+        if self.runner_type == "generate" and not is_generative_model:
+            generate_converts = _RUNNER_CONVERTS["generate"]
+            if self.convert_type not in generate_converts:
+                raise ValueError("This model does not support `--runner generate`.")
+        if self.runner_type == "pooling" and not is_pooling_model:
+            pooling_converts = _RUNNER_CONVERTS["pooling"]
+            if self.convert_type not in pooling_converts:
+                convert_option = "<" + "|".join(pooling_converts) + ">"
+                raise ValueError(
+                    "This model does not support `--runner pooling`. "
+                    f"You can pass `--convert {convert_option} to adapt "
+                    "it into a pooling model."
+                )
+
+        # Note: Initialize these attributes early because transformers fallback
+        # may fail to load dynamic modules in child processes
+        model_info, arch = registry.inspect_model_cls(architectures, self)
+        self._model_info = model_info
+        self._architecture = arch
+        logger.info("Resolved architecture: %s", arch)
+
+        # Init pooler config if needed
+        if self.runner_type == "pooling":
+            if self.override_pooler_config is not None:
+                logger.warning_once(
+                    "`override_pooler_config` is deprecated and will be "
+                    "removed in v0.12.0 or v1.0.0, whichever is sooner. "
+                    "Please use `pooler_config` instead."
+                )
+
+                if isinstance(self.override_pooler_config, dict):
+                    self.pooler_config = PoolerConfig(**self.override_pooler_config)
+                else:
+                    self.pooler_config = self.override_pooler_config
+
+            if self.pooler_config is None:
+                self.pooler_config = PoolerConfig()
+
+            base_config = get_pooling_config(self.model, self.revision)
+            if base_config is not None:
+                for k, v in base_config.items():
+                    if getattr(self.pooler_config, k) is None:
+                        setattr(self.pooler_config, k, v)
+
+            default_pooling_type = self._model_info.default_pooling_type
+            if self.pooler_config.pooling_type is None:
+                self.pooler_config.pooling_type = default_pooling_type
+
+        self.dtype: torch.dtype = _get_and_verify_dtype(
+            self.model,
+            self.hf_config,
+            self.dtype,
+            is_pooling_model=self.runner_type == "pooling",
+            revision=self.revision,
+        )
+
+        # Interleaved attention is not supported by some backends in V0
+        if (
+            not self.disable_sliding_window
+            and is_interleaved(self.hf_text_config)
+            and not envs.VLLM_USE_V1
+            and (backend := envs.VLLM_ATTENTION_BACKEND) in ("XFORMERS", "FLASHINFER")
+        ):
+            logger.warning_once(
+                "%s has interleaved attention, which is currently not "
+                "supported by the %s backend. Disabling sliding window and "
+                "capping the max length to the sliding window size (%d).",
+                self.hf_text_config.model_type,
+                backend,
+                self.hf_text_config.sliding_window,
+            )
+            self.disable_sliding_window = True
+
+        self.original_max_model_len = self.max_model_len
+        self.max_model_len = self.get_and_verify_max_len(self.max_model_len)
+        # Init multimodal config if needed
+        if self._model_info.supports_multimodal:
+            if mm_encoder_tp_mode == "data" and not self._model_info.supports_multimodal_encoder_tp_data:
+                logger.warning_once(
+                    "This model does not support `--mm-encoder-tp-mode data`. "
+                    "Falling back to `--mm-encoder-tp-mode weights`."
+                )
+                mm_encoder_tp_mode = "weights"
+
+            mm_config_kwargs = dict(
+                limit_per_prompt=limit_mm_per_prompt,
+                media_io_kwargs=media_io_kwargs,
+                mm_processor_kwargs=mm_processor_kwargs,
+                mm_processor_cache_gb=mm_processor_cache_gb,
+                mm_processor_cache_type=mm_processor_cache_type,
+                mm_shm_cache_max_object_size_mb=mm_shm_cache_max_object_size_mb,
+                mm_encoder_tp_mode=mm_encoder_tp_mode,
+                interleave_mm_strings=interleave_mm_strings,
+                skip_mm_profiling=skip_mm_profiling,
+                video_pruning_rate=video_pruning_rate,
+            )
+
+            mm_config_kwargs = {k: v for k, v in mm_config_kwargs.items() if v is not None}
+
+            self.multimodal_config = MultiModalConfig(**mm_config_kwargs)
+
+        if self.disable_sliding_window:
+            self.hf_text_config.sliding_window = None
+
+        if not self.skip_tokenizer_init:
+            self._verify_tokenizer_mode()
+
+        self.config_updated = False
+
+        self._verify_quantization()
+        self._verify_cuda_graph()
+        self._verify_bnb_config()

vllm_omni/core/dit_cache_manager.py

@@ -0,0 +1,13 @@
+"""
+DiT Cache Manager for vLLM-Omni.
+
+This module provides caching functionality for DiT (Diffusion Transformer) models
+to optimize inference performance and memory usage.
+"""
+
+
+class DiTCacheManager:
+    """Manages DiT-specific caching for optimized inference."""
+
+    def __init__(self, config):
+        pass

vllm_omni/core/sched/__init__.py

@@ -0,0 +1,13 @@
+"""
+Scheduling components for vLLM-Omni.
+"""
+
+from .diffusion_scheduler import DiffusionScheduler
+from .output import OmniNewRequestData
+from .scheduler import OmniScheduler
+
+__all__ = [
+    "OmniScheduler",
+    "DiffusionScheduler",
+    "OmniNewRequestData",
+]

vllm_omni/core/sched/diffusion_scheduler.py

@@ -0,0 +1,315 @@
+import time
+from collections import defaultdict
+from typing import Optional
+
+from vllm.v1.core.kv_cache_manager import KVCacheBlocks
+from vllm.v1.core.sched.request_queue import create_request_queue
+from vllm.v1.core.sched.scheduler import EngineCoreOutputs, Request, RequestStatus, SchedulerOutput, SpecDecodingStats
+from vllm.v1.core.sched.utils import remove_all
+from vllm.v1.engine import EngineCoreEventType, EngineCoreOutput
+
+from vllm_omni.core.sched.output import OmniNewRequestData
+from vllm_omni.core.sched.scheduler import OmniScheduler
+from vllm_omni.outputs import OmniModelRunnerOutput
+
+
+class DiffusionScheduler(OmniScheduler):
+    """Scheduler for diffusion-based non-autoregressive stages.
+
+    Implements a fast-path scheduling strategy for diffusion models where
+    all input tokens are fed at once. Falls back to standard vLLM scheduling
+    if the token budget cannot accommodate all tokens in a single iteration.
+    """
+
+    def schedule(self) -> SchedulerOutput:
+        """Schedule requests for diffusion model execution.
+
+        Uses a fast-path approach that feeds all input tokens at once
+        (or 1 placeholder token if input is empty). Falls back to
+        standard scheduling if token budget is insufficient.
+
+        Returns:
+            SchedulerOutput containing scheduled requests and token allocations
+        """
+
+        # Select requests with zero prompt and using pooling
+        # (diffusion results returned via pooler_output)
+        token_budget = self.max_num_scheduled_tokens
+        capacity = self.max_num_running_reqs - len(self.running)
+        scheduled_timestamp = time.monotonic()
+
+        scheduled_new_reqs: list[Request] = []
+        scheduled_resumed_reqs: list[Request] = []
+        scheduled_running_reqs: list[Request] = []
+
+        req_to_new_blocks: dict[str, KVCacheBlocks] = {}
+        num_scheduled_tokens: dict[str, int] = {}
+        scheduled_spec_decode_tokens: dict[str, list[int]] = {}
+        scheduled_encoder_inputs: dict[str, list[int]] = {}
+
+        # Temporary queue: preserve waiting order, do not disturb non-diffusion requests
+        skipped_waiting_requests = create_request_queue(self.policy)
+
+        # Fast path selection and scheduling (treat all as diffusion requests,
+        # independent of pooling_params)
+        while self.waiting and token_budget > 0 and capacity > 0:
+            request = self.waiting.peek_request()
+            # Uniformly treat as diffusion. A feature flag can be added later
+            # via config or request tag.
+
+            # Allocate all input tokens for the request in one shot
+            # (allocate 1 placeholder if zero)
+            required_tokens = max(getattr(request, "num_prompt_tokens", 0), 1)
+            if required_tokens > token_budget:
+                # Insufficient budget to process all inputs at once;
+                # stop fast path attempt
+                break
+            num_new_tokens = required_tokens
+            new_blocks = self.kv_cache_manager.allocate_slots(
+                request,
+                num_new_tokens,
+                num_lookahead_tokens=self.num_lookahead_tokens,
+            )
+            if new_blocks is None:
+                # Allocation failed (e.g., VRAM pressure); stop fast path and
+                # fall back to default scheduling
+                # Put the current request back to the head of the waiting queue
+                # Note: the original queue order is preserved
+                break
+
+            # Officially schedule this request
+            request = self.waiting.pop_request()
+            self.running.append(request)
+            request.status = RequestStatus.RUNNING
+            if self.log_stats:
+                request.record_event(EngineCoreEventType.SCHEDULED, scheduled_timestamp)
+
+            req_to_new_blocks[request.request_id] = new_blocks
+            num_scheduled_tokens[request.request_id] = num_new_tokens
+            token_budget -= num_new_tokens
+            capacity -= 1
+            scheduled_new_reqs.append(request)
+
+        # Return skipped waiting requests
+        if skipped_waiting_requests:
+            self.waiting.prepend_requests(skipped_waiting_requests)
+
+        # If fast path scheduled none, fall back to the original scheduling
+        if not num_scheduled_tokens:
+            return super().schedule()
+
+        # Compute common prefix blocks (aligned with v1)
+        num_common_prefix_blocks = [0] * len(self.kv_cache_config.kv_cache_groups)
+        if self.running:
+            any_request = self.running[0]
+            num_common_prefix_blocks = self.kv_cache_manager.get_num_common_prefix_blocks(
+                any_request, len(self.running)
+            )
+
+        # Assemble SchedulerOutput
+        new_reqs_data = [
+            OmniNewRequestData.from_request(req, req_to_new_blocks[req.request_id].get_block_ids())
+            for req in scheduled_new_reqs
+        ]
+        cached_reqs_data = self._make_cached_request_data(
+            scheduled_running_reqs,
+            scheduled_resumed_reqs,
+            num_scheduled_tokens,
+            scheduled_spec_decode_tokens,
+            req_to_new_blocks,
+        )
+        scheduled_requests = scheduled_new_reqs + scheduled_running_reqs + scheduled_resumed_reqs
+        structured_output_request_ids, grammar_bitmask = self.get_grammar_bitmask(
+            scheduled_requests, scheduled_spec_decode_tokens
+        )
+
+        total_num_scheduled_tokens = sum(num_scheduled_tokens.values())
+        scheduler_output = SchedulerOutput(
+            scheduled_new_reqs=new_reqs_data,
+            scheduled_cached_reqs=cached_reqs_data,
+            num_scheduled_tokens=num_scheduled_tokens,
+            total_num_scheduled_tokens=total_num_scheduled_tokens,
+            scheduled_spec_decode_tokens=scheduled_spec_decode_tokens,
+            scheduled_encoder_inputs=scheduled_encoder_inputs,
+            num_common_prefix_blocks=num_common_prefix_blocks,
+            finished_req_ids=self.finished_req_ids,
+            free_encoder_mm_hashes=self.encoder_cache_manager.get_freed_mm_hashes(),
+            structured_output_request_ids=structured_output_request_ids,
+            grammar_bitmask=grammar_bitmask,
+        )
+
+        # KVTransfer: package metadata
+        if self.connector is not None:
+            meta = self.connector.build_connector_meta(scheduler_output)
+            scheduler_output.kv_connector_metadata = meta
+
+        # Update internal state (advance num_computed_tokens, free encoder inputs,
+        # etc.)
+        self._update_after_schedule(scheduler_output)
+        return scheduler_output
+
+    """
+    Scheduler for the diffusion model.
+    This scheduler is modified to stop the request immediately for the diffusion model.
+    This is because the diffusion model can generate the final image/audio in one step.
+    Note: This is just a minimal modification to the original scheduler,
+    and there should be some further efforts to optimize the scheduler.
+    The original scheduler is still used for the AR model.
+    """
+
+    def update_from_output(
+        self,
+        scheduler_output: SchedulerOutput,
+        model_runner_output: OmniModelRunnerOutput,
+    ) -> dict[int, EngineCoreOutputs]:
+        """Update the scheduler state based on the model runner output.
+
+        This method is modified to stop the request immediately for the diffusion model.
+        """
+        sampled_token_ids = model_runner_output.sampled_token_ids
+        logprobs = model_runner_output.logprobs
+        prompt_logprobs_dict = model_runner_output.prompt_logprobs_dict
+        num_scheduled_tokens = scheduler_output.num_scheduled_tokens
+        pooler_outputs = model_runner_output.pooler_output
+        num_nans_in_logits = model_runner_output.num_nans_in_logits
+        kv_connector_output = model_runner_output.kv_connector_output
+
+        outputs: dict[int, list[EngineCoreOutput]] = defaultdict(list)
+        spec_decoding_stats: Optional[SpecDecodingStats] = None
+        kv_connector_stats = kv_connector_output.kv_connector_stats if kv_connector_output else None
+
+        # NOTE(woosuk): As len(num_scheduled_tokens) can be up to 1K or more,
+        # the below loop can be a performance bottleneck. We should do our best
+        # to avoid expensive operations inside the loop.
+        stopped_running_reqs: set[Request] = set()
+        stopped_preempted_reqs: set[Request] = set()
+        for req_id, num_tokens_scheduled in num_scheduled_tokens.items():
+            assert num_tokens_scheduled > 0
+            request = self.requests.get(req_id)
+            if request is None:
+                # The request is already finished. This can happen if the
+                # request is aborted while the model is executing it (e.g.,
+                # in pipeline parallelism).
+                continue
+
+            req_index = model_runner_output.req_id_to_index[req_id]
+            generated_token_ids = sampled_token_ids[req_index] if sampled_token_ids else []
+
+            scheduled_spec_token_ids = scheduler_output.scheduled_spec_decode_tokens.get(req_id)
+            if scheduled_spec_token_ids:
+                num_draft_tokens = len(scheduled_spec_token_ids)
+                num_accepted = len(generated_token_ids) - 1
+                num_rejected = num_draft_tokens - num_accepted
+                # num_computed_tokens represents the number of tokens
+                # processed in the current step, considering scheduled
+                # tokens and rejections. If some tokens are rejected,
+                # num_computed_tokens is decreased by the number of rejected
+                # tokens.
+                request.num_computed_tokens -= num_rejected
+                spec_decoding_stats = self.make_spec_decoding_stats(
+                    spec_decoding_stats,
+                    num_draft_tokens=num_draft_tokens,
+                    num_accepted_tokens=num_accepted,
+                )
+
+            new_logprobs = None
+            new_token_ids = generated_token_ids
+            kv_transfer_params = None
+            status_before_stop = request.status
+            pooler_output = None
+            if pooler_outputs:
+                pooler_output = pooler_outputs[req_index]
+
+            # Diffusion request: completes in one step; mark finished and free resources
+            request.status = RequestStatus.FINISHED_STOPPED
+            # Optional: set a stop_reason for front-end clarity
+            # (does not affect protocol)
+            request.stop_reason = request.stop_reason or "diffusion_done"
+            kv_transfer_params = self._free_request(request)
+            if status_before_stop == RequestStatus.RUNNING:
+                stopped_running_reqs.add(request)
+            else:
+                stopped_preempted_reqs.add(request)
+
+            # Extract sample logprobs if needed.
+            if request.sampling_params is not None and request.sampling_params.logprobs is not None and logprobs:
+                # NOTE: once we support N tokens per step (spec decode),
+                # the outer lists can be of length > 1.
+                new_logprobs = logprobs.slice(req_index, req_index + 1)
+
+            if new_token_ids and self.structured_output_manager.should_advance(request):
+                # NOTE: structured_output_request should not be None if
+                # use_structured_output, we have check above, so safe to ignore
+                # type warning
+                request.structured_output_request.grammar.accept_tokens(req_id, new_token_ids)  # type: ignore[union-attr]  # noqa: E501
+
+            # spec_token_ids comes from the model runner output
+            if num_nans_in_logits is not None and req_id in num_nans_in_logits:
+                request.num_nans_in_logits = num_nans_in_logits[req_id]
+
+            # Get prompt logprobs for this request.
+            prompt_logprobs_tensors = prompt_logprobs_dict.get(req_id)
+            # Convert pooler_output tensor to dict format expected by OmniEngineCoreOutput
+            pooling_output_dict = None
+            if pooler_output is not None:
+                # Wrap tensor in dict to match OmniEngineCoreOutput.pooling_output type
+                # which expects Optional[dict[str, torch.Tensor]]
+                pooling_output_dict = {"model_outputs": pooler_output}
+            if new_token_ids or pooler_output is not None or kv_transfer_params:
+                # Add EngineCoreOutput for this Request.
+                outputs[request.client_index].append(
+                    EngineCoreOutput(
+                        request_id=req_id,
+                        new_token_ids=new_token_ids,
+                        finish_reason=request.get_finished_reason(),
+                        new_logprobs=new_logprobs,
+                        new_prompt_logprobs_tensors=prompt_logprobs_tensors,
+                        pooling_output=pooling_output_dict,
+                        stop_reason=request.stop_reason,
+                        events=request.take_events(),
+                        kv_transfer_params=kv_transfer_params,
+                        trace_headers=request.trace_headers,
+                        num_cached_tokens=request.num_cached_tokens,
+                    )
+                )
+            else:
+                # Invariant: EngineCore returns no partial prefill outputs.
+                assert not prompt_logprobs_tensors
+
+        # Remove the stopped requests from the running and waiting queues.
+        if stopped_running_reqs:
+            if stopped_running_reqs:
+                self.running = remove_all(self.running, stopped_running_reqs)
+        if stopped_preempted_reqs:
+            # This is a rare case and unlikely to impact performance.
+            self.waiting.remove_requests(stopped_preempted_reqs)
+
+        # KV Connector: update state for finished KV Transfers.
+        if model_runner_output.kv_connector_output:
+            self._update_from_kv_xfer_finished(model_runner_output.kv_connector_output)
+
+        # Create EngineCoreOutputs for all clients that have requests with
+        # outputs in this step.
+        engine_core_outputs = {client_index: EngineCoreOutputs(outputs=outs) for client_index, outs in outputs.items()}
+
+        finished_req_ids = self.finished_req_ids_dict
+        if finished_req_ids:
+            # Include ids of requests that finished since last outputs
+            # were sent.
+            for client_index, finished_set in finished_req_ids.items():
+                # Set finished request set in EngineCoreOutputs for this client.
+                if (eco := engine_core_outputs.get(client_index)) is not None:
+                    eco.finished_requests = finished_set
+                else:
+                    engine_core_outputs[client_index] = EngineCoreOutputs(finished_requests=finished_set)
+            finished_req_ids.clear()
+
+        if (stats := self.make_stats(spec_decoding_stats, kv_connector_stats)) is not None:
+            # Return stats to only one of the front-ends.
+            if (eco := next(iter(engine_core_outputs.values()), None)) is None:
+                # We must return the stats even if there are no request
+                # outputs this step.
+                engine_core_outputs[0] = eco = EngineCoreOutputs()
+            eco.scheduler_stats = stats
+
+        return engine_core_outputs

vllm_omni/core/sched/generation_scheduler.py

@@ -0,0 +1,313 @@
+import time
+from collections import defaultdict
+from typing import Optional
+
+from vllm.v1.core.kv_cache_manager import KVCacheBlocks
+from vllm.v1.core.sched.request_queue import create_request_queue
+from vllm.v1.core.sched.scheduler import (
+    EngineCoreOutputs,
+    Request,
+    RequestStatus,
+    SchedulerOutput,
+    SpecDecodingStats,
+)
+from vllm.v1.core.sched.utils import remove_all
+from vllm.v1.engine import EngineCoreEventType, EngineCoreOutput
+
+from vllm_omni.core.sched.output import OmniNewRequestData
+from vllm_omni.core.sched.scheduler import OmniScheduler
+from vllm_omni.outputs import OmniModelRunnerOutput
+
+
+class GenerationScheduler(OmniScheduler):
+    def schedule(self) -> SchedulerOutput:
+        """Diffusion fast path:
+        - Feed all input tokens of the request at once
+          (if 0, allocate 1 placeholder token).
+        - If the token budget cannot be satisfied at once, fall back to the
+          default vLLM scheduling.
+        """
+
+        # Select requests with zero prompt and using pooling
+        # (diffusion results returned via pooler_output)
+        token_budget = self.max_num_scheduled_tokens
+        capacity = self.max_num_running_reqs - len(self.running)
+        scheduled_timestamp = time.monotonic()
+
+        scheduled_new_reqs: list[Request] = []
+        scheduled_resumed_reqs: list[Request] = []
+        scheduled_running_reqs: list[Request] = []
+
+        req_to_new_blocks: dict[str, KVCacheBlocks] = {}
+        num_scheduled_tokens: dict[str, int] = {}
+        scheduled_spec_decode_tokens: dict[str, list[int]] = {}
+        scheduled_encoder_inputs: dict[str, list[int]] = {}
+
+        # Temporary queue: preserve waiting order, do not disturb non-diffusion requests
+        skipped_waiting_requests = create_request_queue(self.policy)
+
+        # Fast path selection and scheduling (treat all as diffusion requests,
+        # independent of pooling_params)
+        while self.waiting and token_budget > 0 and capacity > 0:
+            request = self.waiting.peek_request()
+            # Uniformly treat as diffusion. A feature flag can be added later
+            # via config or request tag.
+
+            # Allocate all input tokens for the request in one shot
+            # (allocate 1 placeholder if zero)
+            required_tokens = max(getattr(request, "num_prompt_tokens", 0), 1)
+            if required_tokens > token_budget:
+                # Insufficient budget to process all inputs at once;
+                # stop fast path attempt
+                break
+            num_new_tokens = required_tokens
+            new_blocks = self.kv_cache_manager.allocate_slots(
+                request,
+                num_new_tokens,
+                num_lookahead_tokens=self.num_lookahead_tokens,
+            )
+            if new_blocks is None:
+                # Allocation failed (e.g., VRAM pressure); stop fast path and
+                # fall back to default scheduling
+                # Put the current request back to the head of the waiting queue
+                # Note: the original queue order is preserved
+                break
+
+            # Officially schedule this request
+            request = self.waiting.pop_request()
+            self.running.append(request)
+            request.status = RequestStatus.RUNNING
+            if self.log_stats:
+                request.record_event(EngineCoreEventType.SCHEDULED, scheduled_timestamp)
+
+            req_to_new_blocks[request.request_id] = new_blocks
+            num_scheduled_tokens[request.request_id] = num_new_tokens
+            token_budget -= num_new_tokens
+            capacity -= 1
+            scheduled_new_reqs.append(request)
+
+        # Return skipped waiting requests
+        if skipped_waiting_requests:
+            self.waiting.prepend_requests(skipped_waiting_requests)
+
+        # If fast path scheduled none, fall back to the original scheduling
+        if not num_scheduled_tokens:
+            return super().schedule()
+
+        # Compute common prefix blocks (aligned with v1)
+        num_common_prefix_blocks = [0] * len(self.kv_cache_config.kv_cache_groups)
+        if self.running:
+            any_request = self.running[0]
+            num_common_prefix_blocks = self.kv_cache_manager.get_num_common_prefix_blocks(
+                any_request, len(self.running)
+            )
+
+        # Assemble SchedulerOutput
+        new_reqs_data = [
+            OmniNewRequestData.from_request(req, req_to_new_blocks[req.request_id].get_block_ids())
+            for req in scheduled_new_reqs
+        ]
+        cached_reqs_data = self._make_cached_request_data(
+            scheduled_running_reqs,
+            scheduled_resumed_reqs,
+            num_scheduled_tokens,
+            scheduled_spec_decode_tokens,
+            req_to_new_blocks,
+        )
+        scheduled_requests = scheduled_new_reqs + scheduled_running_reqs + scheduled_resumed_reqs
+        structured_output_request_ids, grammar_bitmask = self.get_grammar_bitmask(
+            scheduled_requests, scheduled_spec_decode_tokens
+        )
+
+        total_num_scheduled_tokens = sum(num_scheduled_tokens.values())
+        scheduler_output = SchedulerOutput(
+            scheduled_new_reqs=new_reqs_data,
+            scheduled_cached_reqs=cached_reqs_data,
+            num_scheduled_tokens=num_scheduled_tokens,
+            total_num_scheduled_tokens=total_num_scheduled_tokens,
+            scheduled_spec_decode_tokens=scheduled_spec_decode_tokens,
+            scheduled_encoder_inputs=scheduled_encoder_inputs,
+            num_common_prefix_blocks=num_common_prefix_blocks,
+            finished_req_ids=self.finished_req_ids,
+            free_encoder_mm_hashes=self.encoder_cache_manager.get_freed_mm_hashes(),
+            structured_output_request_ids=structured_output_request_ids,
+            grammar_bitmask=grammar_bitmask,
+        )
+
+        # KVTransfer: package metadata
+        if self.connector is not None:
+            meta = self.connector.build_connector_meta(scheduler_output)
+            scheduler_output.kv_connector_metadata = meta
+
+        # Update internal state (advance num_computed_tokens, free encoder inputs,
+        # etc.)
+        self._update_after_schedule(scheduler_output)
+        return scheduler_output
+
+    """
+    Scheduler for the diffusion model.
+    This scheduler is modified to stop the request immediately for the diffusion model.
+    This is because the diffusion model can generate the final image/audio in one step.
+    Note: This is just a minimal modification to the original scheduler,
+    and there should be some further efforts to optimize the scheduler.
+    The original scheduler is still used for the AR model.
+    """
+
+    def update_from_output(
+        self,
+        scheduler_output: SchedulerOutput,
+        model_runner_output: OmniModelRunnerOutput,
+    ) -> dict[int, EngineCoreOutputs]:
+        """Update the scheduler state based on the model runner output.
+
+        This method is modified to stop the request immediately for the diffusion model.
+        """
+        sampled_token_ids = model_runner_output.sampled_token_ids
+        logprobs = model_runner_output.logprobs
+        prompt_logprobs_dict = model_runner_output.prompt_logprobs_dict
+        num_scheduled_tokens = scheduler_output.num_scheduled_tokens
+        pooler_outputs = model_runner_output.pooler_output
+        num_nans_in_logits = model_runner_output.num_nans_in_logits
+        kv_connector_output = model_runner_output.kv_connector_output
+
+        outputs: dict[int, list[EngineCoreOutput]] = defaultdict(list)
+        spec_decoding_stats: Optional[SpecDecodingStats] = None
+        kv_connector_stats = kv_connector_output.kv_connector_stats if kv_connector_output else None
+
+        # NOTE(woosuk): As len(num_scheduled_tokens) can be up to 1K or more,
+        # the below loop can be a performance bottleneck. We should do our best
+        # to avoid expensive operations inside the loop.
+        stopped_running_reqs: set[Request] = set()
+        stopped_preempted_reqs: set[Request] = set()
+        for req_id, num_tokens_scheduled in num_scheduled_tokens.items():
+            assert num_tokens_scheduled > 0
+            request = self.requests.get(req_id)
+            if request is None:
+                # The request is already finished. This can happen if the
+                # request is aborted while the model is executing it (e.g.,
+                # in pipeline parallelism).
+                continue
+
+            req_index = model_runner_output.req_id_to_index[req_id]
+            generated_token_ids = sampled_token_ids[req_index] if sampled_token_ids else []
+
+            scheduled_spec_token_ids = scheduler_output.scheduled_spec_decode_tokens.get(req_id)
+            if scheduled_spec_token_ids:
+                num_draft_tokens = len(scheduled_spec_token_ids)
+                num_accepted = len(generated_token_ids) - 1
+                num_rejected = num_draft_tokens - num_accepted
+                # num_computed_tokens represents the number of tokens
+                # processed in the current step, considering scheduled
+                # tokens and rejections. If some tokens are rejected,
+                # num_computed_tokens is decreased by the number of rejected
+                # tokens.
+                request.num_computed_tokens -= num_rejected
+                spec_decoding_stats = self.make_spec_decoding_stats(
+                    spec_decoding_stats,
+                    num_draft_tokens=num_draft_tokens,
+                    num_accepted_tokens=num_accepted,
+                )
+
+            new_logprobs = None
+            new_token_ids = generated_token_ids
+            kv_transfer_params = None
+            status_before_stop = request.status
+            pooler_output = None
+            if pooler_outputs:
+                pooler_output = pooler_outputs[req_index]
+
+            # Diffusion request: completes in one step; mark finished and free resources
+            request.status = RequestStatus.FINISHED_STOPPED
+            # Optional: set a stop_reason for front-end clarity
+            # (does not affect protocol)
+            request.stop_reason = request.stop_reason  # or "generation_done"
+            kv_transfer_params = self._free_request(request)
+            if status_before_stop == RequestStatus.RUNNING:
+                stopped_running_reqs.add(request)
+            else:
+                stopped_preempted_reqs.add(request)
+
+            # Extract sample logprobs if needed.
+            if request.sampling_params is not None and request.sampling_params.logprobs is not None and logprobs:
+                # NOTE: once we support N tokens per step (spec decode),
+                # the outer lists can be of length > 1.
+                new_logprobs = logprobs.slice(req_index, req_index + 1)
+
+            if new_token_ids and self.structured_output_manager.should_advance(request):
+                # NOTE: structured_output_request should not be None if
+                # use_structured_output, we have check above, so safe to ignore
+                # type warning
+                request.structured_output_request.grammar.accept_tokens(  # type: ignore[union-attr]  # noqa: E501
+                    req_id, new_token_ids
+                )
+
+            # spec_token_ids comes from the model runner output
+            if num_nans_in_logits is not None and req_id in num_nans_in_logits:
+                request.num_nans_in_logits = num_nans_in_logits[req_id]
+
+            # Get prompt logprobs for this request.
+            prompt_logprobs_tensors = prompt_logprobs_dict.get(req_id)
+            # Convert pooler_output tensor to dict format expected by OmniEngineCoreOutput
+            pooling_output_dict = None
+            if pooler_output is not None:
+                # Wrap tensor in dict to match OmniEngineCoreOutput.pooling_output type
+                # which expects Optional[dict[str, torch.Tensor]]
+                pooling_output_dict = {"model_outputs": pooler_output}
+            if new_token_ids or pooler_output is not None or kv_transfer_params:
+                # Add EngineCoreOutput for this Request.
+                outputs[request.client_index].append(
+                    EngineCoreOutput(
+                        request_id=req_id,
+                        new_token_ids=new_token_ids,
+                        finish_reason=request.get_finished_reason(),
+                        new_logprobs=new_logprobs,
+                        new_prompt_logprobs_tensors=prompt_logprobs_tensors,
+                        pooling_output=pooling_output_dict,
+                        stop_reason=request.stop_reason,
+                        events=request.take_events(),
+                        kv_transfer_params=kv_transfer_params,
+                        trace_headers=request.trace_headers,
+                        num_cached_tokens=request.num_cached_tokens,
+                    )
+                )
+            else:
+                # Invariant: EngineCore returns no partial prefill outputs.
+                assert not prompt_logprobs_tensors
+
+        # Remove the stopped requests from the running and waiting queues.
+        if stopped_running_reqs:
+            if stopped_running_reqs:
+                self.running = remove_all(self.running, stopped_running_reqs)
+        if stopped_preempted_reqs:
+            # This is a rare case and unlikely to impact performance.
+            self.waiting.remove_requests(stopped_preempted_reqs)
+
+        # KV Connector: update state for finished KV Transfers.
+        if model_runner_output.kv_connector_output:
+            self._update_from_kv_xfer_finished(model_runner_output.kv_connector_output)
+
+        # Create EngineCoreOutputs for all clients that have requests with
+        # outputs in this step.
+        engine_core_outputs = {client_index: EngineCoreOutputs(outputs=outs) for client_index, outs in outputs.items()}
+
+        finished_req_ids = self.finished_req_ids_dict
+        if finished_req_ids:
+            # Include ids of requests that finished since last outputs
+            # were sent.
+            for client_index, finished_set in finished_req_ids.items():
+                # Set finished request set in EngineCoreOutputs for this client.
+                if (eco := engine_core_outputs.get(client_index)) is not None:
+                    eco.finished_requests = finished_set
+                else:
+                    engine_core_outputs[client_index] = EngineCoreOutputs(finished_requests=finished_set)
+            finished_req_ids.clear()
+
+        if (stats := self.make_stats(spec_decoding_stats, kv_connector_stats)) is not None:
+            # Return stats to only one of the front-ends.
+            if (eco := next(iter(engine_core_outputs.values()), None)) is None:
+                # We must return the stats even if there are no request
+                # outputs this step.
+                engine_core_outputs[0] = eco = EngineCoreOutputs()
+            eco.scheduler_stats = stats
+
+        return engine_core_outputs