update

Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>

.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

.github/ISSUE_TEMPLATE/config.yml

@@ -2,4 +2,4 @@ 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
+    about: Ask questions and discuss with other vLLM-Omni community members

.github/workflows/ci-all-files.yml

@@ -1,60 +0,0 @@
-name: CI All Files
-
-on:
-  workflow_dispatch:
-
-permissions:
-  contents: read
-
-jobs:
-  test-changed:
-    runs-on: ubuntu-latest
-    timeout-minutes: 30
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.12", "3.13"]
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0 # Need full history to detect changes
-
-      - name: Free disk space
-        run: |
-          # Remove large, preinstalled toolchains we don't need on the runner
-          sudo rm -rf /usr/share/dotnet
-          sudo rm -rf /opt/ghc
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /opt/hostedtoolcache/CodeQL
-          # Prune docker to reclaim space (safe on GH runners)
-          docker system prune -af || true
-          df -h
-
-      - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python-version }}
-
-      - name: Prepare environment with uv
-        run: |
-          python -m pip install --upgrade pip uv
-          uv venv --python "${{ matrix.python-version }}" --seed
-          echo "VIRTUAL_ENV=${PWD}/.venv" >> "$GITHUB_ENV"
-          echo "PATH=${PWD}/.venv/bin:$PATH" >> "$GITHUB_ENV"
-          source .venv/bin/activate
-          uv pip install -e ".[dev]"
-
-      - name: Run tests
-        run: |
-          if [ ! -d tests ]; then
-            echo "tests/ directory not found. Skipping pytest run."
-
-
-            exit 0
-          fi
-          TEST_FILES=$(find tests -type f \( -name "test_*.py" -o -name "*_test.py" \))
-          if [ -z "$TEST_FILES" ]; then
-            echo "No pytest-compatible test files detected under tests/. Skipping pytest run."
-            exit 0
-          fi
-          pytest tests/ -v -m "not slow"

.github/workflows/ci.yml

@@ -1,124 +0,0 @@
-name: CI
-
-on:
-  pull_request:
-    branches: [main]
-  workflow_dispatch:
-    inputs:
-      test_changed_only:
-        description: 'Only test changed files (true) or all files (false)'
-        required: false
-        default: 'true'
-        type: choice
-        options:
-          - 'true'
-          - 'false'
-      enable_changed_files_detection:
-        description: 'Enable changed files detection feature'
-        required: false
-        default: 'true'
-        type: choice
-        options:
-          - 'true'
-          - 'false'
-  workflow_call:
-    inputs:
-      test_changed_only:
-        description: 'Only test changed files (true) or all files (false)'
-        required: false
-        type: string
-        default: 'true'
-      enable_changed_files_detection:
-        description: 'Enable changed files detection feature'
-        required: false
-        type: string
-        default: 'true'
-
-permissions:
-  contents: read
-
-jobs:
-  test:
-    runs-on: ubuntu-latest
-    timeout-minutes: 30
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.12", "3.13"]
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Free disk space
-        run: |
-          # Remove large, preinstalled toolchains we don't need on the runner
-          sudo rm -rf /usr/share/dotnet
-          sudo rm -rf /opt/ghc
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /opt/hostedtoolcache/CodeQL
-          # Prune docker to reclaim space (safe on GH runners)
-          docker system prune -af || true
-          df -h
-
-      - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python-version }}
-
-      - name: Prepare environment with uv
-        run: |
-          python -m pip install --upgrade pip uv
-          uv venv --python "${{ matrix.python-version }}" --seed
-          echo "VIRTUAL_ENV=${PWD}/.venv" >> "$GITHUB_ENV"
-          echo "PATH=${PWD}/.venv/bin:${PATH}" >> "$GITHUB_ENV"
-          # Activate env to ensure subsequent commands succeed in this step
-          source .venv/bin/activate
-          uv pip install -e ".[dev]"
-
-      - name: Determine test scope
-        id: test-scope
-        env:
-          ENABLE_CHANGED_FILES_DETECTION: ${{ inputs.enable_changed_files_detection || 'true' }}
-          TEST_CHANGED_ONLY: ${{ inputs.test_changed_only || 'true' }}
-        run: |
-          # Determine if we should test only changed files
-          # If detection is disabled, always test all files
-          if [ "$ENABLE_CHANGED_FILES_DETECTION" != "true" ]; then
-            TEST_CHANGED_ONLY="false"
-          fi
-
-          echo "test_changed_only=$TEST_CHANGED_ONLY" >> "$GITHUB_OUTPUT"
-          echo "enable_detection=$ENABLE_CHANGED_FILES_DETECTION" >> "$GITHUB_OUTPUT"
-          echo "Test scope: $TEST_CHANGED_ONLY"
-          echo "Changed files detection: $ENABLE_CHANGED_FILES_DETECTION"
-
-      - name: Get changed test files
-        id: changed-tests
-        if: steps.test-scope.outputs.test_changed_only == 'true' && steps.test-scope.outputs.enable_detection == 'true'
-        env:
-          GITHUB_BASE_REF: ${{ github.event.pull_request.base.sha || '' }}
-          GITHUB_SHA: ${{ github.sha }}
-        run: |
-          chmod +x .github/scripts/detect_changed_tests.sh
-          .github/scripts/detect_changed_tests.sh \
-            "${{ github.event.pull_request.base.sha || github.event.before || '' }}" \
-            "${{ github.sha }}" \
-            "${{ github.event_name }}"
-
-      - name: Run tests (changed files only)
-        if: steps.test-scope.outputs.test_changed_only == 'true' && steps.changed-tests.outputs.has_tests == 'true'
-        run: |
-          pytest "${{ steps.changed-tests.outputs.test_files }}" -v -m "not slow"
-
-      - name: Run tests (all files)
-        if: steps.test-scope.outputs.test_changed_only == 'false' || steps.changed-tests.outputs.has_tests == 'false'
-        run: |
-          if [ ! -d tests ]; then
-            echo "tests/ directory not found. Skipping pytest run."
-            exit 0
-          fi
-          TEST_FILES=$(find tests -type f \( -name "test_*.py" -o -name "*_test.py" \))
-          if [ -z "$TEST_FILES" ]; then
-            echo "No pytest-compatible test files detected under tests/. Skipping pytest run."
-            exit 0
-          fi
-          pytest tests/ -v -m "not slow"

.github/workflows/pre-commit.yml

@@ -3,6 +3,8 @@ name: pre-commit
 on:
   pull_request:
     branches: [main]
+  push:
+    branches: [main]
   workflow_dispatch: # for manual trigger
 
 concurrency:

.gitignore

@@ -160,7 +160,7 @@ cython_debug/
 docker/
 
 # scripts
-scripts/
+/scripts/
 
 # tests
 tests/
@@ -175,7 +175,7 @@ tests/
 #  *.iws
 #  .idea/
 
-# vLLM-omni specific
+# vLLM-Omni specific
 # Model files and checkpoints
 checkpoints/
 *.bin

.pre-commit-config.yaml

@@ -1,3 +1,6 @@
+default_install_hook_types:
+  - pre-commit
+  - commit-msg
 repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
     rev: v4.6.0
@@ -30,3 +33,28 @@ repos:
     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

CONTRIBUTING.md

@@ -1,19 +1,3 @@
-# Contributing to vLLM
+# Contributing to vLLM-Omni
 
-## Getting Started
-
-Setup environment with uv
-```bash
-uv venv --python 3.12 --seed
-source .venv/bin/activate
-```
-
-Pre-commit has been linked to git commit. For run checks locally, please install pre-commit:
-```bash
-uv pip install pre-commit
-```
-
-Then run:
-```bash
-pre-commit run --show-diff-on-failure --color=always --all-files
-```
+You may find information about contributing to vLLM-Omni on [Contributing](docs/contributing/)

README.md

@@ -1,107 +1,78 @@
-# vLLM-omni: Multi-modal Extension for vLLM
+<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>
 
-vLLM-omni is designed to extend vLLM capabilities to support multi-modality model inference and serving, particularly focusing on non-autoregressive structures and non-textual outputs.
+<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>
 
-## 🎯 Overview
+---
 
-Traditional vLLM systems are limited to text-based, autoregressive generation. vLLM-omni addresses this limitation by enabling support for:
+*Latest News* 🔥
 
-- **Multi-modal Models**: Text, image, video, audio, and sensor data processing
-- **Non-autoregressive Architectures**: Diffusion Transformers (DiT) and other parallel generation models
-- **Heterogeneous Outputs**: Beyond traditional text generation to multimodal outputs
+- [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.
 
-## 🏗️ Architecture
+---
 
-vLLM-omni is built on a modular architecture that extends vLLM's core functionality:
+## 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:
 
-## 🚀 Key Features
+- **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
 
-### Multi-Engine Support
+<p align="center">
+  <picture>
+    <img alt="vllm-omni" src="./docs/source/architecture/omni-modality model architecture.png" width=55%>
+  </picture>
+</p>
 
-- **Autoregressive Engine**: Traditional text generation with enhanced KV-caching
-- **Diffusion Engine**: Support for DiT models and iterative generation
-- **Hybrid Engine**: Combined AR+DiT processing pipelines
+vLLM-Omni is fast with:
 
-### Modality Processing
+- 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
 
-- **Text**: Advanced tokenization and embedding generation
-- **Image**: Vision encoder integration (CLIP, etc.)
-- **Audio**: Speech processing and audio embedding
+vLLM-Omni is flexible and easy to use with:
 
-## 📋 Supported Models
+- 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
 
-### AR + Diffusion Transformer (DiT) Models
-- Qwen-omni (Thinker-Talker-Codec structure)
-- HunyunaImage 3.0 (Ongoing)
-- Qwen-Image (Ongoing)
+vLLM-Omni seamlessly supports most popular open-source models on HuggingFace, including:
 
-## 🛠️ Installation
+- Omni-modality models (e.g. Qwen-Omni)
+- Multi-modality generation models (e.g. Qwen-Image)
 
-### Environment setup
+## Getting Started
 
-Use Docker to keep consistent basic environment (Optional, Recommended)
-```bash
-docker run --gpus all --ipc=host --network=host -v $source_dir:$container_dir --name $container_name -it nvcr.io/nvidia/pytorch:25.01-py3 bash
-```
+Visit our [documentation](https://vllm-omni.readthedocs.io/en/latest/) to learn more.
 
-Set up basic uv environment
-```bash
-pip install uv
-uv venv --python 3.12 --seed
-source .venv/bin/activate
-```
+Visit our documentation to learn more.
 
-### Installation of vLLM (for users)
+- [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/)
 
-Now we build it based on vLLM v0.11.0. Please install it with command below.
-```bash
-uv pip install vllm==0.11.0 --torch-backend=auto
-```
+## Contributing
 
-### Installation of vLLM (for developers)
+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.
 
-If you want to check or debug with source code of vLLM, install stable release version of vllm with 0.11.0 from source with pre-built wheel file.
+## Weekly Meeting
 
-```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 .
-```
+- 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))
 
-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 .
-```
+## License
 
-### Verification for successful installation of vLLM
-Just run the command below. If no error, it demonstrates that the installation is successful.
-```bash
-python -c "import vllm._C"
-```
-
-### 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 .
-```
-
-## Run examples (Qwen2.5-omni)
-
-Please check the folder of [examples](examples)
-
-## Further details
-
-For detailed architecture design, see [vllm_omni_design.md](docs/architecture/vllm_omni_design.md) and [high_level_arch_design.md](docs/architecture/high_level_arch_design.md).
+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/api/overview.md

@@ -1,68 +0,0 @@
-# Summary
-
-## Entry Points
-
-Main entry points for vLLM-omni inference and serving.
-
-- [vllm_omni.entrypoints.omni_llm.OmniLLM][]
-- [vllm_omni.entrypoints.omni_llm.OmniStageLLM][]
-- [vllm_omni.entrypoints.omni_stage.OmniStage][]
-- [vllm_omni.entrypoints.cli.serve.OmniServeCommand][]
-
-## Inputs
-
-Input data structures for multi-modal inputs.
-
-- [vllm_omni.inputs.data.OmniTokenInputs][]
-- [vllm_omni.inputs.data.OmniTokensPrompt][]
-- [vllm_omni.inputs.data.OmniEmbedsPrompt][]
-- [vllm_omni.inputs.parse.parse_singleton_prompt_omni][]
-- [vllm_omni.inputs.preprocess.OmniInputPreprocessor][]
-
-## Engine
-
-Engine classes for offline and online inference.
-
-- [vllm_omni.engine.arg_utils.OmniEngineArgs][]
-- [vllm_omni.engine.processor.OmniProcessor][]
-- [vllm_omni.engine.output_processor.OmniRequestState][]
-- [vllm_omni.engine.output_processor.MultimodalOutputProcessor][]
-- [vllm_omni.engine.PromptEmbedsPayload][]
-- [vllm_omni.engine.AdditionalInformationPayload][]
-- [vllm_omni.engine.OmniEngineCoreRequest][]
-
-## Core
-
-Core scheduling and caching components.
-
-- [vllm_omni.core.sched.scheduler.OmniScheduler][]
-- [vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler][]
-- [vllm_omni.core.sched.output.OmniNewRequestData][]
-- [vllm_omni.core.dit_cache_manager.DiTCacheManager][]
-
-## Model Executor
-
-Model execution components.
-
-- [vllm_omni.model_executor.models.qwen2_5_omni.Qwen2_5OmniForConditionalGeneration][]
-- [vllm_omni.model_executor.models.qwen2_5_omni_talker.Qwen2_5OmniTalkerForConditionalGeneration][]
-- [vllm_omni.model_executor.models.qwen2_5_omni_thinker.Qwen2_5OmniThinkerForConditionalGeneration][]
-- [vllm_omni.model_executor.models.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavForConditionalGenerationVLLM][]
-- [vllm_omni.model_executor.layers.mrope.MRotaryEmbedding][]
-- [vllm_omni.model_executor.models.registry.OmniModelRegistry][]
-
-## Configuration
-
-Configuration classes.
-
-- [vllm_omni.config.OmniModelConfig][]
-
-## Workers
-
-Worker classes and model runners for distributed inference.
-
-- [vllm_omni.worker.gpu_ar_worker.GPUARWorker][]
-- [vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker][]
-- [vllm_omni.worker.gpu_model_runner.OmniGPUModelRunner][]
-- [vllm_omni.worker.gpu_ar_model_runner.GPUARModelRunner][]
-- [vllm_omni.worker.gpu_diffusion_model_runner.GPUDiffusionModelRunner][]

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

@@ -1,6 +1,6 @@
 # Documentation Build Guide
 
-This directory contains the source files for the vLLM-omni documentation.
+This directory contains the source files for the vLLM-Omni documentation.
 
 ## Building Documentation Locally
 
@@ -38,8 +38,8 @@ The documentation automatically extracts docstrings from the code using mkdocstr
 Example docstring:
 
 ```python
-class OmniLLM:
-    """Main entry point for vLLM-omni inference.
+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.
@@ -50,7 +50,7 @@ class OmniLLM:
         **kwargs: Additional arguments passed to the engine
 
     Example:
-        >>> llm = OmniLLM(model="Qwen/Qwen2.5-Omni")
+        >>> llm = Omni(model="Qwen/Qwen2.5-Omni")
         >>> outputs = llm.generate(prompts="Hello")
     """
 ```

docs/contributing/README.md

@@ -1,55 +1,49 @@
-# Contributing to vLLM-omni
+# Contributing to vLLM-Omni
 
-Thank you for your interest in contributing to vLLM-omni! This document provides guidelines and instructions for contributing.
+Thank you for your interest in contributing to vLLM-Omni! This document provides guidelines and instructions for contributing.
 
 ## Getting Started
 
-### Development Environment
-
-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:
+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
 
-Install vLLM from source:
+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.
 
-```bash
-git clone https://github.com/vllm-project/vllm.git
-cd vllm
-git checkout v0.11.0
-VLLM_USE_PRECOMPILED=1 uv pip install -e .
-```
+### Development Environment for vLLM-Omni
 
-Install vLLM-omni from source:
+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 .
+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.
+    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:
+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.
+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 -a  # runs on all files (short for --all-files)
+    pre-commit run --show-diff-on-failure --color=always --all-files  # runs on all files (short for --all-files)
     ```
 
 ### Documentation
@@ -78,20 +72,14 @@ For additional features and advanced configurations, refer to the:
 
 ### Testing
 
-vLLM-omni uses `pytest` to test the codebase.
+vLLM-Omni uses `pytest` to test the codebase.
 
 ```bash
-# Install the test dependencies used in CI (CUDA only)
-uv pip install -e ".[dev]" --torch-backend=auto
-
-# Install some common test dependencies (hardware agnostic)
-uv pip install pytest pytest-asyncio
-
 # Run all tests
 pytest tests/
 
 # Run tests for a single test file with detailed output
-pytest -s -v tests/test_logger.py
+pytest -s -v tests/test_omni_llm.py
 ```
 
 !!! warning
@@ -106,7 +94,7 @@ If you encounter a bug or have a feature request, please search existing issues
 
 ## 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.
+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
 
@@ -128,9 +116,9 @@ Only specific types of PRs will be reviewed. The PR title is prefixed appropriat
 - `[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.)
+- `[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.)
+- `[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.
 
@@ -145,7 +133,7 @@ The PR needs to meet the following code quality standards:
 - 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.
+- 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
 
@@ -153,18 +141,17 @@ Please keep the changes as concise as possible. For major architectural changes
 
 ### 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:
+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 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.
-- Note that not all CI checks will be executed due to limited computational resources. The reviewer will add `ready` label to the PR when the PR is ready to merge or a full CI run is needed.
 
 ## Additional Resources
 
-- [Design Documents](design_documents/index.md) - Architecture and design documentation
+- [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!
+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

@@ -1,6 +1,6 @@
 # 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.
+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
 

docs/design/api_design_template.md

@@ -1,6 +1,6 @@
-# API Design Template for vLLM-omni Modules
+# 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.
+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
 
@@ -44,7 +44,7 @@ This template provides a standardized structure for designing APIs for core, eng
 
 ### 1. Module Overview
 
-**Purpose**: The core module provides fundamental scheduling and caching functionality for vLLM-omni.
+**Purpose**: The core module provides fundamental scheduling and caching functionality for vLLM-Omni.
 
 **Responsibilities**:
 - Request scheduling and prioritization
@@ -333,4 +333,4 @@ print(f"Cache hit rate: {status['cache_hit_rate']:.2%}")
 6. Include error handling strategies
 7. Submit for review before implementation
 
-This template ensures consistency and completeness across all vLLM-omni modules.
+This template ensures consistency and completeness across all vLLM-Omni modules.

docs/design/architecture_overview.md

@@ -2,11 +2,11 @@
 
 # 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.
+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:
+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.  
@@ -57,12 +57,12 @@ According to analysis for current popular open-source models, most of them have
 
 # 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.  
+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.  
+* 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.
@@ -89,32 +89,6 @@ vLLM-omni will import core modules from existing vLLM framework and introduce se
 
 ![][image11]
 
-# Roadmap
-
-* AsyncLLM:  
-- [ ] Initializer: resource allocation for different stages (2025 H2)  
-- [ ] Hidden state Output Processor (2025 H2)  
-- [ ] Multimodal Output Processor (2025 H2)  
-- [ ] Interface for API, Gradio, ComfyUI  
-* OmniEngine:  
-- [ ] Basic Diffusion Engine core: Diffusion step manager, Cache manager (2025 H2)  
-- [ ] Pipeline of different Engine Cores (2025 H2)  
-- [ ] Communication between Engine Cores via AsyncLLM (2025 H2)  
-* Model support (Included but not limited to):  
-- [ ] DiT: Qwen-Image (2025 H2), Flux-Kontext  
-- [ ] Unified-MLLM: BAGEL (2025 H2), OmniGen2, Janus-Pro  
-- [ ] Extensions of AR+DiT: Qwen-omni (2025 H2)  
-* Optimization  
-- [ ] Integration of other DiT acceleration modules, such as xDiT, Cache-DiT  
-- [ ] Disaggregation for AR and DiT  
-- [ ] Integration with vLLM's existing KV cache for potential future hybrid models.  
-- [ ] Support for streaming multi-modal inputs and outputs
-
-# Contact
-
-For any questions or feedback regarding this design document, please contact  
-Han Gao: [hgaoaf@connect.ust.hk](mailto:hgaoaf@connect.ust.hk),  
-Hongsheng Liu: [vllm-project@gmail.com](mailto:vllm-project@gmail.com)
 
 [image1]: <data:image/png;base64,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 truncated
 

docs/design/index.md

@@ -1,12 +1,13 @@
 # Design Documents
 
-This section contains design documents and architecture specifications for vLLM-omni.
+This section contains design documents and architecture specifications for vLLM-Omni.
 
 ## Architecture Documents
 
-- [vLLM-omni Software Design Document](vllm_omni_design.md)
 - [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 Documentation](api_design_doc.md)
+- [vLLM-Omni API Design Documentation](api_design_doc.md)

docs/design/vllm_omni_design.md

@@ -2,7 +2,7 @@
 
 ## 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.
+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
 
@@ -36,7 +36,7 @@ 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]
+    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]
@@ -45,7 +45,7 @@ graph TD
     G --> H[Multi-stage Processing]
     E --> I[Standard vLLM Pipeline]
 
-    subgraph "vLLM-omni Components"
+    subgraph "vLLM-Omni Components"
         F
         G
         H
@@ -68,7 +68,7 @@ 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
+        # 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)

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/index.md

@@ -1,32 +0,0 @@
-# Welcome to vLLM-omni
-
-**vLLM-omni** extends vLLM capabilities to support multi-modality model inference and serving, particularly focusing on non-autoregressive structures and non-textual outputs. It aims to support:
-
-- **Multi-modal Models**: Text, image, video, audio, and sensor data processing
-- **Non-autoregressive Architectures**: Diffusion Transformers (DiT) and other parallel generation models
-- **Heterogeneous Outputs**: Beyond traditional text generation to multimodal outputs
-
-## <span class="twemoji">🎯</span> Key Features
-
-### Multi-Engine Support
-
-- **Autoregressive Engine**: Traditional text generation with enhanced KV-caching
-- **Diffusion Engine**: Support for DiT models and iterative generation
-- **Hybrid Engine**: Combined AR+DiT processing pipelines
-
-### Modality Processing
-
-- **Text**: Advanced tokenization and embedding generation
-- **Image**: Vision encoder integration (CLIP, etc.)
-- **Audio**: Speech processing and audio embedding
-
-## <span class="twemoji">🤖</span> Supported Models
-
-- Qwen-omni (Thinker-Talker-Codec structure)
-- HunyunaImage 3.0 (Ongoing)
-- Qwen-Image (Ongoing)
-
-## <span class="twemoji">📚</span> Documentation Navigation
-
-- To run open-source models on vLLM-Omni, we recommend starting with the [:material-code-tags: User Quide](user_guide/getting_started/quickstart.md)
-- To develop and contribute to vLLM-Omni, we recommend starting with the [:material-tools: Developer Guide](contributing/README.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/generate_examples.py

@@ -1,233 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM-omni project
-import itertools
-import logging
-from dataclasses import dataclass, field
-from pathlib import Path
-from typing import Literal
-
-import regex as re
-
-logger = logging.getLogger("mkdocs")
-
-ROOT_DIR = Path(__file__).parent.parent.parent.parent
-ROOT_DIR_RELATIVE = "../../../../.."
-EXAMPLE_DIR = ROOT_DIR / "examples"
-EXAMPLE_DOC_DIR = ROOT_DIR / "docs/user_guide/examples"
-
-
-def fix_case(text: str) -> str:
-    subs = {
-        "api": "API",
-        "cli": "CLI",
-        "cpu": "CPU",
-        "llm": "LLM",
-        "mae": "MAE",
-        "tpu": "TPU",
-        "gguf": "GGUF",
-        "lora": "LoRA",
-        "rlhf": "RLHF",
-        "vllm": "vLLM",
-        "omni": "Omni",
-        "openai": "OpenAI",
-        "lmcache": "LMCache",
-        "multilora": "MultiLoRA",
-        "mlpspeculator": "MLPSpeculator",
-        r"fp\d+": lambda x: x.group(0).upper(),  # e.g. fp16, fp32
-        r"int\d+": lambda x: x.group(0).upper(),  # e.g. int8, int16
-    }
-    for pattern, repl in subs.items():
-        text = re.sub(rf"\b{pattern}\b", repl, text, flags=re.IGNORECASE)
-    return text
-
-
-@dataclass
-class Example:
-    """
-    Example class for generating documentation content from a given path.
-
-    Attributes:
-        path (Path): The path to the main directory or file.
-        category (str): The category of the document.
-        main_file (Path): The main file in the directory.
-        other_files (list[Path]): list of other files in the directory.
-        title (str): The title of the document.
-
-    Methods:
-        __post_init__(): Initializes the main_file, other_files, and title attributes.
-        determine_main_file() -> Path: Determines the main file in the given path.
-        determine_other_files() -> list[Path]: Determines other files in the directory excluding the main file.
-        determine_title() -> str: Determines the title of the document.
-        generate() -> str: Generates the documentation content.
-    """  # noqa: E501
-
-    path: Path
-    category: str = None
-    main_file: Path = field(init=False)
-    other_files: list[Path] = field(init=False)
-    title: str = field(init=False)
-
-    def __post_init__(self):
-        self.main_file = self.determine_main_file()
-        self.other_files = self.determine_other_files()
-        self.title = self.determine_title()
-
-    @property
-    def is_code(self) -> bool:
-        return self.main_file.suffix != ".md"
-
-    def determine_main_file(self) -> Path:
-        """
-        Determines the main file in the given path.
-        If the path is a file, it returns the path itself. Otherwise, it searches
-        for Markdown files (*.md) in the directory and returns the first one found.
-        Returns:
-            Path: The main file path, either the original path if it's a file or the first
-            Markdown file found in the directory.
-        Raises:
-            IndexError: If no Markdown files are found in the directory.
-        """  # noqa: E501
-        return self.path if self.path.is_file() else list(self.path.glob("*.md")).pop()
-
-    def determine_other_files(self) -> list[Path]:
-        """
-        Determine other files in the directory excluding the main file.
-
-        This method checks if the given path is a file. If it is, it returns an empty list.
-        Otherwise, it recursively searches through the directory and returns a list of all
-        files that are not the main file.
-
-        Returns:
-            list[Path]: A list of Path objects representing the other files in the directory.
-        """  # noqa: E501
-        if self.path.is_file():
-            return []
-        # Text file extensions that can be safely included
-        text_extensions = {
-            ".py",
-            ".md",
-            ".sh",
-            ".yaml",
-            ".yml",
-            ".json",
-            ".txt",
-            ".toml",
-            ".cfg",
-            ".ini",
-        }
-        is_other_file = lambda file: (file.is_file() and file != self.main_file and file.suffix in text_extensions)  # noqa: E731
-        return [file for file in self.path.rglob("*") if is_other_file(file)]
-
-    def determine_title(self) -> str:
-        if not self.is_code:
-            # Specify encoding for building on Windows
-            with open(self.main_file, encoding="utf-8") as f:
-                first_line = f.readline().strip()
-            match = re.match(r"^#\s+(?P<title>.+)$", first_line)
-            if match:
-                return match.group("title")
-        return fix_case(self.path.stem.replace("_", " ").title())
-
-    def fix_relative_links(self, content: str) -> str:
-        """
-        Fix relative links in markdown content by converting them to gh-file
-        format.
-
-        Args:
-            content (str): The markdown content to process
-
-        Returns:
-            str: Content with relative links converted to gh-file format
-        """
-        # Regex to match markdown links [text](relative_path)
-        # This matches links that don't start with http, https, ftp, or #
-        link_pattern = r"\[([^\]]*)\]\((?!(?:https?|ftp)://|#)([^)]+)\)"
-
-        def replace_link(match):
-            link_text = match.group(1)
-            relative_path = match.group(2)
-
-            # Make relative to repo root
-            gh_file = (self.main_file.parent / relative_path).resolve()
-            gh_file = gh_file.relative_to(ROOT_DIR)
-
-            # Make GitHub URL
-            url = "https://github.com/vllm-project/vllm-omni/"
-            url += "tree/main" if self.path.is_dir() else "blob/main"
-            gh_url = f"{url}/{gh_file}"
-
-            return f"[{link_text}]({gh_url})"
-
-        return re.sub(link_pattern, replace_link, content)
-
-    def generate(self) -> str:
-        content = f"# {self.title}\n\n"
-        url = "https://github.com/vllm-project/vllm-omni/"
-        url += "tree/main" if self.path.is_dir() else "blob/main"
-        content += f"Source <{url}/{self.path.relative_to(ROOT_DIR)}>.\n\n"
-
-        # Use long code fence to avoid issues with
-        # included files containing code fences too
-        code_fence = "``````"
-
-        if self.is_code:
-            # Use relative path from ROOT_DIR for snippet inclusion
-            main_file_rel_path = self.main_file.relative_to(ROOT_DIR)
-            content += f'{code_fence}{self.main_file.suffix[1:]}\n--8<-- "{main_file_rel_path}"\n{code_fence}\n'
-        else:
-            with open(self.main_file, encoding="utf-8") as f:
-                # Skip the title from md snippets as it's been included above
-                main_content = f.readlines()[1:]
-            content += self.fix_relative_links("".join(main_content))
-        content += "\n"
-
-        if not self.other_files:
-            return content
-
-        content += "## Example materials\n\n"
-        for file in sorted(self.other_files):
-            # Use relative path from ROOT_DIR for snippet inclusion
-            file_rel_path = file.relative_to(ROOT_DIR)
-            content += f'??? abstract "{file.relative_to(self.path)}"\n'
-            if file.suffix != ".md":
-                content += f"    {code_fence}{file.suffix[1:]}\n"
-            content += f'    --8<-- "{file_rel_path}"\n'
-            if file.suffix != ".md":
-                content += f"    {code_fence}\n"
-
-        return content
-
-
-def on_startup(command: Literal["build", "gh-deploy", "serve"], dirty: bool):
-    logger.info("Generating example documentation")
-    logger.debug("Root directory: %s", ROOT_DIR.resolve())
-    logger.debug("Example directory: %s", EXAMPLE_DIR.resolve())
-    logger.debug("Example document directory: %s", EXAMPLE_DOC_DIR.resolve())
-
-    # Create the EXAMPLE_DOC_DIR if it doesn't exist
-    if not EXAMPLE_DOC_DIR.exists():
-        EXAMPLE_DOC_DIR.mkdir(parents=True)
-
-    categories = sorted(p for p in EXAMPLE_DIR.iterdir() if p.is_dir())
-
-    examples = []
-    glob_patterns = ["*.py", "*.md"]
-    # Find categorised examples
-    for category in categories:
-        globs = [category.glob(pattern) for pattern in glob_patterns]
-        for path in itertools.chain(*globs):
-            examples.append(Example(path, category.stem))
-        # Find examples in subdirectories
-        for path in category.glob("*/*.md"):
-            examples.append(Example(path.parent, category.stem))
-
-    # Generate the example documentation
-    for example in sorted(examples, key=lambda e: e.path.stem):
-        example_name = f"{example.path.stem}.md"
-        doc_path = EXAMPLE_DOC_DIR / example.category / example_name
-        if not doc_path.parent.exists():
-            doc_path.parent.mkdir(parents=True)
-        # Specify encoding for building on Windows
-        with open(doc_path, "w+", encoding="utf-8") as f:
-            f.write(example.generate())
-        logger.debug("Example generated: %s", doc_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

@@ -1,55 +1,218 @@
-/* Custom CSS for vLLM-omni documentation - based on vLLM's extra.css */
-
-/* Improve code block readability */
-.highlight {
-    border-radius: 4px;
+/* Warning for latest docs */
+.md-banner {
+    background-color: var(--md-warning-bg-color);
+    color: var(--md-warning-fg-color);
 }
 
-/* Style API documentation */
-.md-typeset .doc-contents {
-    margin-bottom: 1.5em;
+/* 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;
+    }
 }
 
-/* Improve table readability */
-.md-typeset table {
-    display: table;
-    width: 100%;
+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 (from vLLM) */
+/* 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 (from vLLM) */
+/* 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;
 }
 
-/* Navigation: Expand first level of API Reference by default, collapse deeper levels */
-/* Show direct children of API Reference section (vllm_omni's children: config, core, entrypoints, etc.) */
-.md-nav__item--nested[data-md-state] > .md-nav {
-    display: none;
+/* 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>');
 }
 
-/* Expand active items */
-.md-nav__item--nested.md-nav__item--active > .md-nav {
-    display: block;
+.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
 }
 
-/* For API Reference: show first level children (config, core, entrypoints, etc.) by default */
-.md-nav__item--section:has([href*="api/vllm_omni/index"]) > .md-nav > .md-nav__list > .md-nav__item--nested > .md-nav {
-    display: block;
+.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;
 }
 
-/* But keep deeper levels collapsed */
-.md-nav__item--section:has([href*="api/vllm_omni/index"]) > .md-nav > .md-nav__list > .md-nav__item--nested > .md-nav > .md-nav__list > .md-nav__item--nested > .md-nav {
-    display: none;
+.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);
 }
 
-/* Expand deeper levels only when active */
-.md-nav__item--section:has([href*="api/vllm_omni/index"]) > .md-nav > .md-nav__list > .md-nav__item--nested > .md-nav > .md-nav__list > .md-nav__item--nested.md-nav__item--active > .md-nav {
-    display: block;
+/* 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/index.md

@@ -1,8 +0,0 @@
-# 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.
-
-For detailed example documentation, check the [examples directory](https://github.com/vllm-project/vllm-omni/tree/main/examples) in the repository.

docs/user_guide/examples/offline_inference/qwen2_5_omni.md

@@ -1,64 +1,60 @@
-# Offline Example of vLLM-omni for Qwen2.5-omni
+# 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>.
+Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen2_5_omni>.
 
 
 ## 🛠️ Installation
 
-Please refer to [README.md](https://github.com/vllm-project/vllm-omni/tree/main/README.md)
+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"
+    ``````
 
-## 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:
+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 100
-```
-Get into the example folder
-```bash
-cd examples/offline_inference/qwen2_5_omni
+  --topk 10
 ```
 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
-```
-
-## Example materials
-
-??? abstract "end2end.py"
-    ``````py
-    --8<-- "examples\offline_inference\qwen2_5_omni\end2end.py"
-    ``````
-??? abstract "extract_prompts.py"
-    ``````py
-    --8<-- "examples\offline_inference\qwen2_5_omni\extract_prompts.py"
-    ``````
-??? abstract "processing_omni.py"
-    ``````py
-    --8<-- "examples\offline_inference\qwen2_5_omni\processing_omni.py"
-    ``````
 ??? abstract "run_multiple_prompts.sh"
     ``````sh
-    --8<-- "examples\offline_inference\qwen2_5_omni\run_multiple_prompts.sh"
-    ``````
-??? abstract "run_single_prompt.sh"
-    ``````sh
-    --8<-- "examples\offline_inference\qwen2_5_omni\run_single_prompt.sh"
-    ``````
-??? abstract "utils.py"
-    ``````py
-    --8<-- "examples\offline_inference\qwen2_5_omni\utils.py"
+    --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/offline_inference/qwen_2_5_omni.md

@@ -1,97 +0,0 @@
-# Offline Example of vLLM-omni for Qwen2.5-omni
-
-Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen_2_5_omni>.
-
-
-## 🛠️ Installation
-
-### Environment setup
-
-Use Docker to keep consistent basic environment (Optional, Recommended)
-```bash
-docker run --gpus all --ipc=host --network=host -v $source_dir:$container_dir --name $container_name -it nvcr.io/nvidia/pytorch:25.01-py3 bash
-```
-
-Set up basic uv environment
-```bash
-pip install uv
-uv venv --python 3.12 --seed
-source .venv/bin/activate
-```
-
-### Installation of vLLM (for users)
-
-Now we build it 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 (for developers)
-
-If you want to check or debug with source code of vLLM, install stable release version of vllm with 0.11.0 from source with pre-built wheel file.
-
-```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 .
-```
-
-### Verification for successful installation of vLLM
-Just run the command below. If no error, it demonstrates that the installation is successful.
-```bash
-python -c "import vllm._C"
-```
-
-### 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 .
-```
-
-## Run examples (Qwen2.5-omni)
-
-Get into the example folder
-```bash
-cd examples/offline_inference/qwen_2_5_omni
-```
-Then run
-```bash
-bash run.sh
-```
-The output audio is saved in ./output_audio
-
-## Example materials
-
-??? abstract "end2end.py"
-    ``````py
-    --8<-- "examples/offline_inference/qwen_2_5_omni/end2end.py"
-    ``````
-??? abstract "processing_omni.py"
-    ``````py
-    --8<-- "examples/offline_inference/qwen_2_5_omni/processing_omni.py"
-    ``````
-??? abstract "run.sh"
-    ``````sh
-    --8<-- "examples/offline_inference/qwen_2_5_omni/run.sh"
-    ``````
-??? abstract "utils.py"
-    ``````py
-    --8<-- "examples/offline_inference/qwen_2_5_omni/utils.py"
-    ``````

docs/user_guide/examples/online_serving/README.md

@@ -1,35 +0,0 @@
-# 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 [README.md](https://github.com/vllm-project/vllm-omni/blob/main/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
-```
-
-Send request via curl
-```bash
-bash run_curl_multimodal_generation.sh
-```

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/index.md

@@ -1,5 +0,0 @@
-# Online Serving
-
-## Available Examples
-
-Online serving is under active development. Coming soon.

docs/user_guide/examples/online_serving/openai_chat_completion_client_for_multimodal_generation.md

@@ -1,7 +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>.
+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"
+--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.

docs/user_guide/getting_started/quickstart.md

@@ -1,55 +0,0 @@
-# Quickstart
-
-This guide will help you get started with vLLM-omni.
-
-## Installation
-
-Use Docker to keep consistent basic environment (Optional, Recommended)
-
-```bash
-docker run --gpus all --ipc=host --network=host -v $source_dir:$container_dir --rm --name $container_name -it nvcr.io/nvidia/pytorch:25.01-py3 bash
-```
-
-Set up basic uv environment
-
-```bash
-pip install uv
-uv venv --python 3.12 --seed
-source .venv/bin/activate
-```
-
-Install vLLM
-
-```bash
-uv pip install vllm==0.11.0 --torch-backend=auto
-```
-
-Install vLLM-omni
-
-```bash
-git clone https://github.com/vllm-project/vllm-omni.git
-cd vllm_omni
-uv pip install -e .
-```
-
-## Run examples (Qwen2.5-omni)
-
-Get into the example folder
-
-```bash
-cd examples/offline_inference/qwen_2_5_omni
-```
-
-Then run.
-
-```bash
-bash run.sh
-```
-
-The output audio is saved in ./output_audio
-
-## Next Steps
-
-- Read the [architecture documentation](../../contributing/design_documents/vllm_omni_design.md)
-- Check out the [API reference](../../api/overview.md)
-- Explore the [examples](../examples/index.md)

examples/offline_inference/qwen2_5_omni/README.md

@@ -1,4 +1,4 @@
-# Offline Example of vLLM-omni for Qwen2.5-omni
+# Offline Example of vLLM-Omni for Qwen2.5-omni
 
 ## 🛠️ Installation
 
@@ -32,3 +32,11 @@ 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

@@ -1,166 +1,145 @@
-import argparse
+# 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
-import os as _os_env_toggle
-import random
+from typing import NamedTuple
 
-import numpy as np
 import soundfile as sf
-import torch
-from utils import make_omni_prompt
+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
 
-_os_env_toggle.environ["VLLM_USE_V1"] = "1"
-
 SEED = 42
-# Set all random seeds
-random.seed(SEED)
-np.random.seed(SEED)
-torch.manual_seed(SEED)
-torch.cuda.manual_seed(SEED)
-torch.cuda.manual_seed_all(SEED)
-
-# Make PyTorch deterministic
-torch.backends.cudnn.deterministic = True
-torch.backends.cudnn.benchmark = False
-
-# Set environment variables for deterministic behavior
-os.environ["PYTHONHASHSEED"] = str(SEED)
-os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
 
 
-def parse_args():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--model",
-        required=True,
-        help="Path to merged model directory (will be created if downloading).",
-    )
-    parser.add_argument("--thinker-model", type=str, default=None)
-    parser.add_argument("--talker-model", type=str, default=None)
-    parser.add_argument("--code2wav-model", type=str, default=None)
-    parser.add_argument(
-        "--hf-hub-id",
-        default="Qwen/Qwen2.5-Omni-7B",
-        help="Hugging Face repo id to download if needed.",
-    )
-    parser.add_argument("--hf-revision", default=None, help="Optional HF revision (branch/tag/commit).")
-    parser.add_argument("--prompts", nargs="+", default=None, help="Input text prompts.")
-    parser.add_argument("--voice-type", default="default", help="Voice type, e.g., m02, f030, default.")
-    parser.add_argument(
-        "--code2wav-dir",
-        default=None,
-        help="Path to code2wav folder (contains spk_dict.pt).",
-    )
-    parser.add_argument("--dit-ckpt", default=None, help="Path to DiT checkpoint file (e.g., dit.pt).")
-    parser.add_argument("--bigvgan-ckpt", default=None, help="Path to BigVGAN checkpoint file.")
-    parser.add_argument("--dtype", default="bfloat16", choices=["float16", "bfloat16", "float32"])
-    parser.add_argument("--max-model-len", type=int, default=32768)
-    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("--thinker-only", action="store_true")
-    parser.add_argument("--text-only", action="store_true")
-    parser.add_argument("--do-wave", action="store_true")
-    parser.add_argument(
-        "--prompt_type",
-        choices=[
-            "text",
-            "audio",
-            "audio-long",
-            "audio-long-chunks",
-            "audio-long-expand-chunks",
-            "image",
-            "video",
-            "video-frames",
-            "audio-in-video",
-            "audio-in-video-v2",
-            "audio-multi-round",
-            "badcase-vl",
-            "badcase-text",
-            "badcase-image-early-stop",
-            "badcase-two-audios",
-            "badcase-two-videos",
-            "badcase-multi-round",
-            "badcase-voice-type",
-            "badcase-voice-type-v2",
-            "badcase-audio-tower-1",
-            "badcase-audio-only",
-        ],
-        default="text",
-    )
-    parser.add_argument("--use-torchvision", action="store_true")
-    parser.add_argument("--tokenize", action="store_true")
-    parser.add_argument(
-        "--output-wav",
-        default="output.wav",
-        help="[Deprecated] Output wav directory (use --output-dir).",
-    )
-    parser.add_argument(
-        "--output-dir",
-        default="outputs",
-        help="Output directory to save text and wav files together.",
-    )
-    parser.add_argument(
-        "--thinker-hidden-states-dir",
-        default="thinker_hidden_states",
-        help="Path to thinker hidden states directory.",
-    )
-    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(
-        "--enable-stats",
-        action="store_true",
-        default=False,
-        help="Enable writing detailed statistics (default: disabled)",
-    )
-    parser.add_argument(
-        "--txt-prompts",
-        type=str,
-        default=None,
-        help="Path to a .txt file with one prompt per line (preferred).",
-    )
-    args = parser.parse_args()
-    return args
+class QueryResult(NamedTuple):
+    inputs: dict
+    limit_mm_per_prompt: dict[str, int]
 
 
-def main():
-    args = parse_args()
-    model_name = args.model
-    try:
-        # Preferred: load from txt file (one prompt per line)
-        if getattr(args, "txt_prompts", None) and args.prompt_type == "text":
-            with open(args.txt_prompts, encoding="utf-8") as f:
-                lines = [ln.strip() for ln in f.readlines()]
-            args.prompts = [ln for ln in lines if ln != ""]
-            print(f"[Info] Loaded {len(args.prompts)} prompts from {args.txt_prompts}")
-    except Exception as e:
-        print(f"[Error] Failed to load prompts: {e}")
-        raise
+# 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]()
 
-    if args.prompts is None:
-        raise ValueError("No prompts provided. Use --prompts ... or --txt-prompts <file.txt> (with --prompt_type text)")
     omni_llm = OmniLLM(
         model=model_name,
         log_stats=args.enable_stats,
@@ -205,8 +184,16 @@ def main():
         code2wav_sampling_params,
     ]
 
-    prompt = [make_omni_prompt(args, prompt) for prompt in args.prompts]
-    omni_outputs = omni_llm.generate(prompt, sampling_params_list)
+    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
@@ -217,7 +204,7 @@ def main():
                 request_id = int(output.request_id)
                 text_output = output.outputs[0].text
                 # Save aligned text file per request
-                prompt_text = args.prompts[request_id]
+                prompt_text = prompts[request_id]["prompt"]
                 out_txt = os.path.join(output_dir, f"{request_id:05d}.txt")
                 lines = []
                 lines.append("Prompt:\n")
@@ -239,5 +226,67 @@ def main():
                 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__":
-    main()
+    args = parse_args()
+    main(args)

examples/offline_inference/qwen2_5_omni/processing_omni.py

@@ -1,367 +0,0 @@
-from __future__ import annotations
-
-import base64
-import logging
-import math
-import os
-import time
-import warnings
-from functools import lru_cache
-from io import BytesIO
-
-import requests
-import torch
-import torchvision
-from packaging import version
-from PIL import Image
-from torchvision import io, transforms
-from torchvision.transforms import InterpolationMode
-
-logger = logging.getLogger(__name__)
-
-IMAGE_FACTOR = 28
-MIN_PIXELS = 4 * 28 * 28
-MAX_PIXELS = 16384 * 28 * 28
-MAX_RATIO = 200
-
-VIDEO_MIN_PIXELS = 128 * 28 * 28
-VIDEO_MAX_PIXELS = 768 * 28 * 28
-VIDEO_TOTAL_PIXELS = 24576 * 28 * 28
-FRAME_FACTOR = 2
-FPS = 2.0
-FPS_MIN_FRAMES = 4
-FPS_MAX_FRAMES = 768
-
-temporal_patch_size = 2
-spatial_patch_size = 14
-spatial_merge_size = 2
-
-
-def round_by_factor(number: int, factor: int) -> int:
-    """Returns the closest integer to 'number' that is divisible by 'factor'."""
-    return round(number / factor) * factor
-
-
-def ceil_by_factor(number: int, factor: int) -> int:
-    """Returns the smallest integer greater than or equal to 'number' that is
-    divisible by 'factor'."""
-    return math.ceil(number / factor) * factor
-
-
-def floor_by_factor(number: int, factor: int) -> int:
-    """Returns the largest integer less than or equal to 'number' that is
-    divisible by 'factor'."""
-    return math.floor(number / factor) * factor
-
-
-def smart_resize(
-    height: int,
-    width: int,
-    factor: int = IMAGE_FACTOR,
-    min_pixels: int = MIN_PIXELS,
-    max_pixels: int = MAX_PIXELS,
-) -> tuple[int, int]:
-    """
-    Rescales the image so that the following conditions are met:
-
-    1. Both dimensions (height and width) are divisible by 'factor'.
-
-    2. The total number of pixels is within the range ['min_pixels', 'max_pixels'].
-
-    3. The aspect ratio of the image is maintained as closely as possible.
-    """
-    if max(height, width) / min(height, width) > MAX_RATIO:
-        raise ValueError(
-            f"absolute aspect ratio must be smaller than {MAX_RATIO}, got {max(height, width) / min(height, width)}"
-        )
-    h_bar = max(factor, round_by_factor(height, factor))
-    w_bar = max(factor, round_by_factor(width, factor))
-    if h_bar * w_bar > max_pixels:
-        beta = math.sqrt((height * width) / max_pixels)
-        h_bar = floor_by_factor(height / beta, factor)
-        w_bar = floor_by_factor(width / beta, factor)
-    elif h_bar * w_bar < min_pixels:
-        beta = math.sqrt(min_pixels / (height * width))
-        h_bar = ceil_by_factor(height * beta, factor)
-        w_bar = ceil_by_factor(width * beta, factor)
-    return h_bar, w_bar
-
-
-def fetch_image(ele: dict[str, str | Image.Image], size_factor: int = IMAGE_FACTOR) -> Image.Image:
-    if "image" in ele:
-        image = ele["image"]
-    else:
-        image = ele["image_url"]
-    image_obj = None
-    if isinstance(image, Image.Image):
-        image_obj = image
-    elif image.startswith("http://") or image.startswith("https://"):
-        image_obj = Image.open(requests.get(image, stream=True).raw)
-    elif image.startswith("file://"):
-        image_obj = Image.open(image[7:])
-    elif image.startswith("data:image"):
-        if "base64," in image:
-            _, base64_data = image.split("base64,", 1)
-            data = base64.b64decode(base64_data)
-            image_obj = Image.open(BytesIO(data))
-    else:
-        image_obj = Image.open(image)
-    if image_obj is None:
-        raise ValueError(f"Unrecognized image input, support local path, http url, base64 and PIL.Image, got {image}")
-    image = image_obj.convert("RGB")
-    # resize
-    if "resized_height" in ele and "resized_width" in ele:
-        resized_height, resized_width = smart_resize(
-            ele["resized_height"],
-            ele["resized_width"],
-            factor=size_factor,
-        )
-    else:
-        width, height = image.size
-        min_pixels = ele.get("min_pixels", MIN_PIXELS)
-        max_pixels = ele.get("max_pixels", MAX_PIXELS)
-        resized_height, resized_width = smart_resize(
-            height,
-            width,
-            factor=size_factor,
-            min_pixels=min_pixels,
-            max_pixels=max_pixels,
-        )
-    image = image.resize((resized_width, resized_height))
-
-    return image
-
-
-def smart_nframes(
-    ele: dict,
-    total_frames: int,
-    video_fps: int | float,
-) -> int:
-    """calculate the number of frames for video used for model inputs.
-
-    Args:
-        ele (dict): a dict contains the configuration of video.
-            support either `fps` or `nframes`:
-                - nframes: the number of frames to extract for model inputs.
-                - fps: the fps to extract frames for model inputs.
-                    - min_frames: the minimum number of frames of the video,
-                      only used when fps is provided.
-                    - max_frames: the maximum number of frames of the video,
-                      only used when fps is provided.
-        total_frames (int): the original total number of frames of the video.
-        video_fps (int | float): the original fps of the video.
-
-    Raises:
-        ValueError: nframes should in interval [FRAME_FACTOR, total_frames].
-
-    Returns:
-        int: the number of frames for video used for model inputs.
-    """
-    assert not ("fps" in ele and "nframes" in ele), "Only accept either `fps` or `nframes`"
-    if "nframes" in ele:
-        nframes = round_by_factor(ele["nframes"], FRAME_FACTOR)
-    else:
-        fps = ele.get("fps", FPS)
-        min_frames = ceil_by_factor(ele.get("min_frames", FPS_MIN_FRAMES), FRAME_FACTOR)
-        max_frames = floor_by_factor(
-            ele.get("max_frames", min(FPS_MAX_FRAMES, total_frames)),
-            FRAME_FACTOR,
-        )
-        nframes = total_frames / video_fps * fps
-        nframes = min(max(nframes, min_frames), max_frames)
-        nframes = round_by_factor(nframes, FRAME_FACTOR)
-    if not (FRAME_FACTOR <= nframes and nframes <= total_frames):
-        raise ValueError(f"nframes should in interval [{FRAME_FACTOR}, {total_frames}], but got {nframes}.")
-    return nframes
-
-
-def _read_video_torchvision(
-    ele: dict,
-) -> torch.Tensor:
-    """read video using torchvision.io.read_video
-
-    Args:
-        ele (dict): a dict contains the configuration of video.
-        support keys:
-            - video: the path of video. support "file://", "http://",
-              "https://" and local path.
-            - video_start: the start time of video.
-            - video_end: the end time of video.
-    Returns:
-        torch.Tensor: the video tensor with shape (T, C, H, W).
-    """
-    video_path = ele["video"]
-    if version.parse(torchvision.__version__) < version.parse("0.19.0"):
-        if "http://" in video_path or "https://" in video_path:
-            warnings.warn("torchvision < 0.19.0 does not support http/https video path, please upgrade to 0.19.0.")
-        if "file://" in video_path:
-            video_path = video_path[7:]
-    st = time.time()
-    video, audio, info = io.read_video(
-        video_path,
-        start_pts=ele.get("video_start", 0.0),
-        end_pts=ele.get("video_end", None),
-        pts_unit="sec",
-        output_format="TCHW",
-    )
-    total_frames, video_fps = video.size(0), info["video_fps"]
-    total_duration = round(total_frames / video_fps, 3)
-    logger.info(
-        f"torchvision:  {video_path=}, {total_frames=}, {video_fps=}, duration={total_duration}s, time={time.time() - st:.3f}s"
-    )
-    nframes = smart_nframes(ele, total_frames=total_frames, video_fps=video_fps)
-    idx = torch.linspace(0, total_frames - 1, nframes).round().long()
-    video = video[idx]
-    return video, total_duration, nframes
-
-
-def is_decord_available() -> bool:
-    import importlib.util
-
-    return importlib.util.find_spec("decord") is not None
-
-
-def _read_video_decord(
-    ele: dict,
-) -> torch.Tensor:
-    """read video using decord.VideoReader
-
-    Args:
-        ele (dict): a dict contains the configuration of video.
-        support keys:
-            - video: the path of video. support "file://", "http://",
-              "https://" and local path.
-            - video_start: the start time of video.
-            - video_end: the end time of video.
-    Returns:
-        torch.Tensor: the video tensor with shape (T, C, H, W).
-    """
-    import decord
-
-    video_path = ele["video"]
-    st = time.time()
-    vr = decord.VideoReader(video_path)
-    # TODO: support start_pts and end_pts
-    if "video_start" in ele or "video_end" in ele:
-        raise NotImplementedError("not support start_pts and end_pts in decord for now.")
-    total_frames, video_fps = len(vr), vr.get_avg_fps()
-    total_duration = round(total_frames / video_fps, 3)
-    logger.info(f"decord:  {video_path=}, {total_frames=}, {video_fps=}, time={time.time() - st:.3f}s")
-    nframes = smart_nframes(ele, total_frames=total_frames, video_fps=video_fps)
-    idx = torch.linspace(0, total_frames - 1, nframes).round().long().tolist()
-    video = vr.get_batch(idx).asnumpy()
-    video = torch.tensor(video).permute(0, 3, 1, 2)  # Convert to TCHW format
-    return video, total_duration, nframes
-
-
-VIDEO_READER_BACKENDS = {
-    "decord": _read_video_decord,
-    "torchvision": _read_video_torchvision,
-}
-
-FORCE_QWENVL_VIDEO_READER = os.getenv("FORCE_QWENVL_VIDEO_READER", None)
-
-
-@lru_cache(maxsize=1)
-def get_video_reader_backend() -> str:
-    if FORCE_QWENVL_VIDEO_READER is not None:
-        video_reader_backend = FORCE_QWENVL_VIDEO_READER
-    elif is_decord_available():
-        video_reader_backend = "decord"
-    else:
-        video_reader_backend = "torchvision"
-    # print(f"qwen-vl-utils using {video_reader_backend} to read video.",
-    #       file=sys.stderr)
-    return video_reader_backend
-
-
-def fetch_video(ele: dict, image_factor: int = IMAGE_FACTOR) -> torch.Tensor | list[Image.Image]:
-    if isinstance(ele["video"], str):
-        video_reader_backend = get_video_reader_backend()
-        video, total_dur, nframes = VIDEO_READER_BACKENDS[video_reader_backend](ele)
-        frame_timestamps = total_dur * torch.arange(1, nframes + 1) / nframes
-        grid_timestamps = frame_timestamps[::FRAME_FACTOR]
-        second_per_grid = grid_timestamps[1] - grid_timestamps[0]
-        nframes, _, height, width = video.shape
-        min_pixels = ele.get("min_pixels", VIDEO_MIN_PIXELS)
-        total_pixels = ele.get("total_pixels", VIDEO_TOTAL_PIXELS)
-        max_pixels = max(
-            min(VIDEO_MAX_PIXELS, total_pixels / nframes * FRAME_FACTOR),
-            int(min_pixels * 1.05),
-        )
-        max_pixels = ele.get("max_pixels", max_pixels)
-        # min_pixels = (factor ** 2) * 52
-        # max_pixels = (factor ** 2) * min(768, (16384 / nframes * temporal_patch_size))
-        if "resized_height" in ele and "resized_width" in ele:
-            resized_height, resized_width = smart_resize(
-                ele["resized_height"],
-                ele["resized_width"],
-                factor=image_factor,
-            )
-        else:
-            resized_height, resized_width = smart_resize(
-                height,
-                width,
-                factor=image_factor,
-                min_pixels=min_pixels,
-                max_pixels=max_pixels,
-            )
-        video = transforms.functional.resize(
-            video,
-            [resized_height, resized_width],
-            interpolation=InterpolationMode.BICUBIC,
-            antialias=True,
-        ).float()
-        return video, total_dur, nframes, second_per_grid
-    else:
-        assert isinstance(ele["video"], (list, tuple))
-        process_info = ele.copy()
-        process_info.pop("type", None)
-        process_info.pop("video", None)
-        images = [
-            fetch_image({"image": video_element, **process_info}, size_factor=image_factor)
-            for video_element in ele["video"]
-        ]
-        nframes = ceil_by_factor(len(images), FRAME_FACTOR)
-        if len(images) < nframes:
-            images.extend([images[-1]] * (nframes - len(images)))
-        return images, None, None, None
-
-
-def extract_vision_info(conversations: list[dict] | list[list[dict]]) -> list[dict]:
-    vision_infos = []
-    if isinstance(conversations[0], dict):
-        conversations = [conversations]
-    for conversation in conversations:
-        for message in conversation:
-            if isinstance(message["content"], list):
-                for ele in message["content"]:
-                    if (
-                        "image" in ele
-                        or "image_url" in ele
-                        or "video" in ele
-                        or ele["type"] in ("image", "image_url", "video")
-                    ):
-                        vision_infos.append(ele)
-    return vision_infos
-
-
-def process_vision_info(
-    conversations: list[dict] | list[list[dict]],
-) -> tuple[list[Image.Image] | None, list[torch.Tensor | list[Image.Image]] | None]:
-    vision_infos = extract_vision_info(conversations)
-    # Read images or videos
-    image_inputs = []
-    video_inputs = []
-    for vision_info in vision_infos:
-        if "image" in vision_info or "image_url" in vision_info:
-            image_inputs.append(fetch_image(vision_info))
-        elif "video" in vision_info:
-            video_inputs.append(fetch_video(vision_info))
-        else:
-            raise ValueError("image, image_url or video should in content.")
-    if len(image_inputs) == 0:
-        image_inputs = None
-    if len(video_inputs) == 0:
-        video_inputs = None
-    return image_inputs, video_inputs

examples/offline_inference/qwen2_5_omni/run_multiple_prompts.sh

@@ -1,8 +1,3 @@
-python end2end.py --model Qwen/Qwen2.5-Omni-7B \
-                                 --voice-type "m02" \
-                                 --dit-ckpt none \
-                                 --bigvgan-ckpt none \
-                                 --output-wav output_audio \
-                                 --prompt_type text \
-                                 --init-sleep-seconds 0 \
-                                 --txt-prompts top100.txt
+python end2end.py --output-wav output_audio \
+                  --query-type text \
+                  --txt-prompts top10.txt

examples/offline_inference/qwen2_5_omni/run_single_prompt.sh

@@ -1,8 +1,2 @@
-python end2end.py --model Qwen/Qwen2.5-Omni-7B \
-                                 --voice-type "m02" \
-                                 --dit-ckpt none \
-                                 --bigvgan-ckpt none \
-                                 --output-wav output_audio \
-                                 --prompt_type text \
-                                 --init-sleep-seconds 0 \
-                                 --prompts "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
+python end2end.py --output-wav output_audio \
+                  --query-type use_audio_in_video

examples/offline_inference/qwen2_5_omni/utils.py

@@ -1,312 +0,0 @@
-import tempfile
-from typing import Optional, Union
-from urllib.request import urlopen
-
-import librosa
-import requests
-import resampy
-import soundfile as sf
-import torch
-import torchvision.io
-from processing_omni import fetch_image, fetch_video
-from transformers import AutoConfig, AutoProcessor
-from vllm.inputs import TextPrompt
-
-from vllm_omni.inputs.data import OmniTokensPrompt
-
-# Simple caches to avoid repeated heavy HF loads per prompt
-_PROCESSOR_CACHE: dict[str, "AutoProcessor"] = {}
-_CONFIG_CACHE: dict[str, "AutoConfig"] = {}
-
-
-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 resample_wav_to_16khz(input_filepath):
-    data, original_sample_rate = sf.read(input_filepath)
-    # Only use the first channel
-    if len(data.shape) > 1:
-        data = data[:, 0]
-    # resample to 16kHz
-    data_resampled = resampy.resample(data, sr_orig=original_sample_rate, sr_new=16000)
-    return data_resampled
-
-
-def fetch_and_read_video(args, video_url: str, fps=2):
-    def read_video_with_torchvision(video_file_name: str):
-        video, audio, info = torchvision.io.read_video(
-            video_file_name,
-            start_pts=0.0,
-            end_pts=None,
-            pts_unit="sec",
-            output_format="TCHW",
-        )
-
-        total_frames, video_fps = video.size(0), info["video_fps"]
-        total_duration = round(total_frames / video_fps, 3)
-        nframes = int(total_frames / video_fps * fps)
-
-        frame_timestamps = total_duration * torch.arange(1, nframes + 1) / nframes
-        grid_timestamps = frame_timestamps[::2]
-        second_per_grid = grid_timestamps[1] - grid_timestamps[0]
-
-        idx = torch.linspace(0, video.size(0) - 1, nframes).round().long()
-        video = video[idx]
-
-        if args.legacy_omni_video:
-            return [video, total_duration, nframes, second_per_grid.item()]
-        else:
-            return video
-
-    def read_video_with_transformers(video_file_name: Union[str, list[str]]):
-        video, total_duration, nframes, second_per_grid = fetch_video({"video": video_file_name})
-        if total_duration is None and nframes is None:
-            nframes = len(video)
-            total_duration = 0.5 * nframes
-            second_per_grid = 1.0
-        if args.legacy_omni_video:
-            return [video, total_duration, nframes, second_per_grid]
-        else:
-            return video
-
-    def read_video(video_file_name: str):
-        if args.use_torchvision:
-            return read_video_with_torchvision(video_file_name)
-        else:
-            return read_video_with_transformers(video_file_name)
-
-    if isinstance(video_url, str) and video_url.startswith("http"):
-        with tempfile.NamedTemporaryFile(delete=True) as temp_video_file:
-            resp = requests.get(video_url)
-            assert resp.status_code == requests.codes.ok, (
-                f"Failed to fetch video from {video_url}, status_code:{resp.status_code}, resp:{resp}"
-            )
-
-            temp_video_file.write(urlopen(video_url).read())
-            temp_video_file_path = temp_video_file.name
-            video_file_name = temp_video_file_path
-            return read_video(video_file_name)
-    else:
-        video_file_name = video_url
-        return read_video(video_file_name)
-
-
-def make_inputs_qwen2_omni(
-    args,
-    messages: list[dict[str, Union[str, list[dict[str, str]]]]],
-    use_audio_in_video: Optional[bool] = False,
-    tokenize: bool = False,
-) -> Union[OmniTokensPrompt, TextPrompt]:
-    from transformers import AutoConfig, AutoProcessor
-
-    # Cached processor/config to prevent per-prompt reloading and repeated warnings
-    if args.model not in _PROCESSOR_CACHE:
-        _PROCESSOR_CACHE[args.model] = AutoProcessor.from_pretrained(args.model)
-    processor = _PROCESSOR_CACHE[args.model]
-
-    config = _CONFIG_CACHE.get(args.model)
-    if config is None:
-        try:
-            config = AutoConfig.from_pretrained(args.model)
-        except Exception:
-            config = None
-        _CONFIG_CACHE[args.model] = config  # cache even if None to avoid retry storms
-
-    # Decide legacy flag only once based on config (default True if unknown)
-    if getattr(args, "legacy_omni_video", None) is None:
-        if config is not None and hasattr(config, "architectures"):
-            args.legacy_omni_video = "Qwen2_5OmniModel" not in config.architectures
-        else:
-            args.legacy_omni_video = True
-
-    audios, images, videos = [], [], []
-    for message in messages:
-        if not isinstance(message["content"], list):
-            message["content"] = [
-                {
-                    "type": "text",
-                    "text": message["content"],
-                }
-            ]
-        index, num_contents = 0, len(message["content"])
-        while index < num_contents:
-            ele = message["content"][index]
-            if "type" not in ele:
-                if "text" in ele:
-                    ele["type"] = "text"
-                elif "audio" in ele:
-                    ele["type"] = "audio"
-                elif "audio_url" in ele:
-                    ele["type"] = "audio_url"
-                elif "image" in ele:
-                    ele["type"] = "image"
-                elif "image_url" in ele:
-                    ele["type"] = "image_url"
-                elif "video" in ele:
-                    ele["type"] = "video"
-                elif "video_url" in ele:
-                    ele["type"] = "video_url"
-                else:
-                    raise ValueError(f"Unknown ele: {ele}")
-
-            if ele["type"] == "audio" or ele["type"] == "audio_url":
-                if "audio_url" in ele:
-                    audio_key = "audio_url"
-                    with tempfile.NamedTemporaryFile(delete=True) as temp_audio_file:
-                        temp_audio_file.write(urlopen(ele[audio_key]).read())
-                        temp_audio_file_path = temp_audio_file.name
-                        audios.append(resample_wav_to_16khz(temp_audio_file_path))
-                        ele["audio"] = temp_audio_file_path
-                elif "audio" in ele:
-                    audio_key = "audio"
-                    audios.append(resample_wav_to_16khz(ele[audio_key]))
-                else:
-                    raise ValueError(f"Unknown ele {ele}")
-            elif use_audio_in_video and (ele["type"] == "video" or ele["type"] == "video_url"):
-                # use video as audio as well
-                if "video_url" in ele:
-                    audio_key = "video_url"
-                    with tempfile.NamedTemporaryFile(delete=True) as temp_video_file:
-                        temp_video_file.write(urlopen(ele[audio_key]).read())
-                        temp_video_file_path = temp_video_file.name
-                        ele[audio_key] = temp_video_file_path
-                        audios.append(librosa.load(temp_video_file_path, sr=16000)[0])
-                        videos.append(fetch_and_read_video(args, temp_video_file_path))
-                        ele["video"] = temp_video_file_path
-                elif "video" in ele:
-                    audio_key = "video"
-                    audios.append(librosa.load(ele[audio_key], sr=16000)[0])
-                    videos.append(fetch_and_read_video(args, audio_key))
-                else:
-                    raise ValueError(f"Unknown ele {ele}")
-                # insert a audio after the video
-                message["content"].insert(
-                    index + 1,
-                    {
-                        "type": "audio",
-                        "audio": ele[audio_key],
-                    },
-                )
-                # no need to load the added audio again
-                index += 1
-            elif ele["type"] == "video" or ele["type"] == "video_url":
-                if "video_url" in ele:
-                    video_key = "video_url"
-                    with tempfile.NamedTemporaryFile(delete=True) as temp_video_file:
-                        temp_video_file.write(urlopen(ele["video_url"]).read())
-                        temp_video_file_path = temp_video_file.name
-                        videos.append(fetch_and_read_video(args, temp_video_file))
-                        ele["video"] = temp_video_file_path
-                else:
-                    video_key = "video"
-                    videos.append(fetch_and_read_video(args, ele[video_key]))
-            elif ele["type"] == "image" or ele["type"] == "image_url":
-                images.append(fetch_image(ele))
-
-            # move to the next content
-            index += 1
-
-    prompt = processor.apply_chat_template(
-        messages,
-        tokenize=tokenize,
-        add_generation_prompt=True,
-        add_vision_id=True,
-    )
-
-    audios = audios if len(audios) > 0 else None
-    images = images if len(images) > 0 else None
-    videos = videos if len(videos) > 0 else None
-
-    multi_modal_data = {}
-    if audios:
-        multi_modal_data["audio"] = audios
-    if images:
-        multi_modal_data["image"] = images
-    if videos:
-        multi_modal_data["video"] = videos
-
-    if isinstance(prompt, list) and isinstance(prompt[0], (list, str)):
-        prompt = prompt[0]
-
-    if tokenize:
-        return OmniTokensPrompt(
-            prompt_token_ids=prompt,
-            multi_modal_data=multi_modal_data,
-        )
-    else:
-        return TextPrompt(
-            prompt=prompt,
-            multi_modal_data=multi_modal_data,
-        )
-
-
-def make_text_prompt(args, prompt):
-    messages = [
-        get_system_prompt(),
-        {
-            "role": "user",
-            "content": [
-                {"type": "text", "text": prompt},
-            ],
-        },
-    ]
-
-    prompt = make_inputs_qwen2_omni(args, messages, tokenize=args.tokenize)
-    return prompt
-
-
-def make_audio_in_video_v2_prompt(args):
-    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": [
-                {
-                    "type": "video_url",
-                    "video_url": ("https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2.5-Omni/draw_small.mp4"),
-                },
-            ],
-        },
-    ]
-    prompt = make_inputs_qwen2_omni(
-        args,
-        messages,
-        use_audio_in_video=True,
-        tokenize=args.tokenize,
-    )
-    return prompt
-
-
-def make_omni_prompt(args, prompt=None) -> Union[OmniTokensPrompt, list[OmniTokensPrompt]]:
-    if args.prompt_type == "text":
-        prompt = make_text_prompt(args, prompt)
-    elif args.prompt_type == "audio-in-video-v2":
-        prompt = make_audio_in_video_v2_prompt(args)
-    else:
-        raise ValueError(f"Unsupported prompt type: {args.prompt_type}")
-    return prompt

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/README.md

@@ -1,32 +0,0 @@
-# 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
-```
-
-Send request via curl
-```bash
-bash run_curl_multimodal_generation.sh
-```

examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py

@@ -1,103 +0,0 @@
-import base64
-
-from openai import OpenAI
-
-# 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 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 run_text_to_audio(model: str) -> None:
-    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,
-    ]
-
-    chat_completion = client.chat.completions.create(
-        messages=[
-            get_system_prompt(),
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "text",
-                        "text": "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words.",
-                    },
-                ],
-            },
-        ],
-        model=model,
-        extra_body={
-            "sampling_params_list": sampling_params_list
-        },  # Optional, it has a default setting in stage_configs of the corresponding model.
-    )
-
-    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)
-
-
-if __name__ == "__main__":
-    run_text_to_audio("Qwen/Qwen2.5-Omni-7B")

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')"

examples/online_serving/run_curl_multimodal_generation.sh

@@ -1,75 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-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.
-
-output=$(curl -sS -X POST http://localhost:8091/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d @- <<EOF
-{
-  "model": "Qwen/Qwen2.5-Omni-7B",
-  "extra_body": {
-    "sampling_params_list": [
-      $thinker_sampling_params,
-      $talker_sampling_params,
-      $code2wav_sampling_params
-    ]
-  },
-  "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": [
-        {
-          "type": "text",
-          "text": "Explain the system architecture for a scalable audio generation pipeline. Answer in 15 words."
-        }
-      ]
-    }
-  ]
-}
-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

@@ -1,6 +1,6 @@
-site_name: vLLM-omni
+site_name: vLLM-Omni
 site_description: Multi-modality models inference and serving with non-autoregressive structures
-site_author: vLLM-omni Team
+site_author: vLLM-Omni Team
 site_url: https://vllm-project.github.io/vllm-omni/
 
 repo_name: vllm-project/vllm-omni
@@ -8,30 +8,36 @@ repo_url: https://github.com/vllm-project/vllm-omni
 edit_uri: edit/main/docs/
 
 # Copyright
-copyright: Copyright © 2025 vLLM-omni Team
+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:
-    # Light mode
-    - scheme: default
-      primary: blue
-      accent: blue
+    # 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
-    # Dark mode
-    - scheme: slate
-      primary: blue
-      accent: blue
+    # Palette toggle for dark mode
+    - media: "(prefers-color-scheme: dark)"
+      scheme: slate
+      primary: black
       toggle:
-        icon: material/brightness-4
-        name: Switch to light mode
+        icon: material/brightness-2
+        name: Switch to system preference
   features:
     - content.action.edit
     - content.code.copy
-    - content.tabs.link
     - navigation.instant
     - navigation.instant.progress
     - navigation.tracking
@@ -47,17 +53,18 @@ theme:
     - content.tabs
     - content.tooltips
     - toc.follow
-  icon:
-    repo: fontawesome/brands/github
+  custom_dir: docs/mkdocs/overrides
 
 hooks:
-  - docs/mkdocs/hooks/generate_examples.py
+  - docs/mkdocs/hooks/generate_api_readme.py
+  - docs/mkdocs/hooks/url_schemes.py
 
 # Plugins
 plugins:
   - meta
   - search
-  # - awesome-nav  # Uncomment if installed
+  - autorefs
+  - awesome-nav
   - glightbox
   - git-revision-date-localized:
       # exclude files
@@ -71,6 +78,8 @@ plugins:
       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"
@@ -81,33 +90,27 @@ plugins:
   - mkdocstrings:
       handlers:
         python:
-          paths: [vllm_omni]
           options:
             show_symbol_type_heading: true
             show_symbol_type_toc: true
-            show_source: true
-            show_root_heading: true
-            show_root_toc_entry: true
-            show_root_full_path: false
-            show_object_full_path: false
+            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
-            filters: []
-            summary:
-              modules: 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
+          - 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
 markdown_extensions:
   - attr_list
   - md_in_html
@@ -137,44 +140,12 @@ markdown_extensions:
   # For math rendering
   - pymdownx.arithmatex:
       generic: true
-  # For task lists
-  - pymdownx.tasklist:
-      custom_checkbox: true
-  # For tables
-  - tables
 
-# Navigation
-nav:
-  - Home: index.md
-  - User Guide:
-    - Getting Started:
-      - user_guide/getting_started/quickstart.md
-    - Examples:
-      - user_guide/examples/index.md
-      - Offline Inference:
-        - user_guide/examples/offline_inference/qwen_2_5_omni.md
-      - Online Serving:
-        - user_guide/examples/online_serving/index.md
-  - API Reference:
-      - Summary: api/overview.md
-  - Developer Guide:
-    - contributing/README.md
-    - Design Documents:
-      - contributing/design_documents/index.md
-      - contributing/design_documents/vllm_omni_design.md
-      - contributing/design_documents/architecture_overview.md
-      - contributing/design_documents/api_design_doc.md
-
-# Extra
-extra:
-  version:
-    provider: mike
-
-# Extra CSS
 extra_css:
   - mkdocs/stylesheets/extra.css
 
-# Makes the url format end in .html rather than act as a dir
-# So index.md generates as index.html and is available under URL /index.html
-# https://www.mkdocs.org/user-guide/configuration/#use_directory_urls
-use_directory_urls: false
+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

@@ -5,12 +5,12 @@ 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"
+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"}
+    {name = "vLLM-Omni Team", email = "vllm-project@gmail.com"}
 ]
 keywords = ["vllm", "multimodal", "diffusion", "transformer", "inference", "serving"]
 classifiers = [
@@ -30,10 +30,14 @@ classifiers = [
 dependencies = [
     # Core runtime dependencies (required for actual usage)
     "omegaconf>=2.3.0",
-    "decord>=0.6.0",
+    # "decord>=0.6.0", # install manually
     "librosa>=0.11.0",
     "resampy>=0.4.3",
-    # "vllm==0.11.0", # install manually
+    "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]
@@ -54,12 +58,13 @@ docs = [
     "mkdocs-material",
     "mkdocstrings-python",
     "mkdocs-gen-files",
-    # "mkdocs-awesome-nav",  # Not available on PyPI
+    "mkdocs-awesome-nav",
     "mkdocs-glightbox",
     "mkdocs-git-revision-date-localized-plugin",
     "mkdocs-minify-plugin",
     "regex",
     "pydantic",
+    "black>=23.0.0",
 ]
 
 
@@ -73,10 +78,14 @@ Documentation = "https://vllm-omni.readthedocs.io"
 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 = [
@@ -153,5 +162,9 @@ extend-ignore-identifiers-re = [
     ".*_thw",
     ".*thw",
     "ein",
-    ".*arange"
+    ".*arange",
+    ".*MoBA",
+    ".*temperal_downsample",
+    ".*nothink.*",
+    ".*NOTHINK.*"
 ]

scripts/README.md

@@ -1,108 +0,0 @@
-# vLLM-omni Testing Scripts
-
-This directory contains automated testing scripts for vLLM-omni serving functionality.
-
-## Scripts Overview
-
-### 1. `test_serving.sh` - Comprehensive Test Suite
-Full-featured testing script that performs complete validation of the vLLM-omni serving functionality.
-
-**Features:**
-- Model existence validation
-- Environment setup verification
-- Import functionality testing
-- Server startup and health checks
-- Text generation testing
-- Performance benchmarking
-- API client integration testing
-- Comprehensive logging and error handling
-
-**Usage:**
-```bash
-# Use default model and port
-./scripts/test_serving.sh
-
-# Use specific model and port
-./scripts/test_serving.sh ./models/Qwen3-0.6B 8000
-
-# Use HuggingFace model
-./scripts/test_serving.sh Qwen/Qwen3-0.6B 8001
-
-# Show help
-./scripts/test_serving.sh --help
-```
-
-**Test Coverage:**
-- ✅ Model loading and server startup
-- ✅ Health and info endpoints
-- ✅ Text generation functionality
-- ✅ Performance metrics
-- ✅ API client integration
-- ✅ All imports working correctly
-
-### 2. `quick_test.sh` - Fast Validation
-Lightweight script for quick validation after making changes.
-
-**Features:**
-- Fast import testing
-- Basic server startup
-- Health endpoint validation
-- Simple text generation test
-- Retry mechanism for reliability
-
-**Usage:**
-```bash
-# Use default port (8000)
-./scripts/quick_test.sh
-
-# Use specific port
-./scripts/quick_test.sh 8001
-```
-
-**Test Coverage:**
-- ✅ Import functionality
-- ✅ Server startup
-- ✅ Health endpoint
-- ✅ Basic text generation
-
-## Prerequisites
-
-1. **Conda Environment**: Ensure `vllm_omni` environment is activated
-2. **Model Available**: Qwen3-0.6B model should be available in `./models/Qwen3-0.6B/`
-3. **Dependencies**: All vLLM-omni dependencies should be installed
-
-## Environment Setup
-
-```bash
-# Activate conda environment
-conda activate vllm_omni
-
-# Verify installation
-python -c "import vllm_omni; print('vLLM-omni ready')"
-```
-
-## Model Setup
-
-The scripts expect the Qwen3-0.6B model to be available. You can:
-
-1. **Use local model**: Place model in `./models/Qwen3-0.6B/`
-2. **Use HuggingFace model**: Pass `Qwen/Qwen3-0.6B` as model path
-3. **Download model**: Use the `download_models.py` script
-
-```bash
-# Download model using the provided script
-python scripts/download_models.py
-```
-
-## Output and Logging
-
-### Test Results
-- **Success**: Green `[SUCCESS]` messages
-- **Info**: Blue `[INFO]` messages  
-- **Warnings**: Yellow `[WARNING]` messages
-- **Errors**: Red `[ERROR]` messages
-
-### Log Files
-- `server.log`: Server startup and runtime logs
-- `api_client_test.log`: API client test results
-- `simple_usage_test.log`: Simple usage test results

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

scripts/test_serving.sh

@@ -1,363 +0,0 @@
-#!/bin/bash
-
-# vLLM-omni Serving Functionality Test Script
-# This script tests the complete serving functionality of vLLM-omni
-# Usage: ./scripts/test_serving.sh [model_path] [port]
-
-set -e  # Exit on any error
-
-# Configuration
-MODEL_PATH=${1:-"./models/Qwen3-0.6B"}
-PORT=${2:-8000}
-HOST="localhost"
-TIMEOUT=30
-SERVER_STARTUP_WAIT=15
-
-# Colors for output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-BLUE='\033[0;34m'
-NC='\033[0m' # No Color
-
-# Logging functions
-log_info() {
-    echo -e "${BLUE}[INFO]${NC} $1"
-}
-
-log_success() {
-    echo -e "${GREEN}[SUCCESS]${NC} $1"
-}
-
-log_warning() {
-    echo -e "${YELLOW}[WARNING]${NC} $1"
-}
-
-log_error() {
-    echo -e "${RED}[ERROR]${NC} $1"
-}
-
-# Cleanup function
-cleanup() {
-    log_info "Cleaning up..."
-    if [ ! -z "$SERVER_PID" ]; then
-        log_info "Stopping server (PID: $SERVER_PID)..."
-        kill $SERVER_PID 2>/dev/null || true
-        wait $SERVER_PID 2>/dev/null || true
-    fi
-    # Kill any remaining vllm processes
-    pkill -f "vllm serve" 2>/dev/null || true
-    log_info "Cleanup completed"
-}
-
-# Set up trap for cleanup on exit
-trap cleanup EXIT
-
-# Check if model exists
-check_model() {
-    log_info "Checking if model exists at: $MODEL_PATH"
-    if [ ! -d "$MODEL_PATH" ]; then
-        log_error "Model directory not found: $MODEL_PATH"
-        log_info "Available models:"
-        ls -la models/ 2>/dev/null || log_warning "No models directory found"
-        exit 1
-    fi
-    log_success "Model found: $MODEL_PATH"
-}
-
-# Check if conda environment is activated
-check_environment() {
-    log_info "Checking conda environment..."
-    if [[ "$CONDA_DEFAULT_ENV" != "vllm_omni" ]]; then
-        log_warning "vllm_omni conda environment not activated"
-        log_info "Attempting to activate vllm_omni environment..."
-        source $(conda info --base)/etc/profile.d/conda.sh
-        conda activate vllm_omni
-        if [[ "$CONDA_DEFAULT_ENV" != "vllm_omni" ]]; then
-            log_error "Failed to activate vllm_omni environment"
-            exit 1
-        fi
-    fi
-    log_success "Environment check passed: $CONDA_DEFAULT_ENV"
-}
-
-# Test import functionality
-test_imports() {
-    log_info "Testing imports..."
-    python -c "
-import vllm_omni
-from vllm_omni.entrypoints.omni_llm import OmniLLM, AsyncOmniLLM
-print('✅ All imports successful')
-" || {
-        log_error "Import test failed"
-        exit 1
-    }
-    log_success "Import test passed"
-}
-
-# Start the server
-start_server() {
-    log_info "Starting vLLM-omni server on port $PORT..."
-    log_info "Command: vllm serve $MODEL_PATH --omni --port $PORT"
-
-    # Start server in background
-    vllm serve "$MODEL_PATH" --omni --port "$PORT" > server.log 2>&1 &
-    SERVER_PID=$!
-
-    log_info "Server started with PID: $SERVER_PID"
-    log_info "Waiting $SERVER_STARTUP_WAIT seconds for server to initialize..."
-    sleep $SERVER_STARTUP_WAIT
-
-    # Check if server is still running
-    if ! kill -0 $SERVER_PID 2>/dev/null; then
-        log_error "Server failed to start"
-        log_info "Server logs:"
-        cat server.log
-        exit 1
-    fi
-
-    log_success "Server appears to be running"
-}
-
-# Test health endpoint
-test_health() {
-    log_info "Testing health endpoint..."
-    for attempt in {1..10}; do
-        local response=$(curl -s -w "%{http_code}" http://$HOST:$PORT/health 2>/dev/null || echo "000")
-        local http_code="${response: -3}"
-        local body="${response%???}"
-
-        if [ "$http_code" = "200" ]; then
-            log_success "Health check passed: $body"
-            return 0
-        fi
-
-        log_warning "Health check attempt $attempt failed (HTTP $http_code). Retrying in 3s..."
-        sleep 3
-    done
-
-    log_error "Health check failed after multiple attempts"
-    return 1
-}
-
-# Test info endpoint
-test_info() {
-    log_info "Testing info endpoint..."
-    local response=$(curl -s -w "%{http_code}" http://$HOST:$PORT/info 2>/dev/null || echo "000")
-    local http_code="${response: -3}"
-    local body="${response%???}"
-
-    if [ "$http_code" = "200" ]; then
-        log_success "Info endpoint working"
-        echo "$body" | python -m json.tool 2>/dev/null || log_warning "Info response not valid JSON"
-    else
-        log_error "Info endpoint failed (HTTP $http_code): $body"
-        return 1
-    fi
-}
-
-# Test text generation
-test_generation() {
-    log_info "Testing text generation..."
-
-    # Test 1: Simple generation
-    local response=$(curl -s -X POST http://$HOST:$PORT/generate \
-        -H "Content-Type: application/json" \
-        -d '{"prompts": ["Test the server functionality"], "max_tokens": 20, "temperature": 0.7}' \
-        2>/dev/null || echo "{}")
-
-    if echo "$response" | python -c "import sys, json; data=json.load(sys.stdin); exit(0 if 'outputs' in data and len(data['outputs']) > 0 else 1)" 2>/dev/null; then
-        log_success "Text generation test passed"
-        echo "$response" | python -c "import sys, json; data=json.load(sys.stdin); print('Generated text:', data['outputs'][0]['text'][:100] + '...' if len(data['outputs'][0]['text']) > 100 else data['outputs'][0]['text'])" 2>/dev/null
-    else
-        log_error "Text generation test failed"
-        echo "Response: $response"
-        return 1
-    fi
-}
-
-# Test API client example
-test_api_client() {
-    log_info "Testing API client example..."
-    if [ -f "examples/basic/api_client.py" ]; then
-        python examples/basic/api_client.py > api_client_test.log 2>&1 || {
-            log_error "API client test failed"
-            log_info "API client logs:"
-            cat api_client_test.log
-            return 1
-        }
-        log_success "API client test passed"
-    else
-        log_warning "API client example not found, skipping"
-    fi
-}
-
-# Test AR → DiT pipeline example
-test_ar_dit_pipeline() {
-    log_info "Testing AR → DiT diffusers pipeline example..."
-
-    if ! python - 2>/dev/null <<'PY'
-import importlib
-for pkg in ("torch", "diffusers"):
-    if importlib.util.find_spec(pkg) is None:
-        raise ImportError(pkg)
-PY
-    then
-        log_warning "Required packages for diffusers pipeline not found, skipping AR → DiT test"
-        return 0
-    fi
-
-    if [ ! -f "examples/omni/ar_dit_diffusers.py" ]; then
-        log_warning "AR → DiT example not found, skipping"
-        return 0
-    fi
-
-    local output_path="logs/ar_dit_pipeline.png"
-    local log_path="logs/ar_dit_pipeline.log"
-    rm -f "$output_path"
-
-    local cmd=(python examples/omni/ar_dit_diffusers.py
-        --prompt "Smoke test prompt of a lighthouse"
-        --seed 0
-        --output "$output_path")
-
-    local env_prefix=()
-    if [ -n "$AR_MODEL" ]; then
-        env_prefix+=("AR_MODEL=$AR_MODEL")
-    fi
-    if [ -n "$DIT_MODEL" ]; then
-        env_prefix+=("DIT_MODEL=$DIT_MODEL")
-    fi
-
-    if ! env "${env_prefix[@]}" "${cmd[@]}" >"$log_path" 2>&1; then
-        log_error "AR → DiT pipeline test failed"
-        log_info "AR → DiT logs:"
-        cat "$log_path"
-        return 1
-    fi
-
-    if [ ! -f "$output_path" ]; then
-        log_error "AR → DiT pipeline did not produce an image"
-        log_info "AR → DiT logs:"
-        cat "$log_path"
-        return 1
-    fi
-
-    log_success "AR → DiT pipeline test passed (output: $output_path)"
-}
-
-# Test simple usage example
-test_simple_usage() {
-    log_info "Testing simple usage example..."
-    if [ -f "examples/basic/simple_usage.py" ]; then
-        # Run with timeout to avoid hanging
-        timeout 60 python examples/basic/simple_usage.py > simple_usage_test.log 2>&1 || {
-            local exit_code=$?
-            if [ $exit_code -eq 124 ]; then
-                log_warning "Simple usage test timed out (60s), but this might be normal for model loading"
-            else
-                log_error "Simple usage test failed (exit code: $exit_code)"
-                log_info "Simple usage logs:"
-                cat simple_usage_test.log
-                return 1
-            fi
-        }
-        log_success "Simple usage test passed"
-    else
-        log_warning "Simple usage example not found, skipping"
-    fi
-}
-
-# Performance test
-test_performance() {
-    log_info "Running performance test..."
-
-    local start_time=$(date +%s)
-    local response=$(curl -s -X POST http://$HOST:$PORT/generate \
-        -H "Content-Type: application/json" \
-        -d '{"prompts": ["Performance test prompt"], "max_tokens": 50, "temperature": 0.7}' \
-        2>/dev/null)
-    local end_time=$(date +%s)
-    local duration=$((end_time - start_time))
-
-    if echo "$response" | python -c "import sys, json; data=json.load(sys.stdin); exit(0 if 'outputs' in data else 1)" 2>/dev/null; then
-        log_success "Performance test passed (${duration}s response time)"
-    else
-        log_error "Performance test failed"
-        return 1
-    fi
-}
-
-# Main test function
-run_tests() {
-    log_info "Starting vLLM-omni serving functionality tests..."
-    log_info "Model: $MODEL_PATH"
-    log_info "Port: $PORT"
-    log_info "Host: $HOST"
-    echo "=========================================="
-
-    # Run all tests
-    check_model
-    check_environment
-    test_imports
-    start_server
-
-    # Wait a bit more for server to be fully ready
-    sleep 5
-
-    test_health
-    test_info
-    test_generation
-    test_performance
-    test_api_client
-    test_ar_dit_pipeline
-
-    # Note: Simple usage test is commented out as it starts its own server
-    # test_simple_usage
-
-    log_success "All tests completed successfully!"
-    echo "=========================================="
-    log_info "Test Summary:"
-    log_info "✅ Model loading and server startup"
-    log_info "✅ Health and info endpoints"
-    log_info "✅ Text generation functionality"
-    log_info "✅ Performance metrics"
-    log_info "✅ API client integration"
-    log_info "✅ AR → DiT diffusers pipeline"
-    log_info "✅ All imports working correctly"
-}
-
-# Show usage
-show_usage() {
-    echo "Usage: $0 [model_path] [port]"
-    echo ""
-    echo "Arguments:"
-    echo "  model_path    Path to the model directory (default: ./models/Qwen3-0.6B)"
-    echo "  port          Port to run the server on (default: 8000)"
-    echo ""
-    echo "Examples:"
-    echo "  $0                                    # Use default model and port"
-    echo "  $0 ./models/Qwen3-0.6B 8001          # Use specific model and port"
-    echo "  $0 Qwen/Qwen3-0.6B 8000              # Use HuggingFace model"
-}
-
-# Main execution
-main() {
-    # Check for help flag
-    if [[ "$1" == "-h" || "$1" == "--help" ]]; then
-        show_usage
-        exit 0
-    fi
-
-    # Change to script directory
-    cd "$(dirname "$0")/.."
-
-    # Create logs directory if it doesn't exist
-    mkdir -p logs
-
-    # Run tests
-    run_tests
-}
-
-# Run main function with all arguments
-main "$@"

vllm_omni/__init__.py

@@ -1,5 +1,5 @@
 """
-vLLM-omni: Multi-modality models inference and serving with
+vLLM-Omni: Multi-modality models inference and serving with
 non-autoregressive structures.
 
 This package extends vLLM beyond traditional text-based, autoregressive
@@ -19,20 +19,21 @@ except ModuleNotFoundError as exc:  # pragma: no cover - optional dependency
         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_llm import OmniLLM
+# from .entrypoints.omni import Omni
 
-# from .version import __version__, __version_tuple__  # isort:skip
+from .version import __version__, __version_tuple__  # isort:skip
 
 
 __all__ = [
-    # "__version__",  # TODO: add __version__ back when it is implemented
-    # "__version_tuple__",  # TODO: add __version_tuple__ back when it is implemented
+    "__version__",
+    "__version_tuple__",
     # Main components
-    "OmniLLM",
+    # "Omni",
     "AsyncOmniLLM",
     # Configuration
     "OmniModelConfig",

vllm_omni/config/__init__.py

@@ -1,5 +1,5 @@
 """
-Configuration module for vLLM-omni.
+Configuration module for vLLM-Omni.
 """
 
 from vllm_omni.config.model import OmniModelConfig

vllm_omni/config/model.py

@@ -1,21 +1,72 @@
-from typing import Optional
+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."""
+    """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):
@@ -24,3 +75,313 @@ class OmniModelConfig(ModelConfig):
     @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

@@ -1,5 +1,5 @@
 """
-DiT Cache Manager for vLLM-omni.
+DiT Cache Manager for vLLM-Omni.
 
 This module provides caching functionality for DiT (Diffusion Transformer) models
 to optimize inference performance and memory usage.

vllm_omni/core/sched/__init__.py

@@ -1,5 +1,5 @@
 """
-Scheduling components for vLLM-omni.
+Scheduling components for vLLM-Omni.
 """
 
 from .diffusion_scheduler import DiffusionScheduler

vllm_omni/core/sched/diffusion_scheduler.py

@@ -14,12 +14,22 @@ 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:
-        """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.
+        """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
@@ -239,6 +249,12 @@ class DiffusionScheduler(OmniScheduler):
 
             # 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(
@@ -248,7 +264,7 @@ class DiffusionScheduler(OmniScheduler):
                         finish_reason=request.get_finished_reason(),
                         new_logprobs=new_logprobs,
                         new_prompt_logprobs_tensors=prompt_logprobs_tensors,
-                        pooling_output=pooler_output,
+                        pooling_output=pooling_output_dict,
                         stop_reason=request.stop_reason,
                         events=request.take_events(),
                         kv_transfer_params=kv_transfer_params,