Merge origin/main into codex-vae-spatial-decode-auto

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>

.buildkite/cuda/test-nightly.yml

@@ -345,7 +345,7 @@ steps:
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
           - export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py --test-config-file tests/dfx/perf/tests/test_qwen_image_vllm_omni.json
@@ -361,7 +361,7 @@ steps:
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
           - export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py --test-config-file tests/dfx/perf/tests/test_qwen_image_edit_vllm_omni.json
@@ -377,7 +377,7 @@ steps:
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
           - export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py --test-config-file tests/dfx/perf/tests/test_qwen_image_edit_2511_vllm_omni.json
@@ -393,7 +393,7 @@ steps:
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
           - export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py --test-config-file tests/dfx/perf/tests/test_qwen_image_layered_vllm_omni.json
@@ -409,7 +409,7 @@ steps:
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
           - export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py --test-config-file tests/dfx/perf/tests/test_bagel_vllm_omni.json

.buildkite/cuda/test-ready.yml

@@ -227,6 +227,18 @@ steps:
             "
         mirror_hardwares: h100_1
 
+      - label: "Omni · MiniCPM-o 4.5 Perf Test"
+        timeout_in_minutes: 60
+        source_file_dependencies:
+          - tests/dfx/perf/tests/test_minicpmo_4_5_ready.json
+          - vllm_omni/deploy/minicpmo_4_5.yaml
+          - vllm_omni/model_executor/models/minicpmo_4_5/
+          - vllm_omni/model_executor/stage_input_processors/minicpmo_4_5_omni.py
+        commands:
+          - export BENCHMARK_DIR=tests/dfx/perf/results
+          - pytest -s -v tests/dfx/perf/scripts/run_benchmark.py --test-config-file tests/dfx/perf/tests/test_minicpmo_4_5_ready.json -m 'core_model and cuda'
+        mirror_hardwares: h100_1
+
       - label: "TTS · VoxCPM2 Test"
         source_file_dependencies:
           - tests/e2e/online_serving/test_voxcpm2_tts.py

.buildkite/npu/test-npu-nightly.yml

@@ -115,7 +115,7 @@ steps:
           DIFFUSION_ATTENTION_BACKEND: TORCH_SDPA
         commands:
           - export DIFFUSION_BENCHMARK_DIR=tests/dfx/perf/results
-          - export CACHE_DIT_VERSION=1.3.0
+          - export CACHE_DIT_VERSION=1.5.0
           - |
             set +e
             pytest -s -v tests/dfx/perf/scripts/run_diffusion_benchmark.py -m npu --test-config-file tests/dfx/perf/tests/test_qwen_image_edit_2511_vllm_omni.json

.buildkite/npu/test-npu-ready.yml

@@ -17,6 +17,15 @@ steps:
           VLLM_ALLOW_LONG_MAX_MODEL_LEN: "1"
         commands:
           - pytest -s -v tests/e2e/online_serving/test_minicpmo_4_5_duplex.py -m 'core_model and npu and A3 and omni' --run-level 'core_model'
+      - label: "Omni · MiniCPM-o 4.5 Perf Test"
+        timeout_in_minutes: 60
+        mirror_hardwares: a3_npu_2
+        env:
+          VLLM_WORKER_MULTIPROC_METHOD: spawn
+          VLLM_ALLOW_LONG_MAX_MODEL_LEN: "1"
+        commands:
+          - export BENCHMARK_DIR=tests/dfx/perf/results
+          - pytest -s -v tests/dfx/perf/scripts/run_benchmark.py --test-config-file tests/dfx/perf/tests/test_minicpmo_4_5_ready.json -m 'core_model and npu and A3 and omni'
 
   - group: ":card_index_dividers: TTS Model Test"
     key: ready-tts-group

.buildkite/release/release-pipeline.yml

@@ -40,9 +40,11 @@ steps:
     commands:
       - "bash .buildkite/release/scripts/publish-release-wheels-pypi.sh"
 
+  # Auto-skip when NIGHTLY=1 (scheduled); otherwise require manual unblock.
   - block: "Unblock to build release Docker images"
     depends_on: ~
     key: block-build-release-images
+    if: build.env("NIGHTLY") != "1"
 
   - group: "Build release Docker images"
     key: "build-release-images"
@@ -63,7 +65,7 @@ steps:
               --progress plain \
               -f docker/Dockerfile.cuda .
           - "docker push public.ecr.aws/q9t5s3a7/vllm-omni-release-repo:$BUILDKITE_COMMIT-$(uname -m)"
-          # re-tag to default image tag and push, just in case arm64 build fails
+          # Fallback default tag (no arch suffix) if arm64 build fails.
           - "docker tag public.ecr.aws/q9t5s3a7/vllm-omni-release-repo:$BUILDKITE_COMMIT-$(uname -m) public.ecr.aws/q9t5s3a7/vllm-omni-release-repo:$BUILDKITE_COMMIT"
           - "docker push public.ecr.aws/q9t5s3a7/vllm-omni-release-repo:$BUILDKITE_COMMIT"
         env:
@@ -86,6 +88,26 @@ steps:
         env:
           DOCKER_BUILDKIT: "1"
 
+  # NIGHTLY=1 only: promote ECR images to DockerHub nightly tags, then prune.
+  - label: "Publish nightly multi-arch image to DockerHub"
+    depends_on:
+      - build-release-image-x86
+      - build-release-image-arm64
+    if: build.env("NIGHTLY") == "1"
+    key: publish-nightly-images-dockerhub
+    agents:
+      queue: small_cpu_queue_release
+    commands:
+      - "bash .buildkite/release/scripts/publish-release-images.sh nightly"
+      - "bash .buildkite/release/scripts/cleanup-nightly-builds.sh"
+    plugins:
+      - docker-login#v3.0.0:
+          username: vllmbot
+          password-env: DOCKERHUB_TOKEN
+    env:
+      DOCKER_BUILDKIT: "1"
+      DOCKERHUB_USERNAME: "vllmbot"
+
   - block: "Publish release images to DockerHub"
     key: block-publish-release-images
     depends_on:

.buildkite/release/scripts/cleanup-nightly-builds.sh

@@ -0,0 +1,147 @@
+#!/bin/bash
+# SPDX-License-Identifier: Apache-2.0
+#
+# Clean up old nightly builds from DockerHub (vllm/vllm-omni), keeping the
+# newest 14 tags with the nightly- prefix.
+
+set -ex
+
+TAG_PREFIX="nightly-"
+REPO="vllm/vllm-omni"
+REPO_API_URL="https://hub.docker.com/v2/repositories/${REPO}/tags"
+
+echo "Cleaning up tags with prefix: $TAG_PREFIX in repository: $REPO"
+
+if [ -z "$DOCKERHUB_TOKEN" ]; then
+  echo "Error: DOCKERHUB_TOKEN environment variable is not set"
+  exit 1
+fi
+
+if [ -z "$DOCKERHUB_USERNAME" ]; then
+  echo "Error: DOCKERHUB_USERNAME environment variable is not set"
+  exit 1
+fi
+
+echo "Getting DockerHub bearer token..."
+set +x
+BEARER_TOKEN=$(curl -s -X POST \
+  -H "Content-Type: application/json" \
+  -d "{\"username\": \"$DOCKERHUB_USERNAME\", \"password\": \"$DOCKERHUB_TOKEN\"}" \
+  "https://hub.docker.com/v2/users/login" | jq -r '.token')
+set -x
+
+if [ -z "$BEARER_TOKEN" ] || [ "$BEARER_TOKEN" = "null" ]; then
+  echo "Error: Failed to get DockerHub bearer token"
+  exit 1
+fi
+
+get_all_tags() {
+  local page=1
+  local all_tags=""
+  # Hub `name` is a partial match (not a strict prefix). Use it to narrow the
+  # list, then enforce startswith() client-side. Exhaust API pages; do not
+  # stop merely because a page has no matching tags after filtering.
+  local name_query
+  name_query=$(printf '%s' "$TAG_PREFIX" | jq -sRr @uri)
+
+  while true; do
+    set +x
+    local response
+    response=$(curl -s -H "Authorization: Bearer $BEARER_TOKEN" \
+      "$REPO_API_URL?page=$page&page_size=100&name=${name_query}")
+    set -x
+
+    local page_count
+    page_count=$(echo "$response" | jq -r '.results | length')
+    if [ -z "$page_count" ] || [ "$page_count" = "0" ] || [ "$page_count" = "null" ]; then
+      break
+    fi
+
+    local tags
+    tags=$(echo "$response" | jq -r --arg prefix "$TAG_PREFIX" \
+      '.results[] | select(.name | startswith($prefix)) | "\(.last_updated)|\(.name)"')
+
+    if [ -n "$tags" ]; then
+      all_tags="$all_tags$tags"$'\n'
+    fi
+    page=$((page + 1))
+  done
+
+  echo "$all_tags" | sort -r | cut -d'|' -f2
+}
+
+delete_tag() {
+  local tag_name="$1"
+  echo "Deleting tag: $tag_name"
+
+  local delete_url="https://hub.docker.com/v2/repositories/${REPO}/tags/$tag_name"
+  set +x
+  local response http_code body
+  response=$(curl -s -w "\n%{http_code}" -X DELETE \
+    -H "Authorization: Bearer $BEARER_TOKEN" "$delete_url")
+  set -x
+  http_code=$(printf '%s\n' "$response" | tail -n1)
+  body=$(printf '%s\n' "$response" | sed '$d')
+
+  if [ "$http_code" -ge 200 ] && [ "$http_code" -lt 300 ]; then
+    echo "Successfully deleted tag: $tag_name (HTTP ${http_code})"
+    return 0
+  fi
+
+  local detail=""
+  if echo "$body" | jq -e '.detail' > /dev/null 2>&1; then
+    detail=$(echo "$body" | jq -r '.detail')
+  elif [ -n "$body" ]; then
+    detail="$body"
+  else
+    detail="(empty response body)"
+  fi
+  echo "Error: Failed to delete tag $tag_name (HTTP ${http_code}): ${detail}"
+  return 1
+}
+
+echo "Fetching all tags from DockerHub..."
+all_tags=$(get_all_tags)
+
+if [ -z "$all_tags" ]; then
+  echo "No tags found to clean up"
+  exit 0
+fi
+
+total_tags=$(echo "$all_tags" | wc -l)
+echo "Found $total_tags tags"
+
+tags_to_keep=14
+tags_to_delete=$((total_tags - tags_to_keep))
+
+if [ $tags_to_delete -le 0 ]; then
+  echo "No tags need to be deleted (only $total_tags tags found, keeping $tags_to_keep)"
+  exit 0
+fi
+
+echo "Will delete $tags_to_delete old tags, keeping the newest $tags_to_keep"
+
+tags_to_delete_list=$(echo "$all_tags" | tail -n +$((tags_to_keep + 1)))
+
+if [ -z "$tags_to_delete_list" ]; then
+  echo "No tags to delete"
+  exit 0
+fi
+
+echo "Deleting old tags..."
+delete_failures=0
+while IFS= read -r tag; do
+  if [ -n "$tag" ]; then
+    if ! delete_tag "$tag"; then
+      delete_failures=$((delete_failures + 1))
+    fi
+    sleep 1
+  fi
+done <<< "$tags_to_delete_list"
+
+if [ "$delete_failures" -gt 0 ]; then
+  echo "Cleanup finished with ${delete_failures} deletion failure(s)"
+  exit 1
+fi
+
+echo "Cleanup completed successfully"

.buildkite/release/scripts/publish-release-images.sh

@@ -1,50 +1,87 @@
 #!/bin/bash
 # SPDX-License-Identifier: Apache-2.0
 #
-# Publish release Docker images from ECR to DockerHub.
-# Pulls per-arch images, tags with `latest` and versioned tags, pushes them,
-# then creates and pushes a multi-arch manifest.
+# Publish Docker images from ECR staging to DockerHub.
+#
+# Usage:
+#   publish-release-images.sh            # release mode (default)
+#   publish-release-images.sh release
+#   publish-release-images.sh nightly
+#
+# Release tags: latest, v<version>
+# Nightly tags: nightly, nightly-<commit>
 
 set -euo pipefail
 
-RELEASE_VERSION=$(buildkite-agent meta-data get release-version --default "" | sed 's/^v//')
-if [ -z "${RELEASE_VERSION}" ]; then
-  echo "ERROR: release-version metadata not set"
-  exit 1
-fi
+MODE="${1:-release}"
 
-COMMIT="$BUILDKITE_COMMIT"
 DOCKERHUB_REPO="vllm/vllm-omni"
 ECR_REPO="public.ecr.aws/q9t5s3a7/vllm-omni-release-repo"
+COMMIT="$BUILDKITE_COMMIT"
+PRIMARY_TAGS=()
+EXTRA_MANIFEST_TAGS=()
+
+case "${MODE}" in
+  release)
+    RELEASE_VERSION=$(buildkite-agent meta-data get release-version --default "" | sed 's/^v//')
+    if [ -z "${RELEASE_VERSION}" ]; then
+      echo "ERROR: release-version metadata not set"
+      exit 1
+    fi
+    PRIMARY_TAGS=("latest" "v${RELEASE_VERSION}")
+    ;;
+  nightly)
+    # Commit-pinned manifest reuses nightly-{x86_64,aarch64} arch images.
+    PRIMARY_TAGS=("nightly")
+    EXTRA_MANIFEST_TAGS=("nightly-${COMMIT}")
+    ;;
+  *)
+    echo "Usage: $0 {release|nightly}"
+    exit 2
+    ;;
+esac
 
 echo "========================================"
-echo "Publishing release images v${RELEASE_VERSION}"
+echo "Publishing ${MODE} images to ${DOCKERHUB_REPO}"
 echo "  Commit: ${COMMIT}"
+echo "  Primary tags: ${PRIMARY_TAGS[*]}"
+if [ ${#EXTRA_MANIFEST_TAGS[@]} -gt 0 ]; then
+  echo "  Extra manifests: ${EXTRA_MANIFEST_TAGS[*]}"
+fi
 echo "========================================"
 
-# Login to ECR to pull staging images
 aws ecr-public get-login-password --region us-east-1 | \
   docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7
 
-docker pull "${ECR_REPO}:${COMMIT}-x86_64"
-docker pull "${ECR_REPO}:${COMMIT}-aarch64"
+SRC_X86="${ECR_REPO}:${COMMIT}-x86_64"
+SRC_ARM="${ECR_REPO}:${COMMIT}-aarch64"
 
-docker tag "${ECR_REPO}:${COMMIT}-x86_64" "${DOCKERHUB_REPO}:latest-x86_64"
-docker tag "${ECR_REPO}:${COMMIT}-x86_64" "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-x86_64"
-docker push "${DOCKERHUB_REPO}:latest-x86_64"
-docker push "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-x86_64"
+docker pull "${SRC_X86}"
+docker pull "${SRC_ARM}"
 
-docker tag "${ECR_REPO}:${COMMIT}-aarch64" "${DOCKERHUB_REPO}:latest-aarch64"
-docker tag "${ECR_REPO}:${COMMIT}-aarch64" "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-aarch64"
-docker push "${DOCKERHUB_REPO}:latest-aarch64"
-docker push "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-aarch64"
+publish_manifest() {
+  local tag="$1"
+  local arch_tag_base="$2"
+  docker manifest rm "${DOCKERHUB_REPO}:${tag}" || true
+  docker manifest create \
+    "${DOCKERHUB_REPO}:${tag}" \
+    "${DOCKERHUB_REPO}:${arch_tag_base}-x86_64" \
+    "${DOCKERHUB_REPO}:${arch_tag_base}-aarch64"
+  docker manifest push "${DOCKERHUB_REPO}:${tag}"
+}
 
-docker manifest rm "${DOCKERHUB_REPO}:latest" || true
-docker manifest rm "${DOCKERHUB_REPO}:v${RELEASE_VERSION}" || true
-docker manifest create "${DOCKERHUB_REPO}:latest" "${DOCKERHUB_REPO}:latest-x86_64" "${DOCKERHUB_REPO}:latest-aarch64"
-docker manifest create "${DOCKERHUB_REPO}:v${RELEASE_VERSION}" "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-x86_64" "${DOCKERHUB_REPO}:v${RELEASE_VERSION}-aarch64"
-docker manifest push "${DOCKERHUB_REPO}:latest"
-docker manifest push "${DOCKERHUB_REPO}:v${RELEASE_VERSION}"
+for tag in "${PRIMARY_TAGS[@]}"; do
+  docker tag "${SRC_X86}" "${DOCKERHUB_REPO}:${tag}-x86_64"
+  docker tag "${SRC_ARM}" "${DOCKERHUB_REPO}:${tag}-aarch64"
+  docker push "${DOCKERHUB_REPO}:${tag}-x86_64"
+  docker push "${DOCKERHUB_REPO}:${tag}-aarch64"
+  publish_manifest "${tag}" "${tag}"
+done
+
+ARCH_BASE="${PRIMARY_TAGS[0]}"
+for tag in "${EXTRA_MANIFEST_TAGS[@]}"; do
+  publish_manifest "${tag}" "${ARCH_BASE}"
+done
 
 echo ""
-echo "Successfully published release images for v${RELEASE_VERSION}"
+echo "Successfully published ${MODE} images"

.claude/skills/precheck-pr/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: precheck-pr
-description: Self-check your branch before creating a PR — catch dead code, verify accuracy/perf claims, validate PR title format, and confirm merge readiness. Use when the user says "precheck", "self review", "pre-submit check", or "check my PR before I open it." Never posts to GitHub.
+description: Self-check your branch before creating a PR — catch dead code, prevent new model-specific Python examples, verify accuracy/perf claims, validate PR title format, and confirm merge readiness. Use when the user says "precheck", "self review", "pre-submit check", or "check my PR before I open it." Never posts to GitHub.
 ---
 
 # PR Pre-Check
@@ -65,6 +65,8 @@ Ask: "Quick mode or full mode?" Then walk the checklist for the detected PR type
 
 **Also run the Code-Quality sweep on every PR**, regardless of type or mode: the five diff-scoped checks in [references/code-quality.md](references/code-quality.md) — kwargs fragility, broad-except swallow, `Any`/wrong type hints, hot-path `.clone()`/`deepcopy`, and event-loop blocking — plus the advisory conventions (log level, structured logging, synchronization, cleanup, dependencies, naming) in [checklists.md](references/checklists.md). These count only lines the PR *adds* — the pre-existing backlog across the repo is out of scope.
 
+**Also run the Examples-Policy check on every PR** using [references/examples-policy.md](references/examples-policy.md). Inspect only Python paths introduced by the diff. Treat a new model-, checkpoint-, vendor-, or family-specific Python example as blocking; do not report pre-existing example debt.
+
 ### Step 5: Print Report
 
 ```
@@ -77,6 +79,7 @@ Pre-check report for <branch>
   ─────────────────  ──────
   PR title format    ✓
   Code quality       ⚠ 1 broad except, 2 Any hints
+  Examples policy    ✗ new model-specific Python example
   PR desc integrity  ✓
   Registry/config    ✓
   Dead code          ⚠ 2 warnings

.claude/skills/precheck-pr/references/checklists.md

@@ -47,6 +47,19 @@ Diff-scoped sweep for five fragility patterns that are pervasive in the existing
 
 Roll the ⚠/✗ counts into the report as a single **Code quality** dimension row.
 
+---
+
+## All PRs: Examples Policy (run on every PR, quick & full)
+
+Inspect Python files introduced under `examples/` using
+[examples-policy.md](examples-policy.md). This is a diff-scoped ratchet: existing
+model-specific examples may be edited or removed without triggering the check.
+
+- [ ] **No new model-specific Python examples:** the diff adds, copies, or renames no Python file whose path or behavior is specific to one model, checkpoint, vendor, or model family. ✗ for any violation. Route inference through a shared task/protocol entrypoint, model request behavior through `vllm_omni/model_extras`, and commands through task docs or `recipes/`.
+- [ ] **New generic example code is justified:** any new task-neutral Python file under `examples/` is an actual user-facing entrypoint rather than reusable production logic, a benchmark, an app, a setup tool, or a reproducer. ⚠ when the placement is questionable.
+
+Roll the result into the report as a separate **Examples policy** dimension.
+
 ### Conventions (eyeball — no grep)
 
 Judgment calls that don't grep cleanly — apply them while reading the diff, not as a mechanical check:

.claude/skills/precheck-pr/references/examples-policy.md

@@ -0,0 +1,68 @@
+# Examples Python Policy
+
+Prevent new model-specific Python examples without forcing cleanup of the
+existing backlog. Run this check on every PR in quick and full modes.
+
+## Scope
+
+Set the merge base as in the main workflow, then list Python paths introduced
+by the PR, including additions, copies, and renames:
+
+```bash
+BASE="$(git merge-base HEAD origin/main)"
+git diff --name-status --diff-filter=ACR --find-renames --find-copies \
+  "${BASE}"...HEAD -- 'examples/**/*.py'
+```
+
+Inspect destination paths for copies and renames. Existing model-specific files
+that are only modified or deleted are grandfathered and must not be reported.
+
+## Blocking classification
+
+Mark ✗ when a newly introduced Python file does any of the following:
+
+- Names or scopes a path to one model, checkpoint, vendor, or model family.
+- Hard-codes a model identifier, model-specific prompt, request shape, output
+  conversion, or launch configuration behind an otherwise generic filename.
+- Duplicates an existing task/protocol runner to demonstrate one model.
+- Adds a per-model helper below a generic task hub such as `text_to_speech/`.
+
+Do not infer safety from a generic filename alone. For example, a file named
+`text_to_audio.py` is still model-specific if it only implements one model's
+contract.
+
+## Acceptable additions
+
+A new Python example may pass when it is a genuinely model-neutral,
+user-facing task or protocol entrypoint. It must accept model selection through
+configuration and keep model-specific behavior out of the script.
+
+Use ⚠ when the file is model-neutral but reusable logic appears misplaced under
+`examples/`. Route code by purpose:
+
+| Code purpose | Preferred location |
+|---|---|
+| Model prompt, defaults, request/output adaptation | `vllm_omni/model_extras/` or production model modules |
+| Runnable model command and validation evidence | Task documentation or `recipes/` |
+| UI or interactive application | `apps/` |
+| Benchmark or evaluator | `benchmarks/` |
+| Regression reproducer | `tests/` |
+| Download, conversion, or setup utility | `tools/` |
+
+If a model cannot use an existing shared runner, report the missing generic
+capability. Do not solve the gap by adding a model-specific Python example.
+
+## Report format
+
+Add one row to the pre-check report:
+
+```text
+Examples policy    ✓ no new Python example paths
+Examples policy    ⚠ generic helper belongs in tools/
+Examples policy    ✗ examples/offline_inference/new_model/end2end.py is model-specific
+Examples policy    ✗ examples/offline_inference/x_to_y_model_name.py is model-specific
+```
+
+For a blocker, name the path, the model-specific behavior, and the appropriate
+shared runner or destination. Do not propose deleting unrelated existing
+examples as part of the PR.

.claude/skills/review-pr/SKILL.md

@@ -97,6 +97,7 @@ If docs and live code disagree, verify the code/tests and report the drift.
 | [test-quality-evaluation.md](references/checks/test-quality-evaluation.md) | Tests change, are absent for risky code, or may not exercise production behavior. |
 | [tests-docs-checklist.md](references/checks/tests-docs-checklist.md) | Coverage, CI markers, examples, user docs, or PR evidence need review. |
 | [verification.md](references/checks/verification.md) | Hardware, a server, or a runnable affected path is available for active verification. |
+| [examples-policy.md](../precheck-pr/references/examples-policy.md) | The PR adds, copies, or renames Python under `examples/`; apply the canonical policy shared with `precheck-pr`. |
 
 ### Delivery and reviewer coordination
 
@@ -152,6 +153,13 @@ contract they protect or use only the applicable evidence checks.
 Apply every category in [general-checks.md](references/process/general-checks.md) before
 lower-priority comments.
 
+If the diff census contains an added, copied, or renamed Python path under
+`examples/`, read and apply the canonical
+[examples policy](../precheck-pr/references/examples-policy.md). Treat a new
+model-specific Python example as blocking. Do not flag model-specific example
+debt that the PR only modifies or removes, and do not run the rest of the
+author-oriented `precheck-pr` workflow.
+
 For each changed value or behavior, trace:
 
 ```text

.readthedocs.yml

@@ -11,10 +11,15 @@ build:
     post_checkout:
       - git fetch --unshallow || true
     pre_install:
+      # `pre_install` runs before RTD upgrades pip, and the seeded pip 23.1
+      # rejects the `typing_extensions` wheel torch needs (fixed in pip 23.2).
+      - python -m pip install --upgrade pip
       # Pre-install CPU torch so the subsequent `pip install .[docs]` does not
       # pull ~2 GB of CUDA 13 libs (cuBLAS/cuDNN/NCCL/...) that exceed RTD's
-      # 1 GB memory cap and time out the mkdocs build.
-      - python -m pip install --index-url https://download.pytorch.org/whl/cpu torch
+      # 1 GB memory cap and time out the mkdocs build. Wheel-only: `--index-url`
+      # replaces PyPI, so an sdist fallback cannot reach its build deps. The
+      # quotes keep YAML from reading `:all:` as a mapping key.
+      - "python -m pip install --only-binary=:all: --index-url https://download.pytorch.org/whl/cpu torch"
 
 mkdocs:
   configuration: mkdocs.yml

benchmarks/diffusion/quantization_quality.py

@@ -158,7 +158,6 @@ def _build_omni_kwargs(args, quantization=None):
 def _generate_image(omni, args, prompt, seed):
     """Generate a single image and return (PIL.Image, time_seconds, memory_gib)."""
     from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-    from vllm_omni.outputs import OmniRequestOutput
     from vllm_omni.platforms import current_omni_platform
 
     generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(seed)
@@ -176,7 +175,7 @@ def _generate_image(omni, args, prompt, seed):
     elapsed = time.perf_counter() - start
     peak_mem = torch.accelerator.max_memory_allocated() / (1024**3)
 
-    req_out = OmniRequestOutput.unwrap_result(outputs)
+    req_out = outputs[0] if isinstance(outputs, list) else outputs
     if not req_out.images:
         raise ValueError("Could not extract image output from result.")
     img = req_out.images[0]
@@ -206,15 +205,15 @@ def _generate_video(omni, args, prompt, seed):
     elapsed = time.perf_counter() - start
     peak_mem = torch.accelerator.max_memory_allocated() / (1024**3)
 
-    first = outputs[0]
-    if hasattr(first, "request_output") and isinstance(first.request_output, list):
-        inner = first.request_output[0]
-        if isinstance(inner, OmniRequestOutput) and hasattr(inner, "images"):
-            frames = inner.images[0] if inner.images else None
-        else:
-            frames = inner
-    elif hasattr(first, "images") and first.images:
-        frames = first.images[0]
+    if isinstance(outputs, list) and isinstance(outputs[0], OmniRequestOutput):
+        first = outputs[0]
+    elif isinstance(outputs, OmniRequestOutput):
+        first = outputs
+    else:
+        raise ValueError("Could not extract video frames from output.")
+
+    if hasattr(first, "images"):
+        frames = first.images[0] if first.images else None
     else:
         raise ValueError("Could not extract video frames from output.")
 

benchmarks/glm_image/vllm-omni/inference.py

@@ -300,7 +300,7 @@ def benchmark(args: argparse.Namespace) -> None:
         output_idx = 0
         for stage_outputs in omni.generate(all_prompts, sampling_params_list, py_generator=True):
             if stage_outputs.final_output_type == "image":
-                request_output = stage_outputs.request_output
+                request_output = stage_outputs
                 request_id = getattr(request_output, "request_id", "")
 
                 images = getattr(request_output, "images", [])

benchmarks/lingbot_video/dense_pipeline_parity.py

@@ -167,9 +167,11 @@ def main() -> None:
     ]
     native_cmd = [
         sys.executable,
-        "examples/offline_inference/text_to_video/text_to_video_lingbot.py",
+        "examples/offline_inference/text_to_video/text_to_video.py",
         "--model",
         args.model,
+        "--model-class-name",
+        "LingBotVideoPipeline",
         "--prompt",
         args.prompt,
         "--output",

docker/Dockerfile.npu

@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_TAG=v0.26.0
+ARG VLLM_ASCEND_TAG=v0.27.0
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_TAG}
 
 # WORKDIR /vllm-workspace/vllm

docker/Dockerfile.npu.a3

@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_TAG=v0.26.0-a3
+ARG VLLM_ASCEND_TAG=v0.27.0-a3
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_TAG}
 
 # WORKDIR /vllm-workspace/vllm

docker/Dockerfile.npu.ci

@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_TAG=v0.26.0
+ARG VLLM_ASCEND_TAG=v0.27.0
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_TAG}
 
 ARG APP_DIR=/vllm-workspace/vllm-omni

docker/Dockerfile.npu.ci.a3

@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_TAG=v0.26.0-a3
+ARG VLLM_ASCEND_TAG=v0.27.0-a3
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_TAG}
 
 ARG APP_DIR=/vllm-workspace/vllm-omni

docs/.nav.yml

@@ -71,7 +71,7 @@ nav:
         - Hybrid Sharded Data Parallel: user_guide/diffusion/parallelism/hsdp.md
         - Sequence Parallel: user_guide/diffusion/parallelism/sequence_parallel.md
         - Tensor Parallel: user_guide/diffusion/parallelism/tensor_parallel.md
-        - VAE Parallelism: user_guide/diffusion/parallelism/vae_parallelism.md
+        - VAE Parallel: user_guide/diffusion/parallelism/vae_parallel.md
       - Attention Backends: user_guide/diffusion/attention_backends.md
       - Regional Compilation: user_guide/diffusion/regional_compilation.md
       - Frame Interpolation: user_guide/diffusion/frame_interpolation.md
@@ -123,7 +123,7 @@ nav:
           - Pipeline Parallel: design/feature/pipeline_parallel.md
           - Sequence Parallel: design/feature/sequence_parallel.md
           - Tensor Parallel: design/feature/tensor_parallel.md
-          - VAE Patch Parallelism: design/feature/vae_parallel.md
+          - VAE Parallel: design/feature/vae_parallel.md
         - Attention Optimization:
           - Skip-Softmax: design/feature/skip_softmax.md
         - Cache-DiT: design/feature/cache_dit.md
@@ -163,7 +163,8 @@ nav:
 - CLI Reference: cli
 - Community:
   - Governance: community/governance.md
-  - community/*
+  - Contact Us: community/contact_us.md
+  - Community Meetings: community/meetings.md
   - Slack: https://slack.vllm.ai
   - Blog: https://blog.vllm.ai
   - Forum: https://discuss.vllm.ai

docs/community/contact_us.md

@@ -1,5 +1,9 @@
 # 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-omni/security/advisories) feature
+- For bugs and feature requests, search or open a GitHub [issue](https://github.com/vllm-project/vllm-omni/issues).
+- For technical and usage questions, join the `sig-omni` channel in [vLLM Slack](https://slack.vllm.ai/) or use the [vLLM Forum](https://discuss.vllm.ai/).
+- For coordinating contributions and development, use `sig-omni`, the forum, or join a [community meeting](meetings.md).
+
+## Security Disclosures
+
+Do not share vulnerability details in public GitHub issues, pull requests, discussions, or Slack channels. To arrange a private report, send a direct message to one of the [lead maintainers](governance.md#lead-maintainers) in [vLLM Slack](https://slack.vllm.ai/). If you cannot send a direct message, you may post a detail-free request in `sig-omni` asking a lead maintainer to contact you privately; do not include any vulnerability details in that request.

docs/community/governance.md

@@ -52,31 +52,11 @@ Committers have write access and merge rights. They typically have deep expertis
 - [@ZJY0516](https://github.com/ZJY0516): Diffusion attention backend, kernel fusion, and CustomOp
 - [@yuanheng-zhao](https://github.com/yuanheng-zhao): Diffusion cache, offload, and Omni Support
 
-## Diffusion Workload Division
+## Reviewer Routing
 
-This section breaks down diffusion-related responsibilities. If you have PRs touching these areas, please ping the listed owners for review.
+For path-based reviewer routing, consult [`.github/CODEOWNERS`](https://github.com/vllm-project/vllm-omni/blob/main/.github/CODEOWNERS). Detailed subsystem boundaries and safe-change guidance live in the [module design documents](../design/index.md#module-design-documents).
 
-### Runtime-related
-
-- Diffusion models: @RuixiangMa, @wtomin, @xuechendi
-
-### Optimization
-
-- Parallel: @RuixiangMa, @wtomin
-- Attention backend: @ZJY0516
-- Cache: @yuanheng-zhao, @RuixiangMa, @SamitHuang
-- Offload: @yuanheng-zhao
-- Intel GPU support: @xuechendi
-- Quantization: @lishunyang12, @david6666666
-- Kernel fusion / communication-computation: @ZJY0516
-
-### Docs & Test
-
-- Diffusion docs: @RuixiangMa, @wtomin
-
-## Meetings
-
-Committers hold **bi-weekly meetings** to discuss future directions and collaborations of the project.
+`CODEOWNERS` is operational routing metadata, not a governance or approval policy. Being listed does not grant committer status or merge rights, and it does not mean every listed reviewer must approve a change. The maintainer roster above is authoritative for project roles and merge rights.
 
 ## Committer Nomination Process
 

docs/community/meetings.md

@@ -0,0 +1,8 @@
+# Community Meetings
+
+vLLM-Omni hosts developer-facing meetings for Chinese- and English-language audiences. Schedules and access details can change, so use these maintained documents as the source of truth:
+
+- [Chinese-language meeting document](https://docs.google.com/document/d/1pdUBiS_7mdOUNDtdwy-9OUf7jsMWN324BIbps5olDME/edit?tab=t.0)
+- [English-language meeting document](https://docs.google.com/document/d/19AlD08T8uJ7wqfs8VtYCi0iGMTwlTYVVlAxFZ3S7OIg/edit?tab=t.0)
+
+Each document contains the current schedule, access details, agendas, and past meeting notes. For broader vLLM community events, see the [vLLM events calendar](https://vllm.ai/events/).

docs/community/meetups.md

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

docs/community/volunteers.md

@@ -1,12 +0,0 @@
-# Volunteers for Bugfix and CI
-
-We encourage you to check current docs and [issues](https://github.com/vllm-project/vllm-omni/issues) to find possible solutions for your questions. If non of these can solve it, please propose an issue to describe your questions about bug or CI problems for developing.
-
-If you have urgent need for locating and solving bugfix or CI problems, please find community volunteers below.
-
-| Dec 4-Dec 12 | Dec 15-Dec 19 | Dec 22-Dec 26 | Dec 29- Jan 2, 2026| Jan 5-Jan 9 | Jan 12-Jan 16 |
-|----------|----------|----------|----------|----------|----------|
-| <a href="https://github.com/congw729">Conw729</a> | <a href="https://github.com/yinpeiqi">yinpeiqi</a> | <a href="https://github.com/tzhouam">tzhouam</a> | <a href="https://github.com/SamitHuang">SamitHuang</a> | <a href="https://github.com/gcanlin">gcanlin</a> | <a href="https://github.com/natureofnature">natureofnature</a> |
-| <a href="https://github.com/david6666666">david6666666</a> | <a href="https://github.com/R2-Y">R2-Y</a> | <a href="https://github.com/hsliuustc0106">hsliuustc0106</a> | <a href="https://github.com/Gaohan123">Gaohan123</a> | <a href="https://github.com/ZJY0516">ZJY0516</a> | <a href="https://github.com/qibaoyuan">qibaoyuan</a> |
-
-We kindly welcome more contributors to fix bugs and contribute new features!

docs/contributing/README.md

@@ -3,7 +3,7 @@
 Thank you for your interest in contributing to vLLM-Omni! This document provides guidelines and instructions for contributing.
 
 !!! note
-    We host weekly developer-facing online meetings to discuss milestones and updates **every Tuesday at 19:30 PDT**. Meeting link as well as the past meeting notes can be found [here](https://tinyurl.com/vllm-omni-meeting).
+    vLLM-Omni hosts developer-facing meetings for Chinese- and English-language audiences. See [Community Meetings](../community/meetings.md) for current schedules, access details, agendas, and past notes.
 
 ## Getting Started
 
@@ -122,7 +122,7 @@ Repository skills guide the agent, but they do not replace contributor judgment,
 If you encounter a bug or have a feature request, please search existing issues first to see if it has already been reported. If not, please file a new issue, providing as much relevant information as possible.
 
 !!! important
-    If you discover a security vulnerability, please report it by creating a GitHub issue with the `security` label.
+    Do not report suspected security vulnerabilities through a public issue, pull request, discussion, or Slack channel. Follow the [security disclosure instructions](../community/contact_us.md#security-disclosures) to arrange a private report.
 
 ## Pull Requests & Code Reviews
 

docs/design/architecture_overview.md

@@ -398,7 +398,7 @@ the source for configuration, compatibility, and implementation details.
 * **Composable parallelism:** Diffusion stages can combine [CFG-Parallel](feature/cfg_parallel.md),
   [Expert Parallel](feature/expert_parallel.md), [HSDP](feature/hsdp.md),
   [Pipeline Parallel](feature/pipeline_parallel.md), [Sequence Parallel](feature/sequence_parallel.md),
-  [Tensor Parallel](feature/tensor_parallel.md), and [VAE Patch Parallelism](feature/vae_parallel.md)
+  [Tensor Parallel](feature/tensor_parallel.md), and [VAE Parallel](feature/vae_parallel.md)
   according to the pipeline and hardware topology.
 * **Attention and cache acceleration:** [Skip-Softmax](feature/skip_softmax.md),
   [Cache-DiT](feature/cache_dit.md), and [TeaCache](feature/teacache.md)

docs/design/feature/vae_parallel.md

@@ -1,42 +1,57 @@
-# VAE Patch Parallelism
+# VAE Parallel
 
-This document describes how to add **VAE Patch Parallelism** support to a diffusion model.
-We use **Qwen-Image** as the reference implementation for decode parallel, and **Wan2.2** for encode parallel.
+This document defines the VAE parallelism strategy contract and describes how to
+add a strategy to a diffusion model. Tile/patch parallelism is the portable
+baseline. Spatial-shard parallelism is a model-specific decode strategy that can
+be selected explicitly or through a request-aware policy.
+
+We use **Qwen-Image** as the reference implementation for tile-parallel decode,
+**Wan2.2** for tile-parallel encode, and **Wan2.2** as the first spatial-shard
+decode implementation. Other VAE families can add spatial strategies later
+without changing the public mode meanings defined here.
 
 ---
 
 ## Table of Contents
 
 - [Overview](#overview)
-- [Step-by-Step Implementation (Decode)](#step-by-step-implementation-decode)
-- [Encode Parallel Implementation](#encode-parallel-implementation)
+- [Strategy Selection Contract](#strategy-selection-contract)
+- [Tile/Patch Implementation (Decode)](#tilepatch-implementation-decode)
+- [Tile/Patch Encode Implementation](#tilepatch-encode-implementation)
 - [Testing](#testing)
 - [Reference Implementations](#reference-implementations)
+- [Spatially-Sharded Decode](#spatially-sharded-decode)
+- [Adding Another VAE Strategy](#adding-another-vae-strategy)
 - [Summary](#summary)
 
 ---
 
 ## Overview
 
-### What is Vae Patch parallel?
+### What is VAE parallelism?
 
-**VAE Patch Parallelism** is an acceleration technique for both **encoding** and **decoding**. Instead of processing the entire tensor at once, the tensor is:
+VAE parallelism distributes VAE **encoding** or **decoding** across multiple
+ranks. The implementation currently has two strategy families:
 
-+ Split into multiple spatial tiles
+| Strategy | Unit of work | Communication | Current scope |
+|----------|--------------|---------------|---------------|
+| Tile/patch | Independent overlapping tiles | Gather, stitch, and optionally broadcast | Encode and decode across supported distributed VAEs |
+| Spatial shard | One feature map sharded along a spatial axis | Per-layer halo exchange followed by output gather | Wan decode |
 
-+ Distributed across multiple ranks
+Both strategies preserve the model-specific receptive field and reconstruct the
+same logical output as single-rank execution. They differ in how work is divided:
+tile parallelism assigns complete overlapping tiles to ranks, while spatial
+sharding keeps one logical feature map partitioned throughout the decoder.
 
-+ Encoded/Decoded in parallel
+This approach can:
 
-+ Merged to reconstruct the final output
++ Distribute computation across multiple devices
++ Reduce peak memory usage per device
++ Accelerate encoding or decoding latency
 
-This approach:
-
-+ Distributes computation across multiple devices
-
-+ Reduces peak memory usage per device
-
-+ Accelerates encoding/decoding latency
+The actual benefit depends on shape, topology, communication cost, and VAE
+architecture. A strategy must not become the global default without parity and
+performance evidence on its supported models.
 
 ### When to Use Encode vs Decode Parallel
 
@@ -45,8 +60,16 @@ This approach:
 | **Decode Parallel** | Text-to-Image, Text-to-Video | Latent → Image/Video |
 | **Encode Parallel** | Image-to-Video (I2V) | Image → Latent (for conditioning) |
 
-### Architecture
-We introduce **DistributedVaeExecutor** as the core component responsible for distributed VAE encoding/decoding.
+### Architecture and ownership
+
+`DiffusionParallelConfig.vae_parallel_mode` carries the public strategy request.
+The shared registry forwards that value to the VAE, but each distributed VAE
+adapter owns its supported modes, request-shape selection, topology validation,
+and fallback behavior. Model-specific policy must not be added to the shared
+runner or registry.
+
+**DistributedVaeExecutor** is the model-agnostic core for tile-parallel VAE
+encoding and decoding.
 
 The executor is model-agnostic and accepts three function parameters:
 
@@ -76,6 +99,11 @@ This design separates:
 
 + Model-specific tiling and merging logic
 
+Spatial-shard implementations use the same configured VAE process group but do
+not use the tile task executor. They keep the feature map sharded across decoder
+layers and own their halo-exchange and final-gather contracts inside the model's
+distributed autoencoder implementation.
+
 #### Why split / exec / merge is necessary?
 
 The latent tensor cannot be arbitrarily partitioned.
@@ -92,7 +120,34 @@ Therefore:
 
 + Merge must perform blending to avoid seams
 
-## Step-by-Step Implementation (Decode)
+## Strategy Selection Contract
+
+`vae_parallel_mode="tile"` is the compatibility-preserving global default.
+Other values are stable strategy requests, but support remains model-specific.
+
+| Mode | Meaning | Current implementation | Unsupported or ineligible request |
+|------|---------|------------------------|-----------------------------------|
+| `tile` | Use the existing tile/patch executor | All distributed VAE adapters | Not applicable |
+| `auto` | Let the VAE adapter select a validated strategy for each request | Wan decode selects height or width spatial sharding | Wan uses tile when ineligible; unsupported VAE families reject or document their fallback |
+| `spatial_shard_height` | Force spatial-shard decode along height | Wan decode | Wan warns and uses tile when ineligible; unsupported VAE families reject |
+| `spatial_shard_width` | Force spatial-shard decode along width | Wan decode | Wan warns and uses tile when ineligible; unsupported VAE families reject |
+
+The shared configuration layer validates the mode name. It does not decide
+whether a model supports that mode. A VAE adapter adding a mode must:
+
+1. Keep `tile` as an exact feature-off path.
+2. Select a strategy per request rather than from stale process-global state.
+3. Validate input rank/shape and process-group membership before collectives.
+4. Reject or explicitly fall back from unsupported combinations; never silently
+   reinterpret one forced strategy as another.
+5. Preserve direct encode/decode behavior when Diffusers does not select a tiled
+   path.
+
+`auto` is a policy hook, not a universal heuristic. Each VAE family may select
+from only the strategies it has validated. Adding another family must not change
+Wan's selector or the meanings of the explicit modes.
+
+## Tile/Patch Implementation (Decode)
 
 ### Step 1: Implement DistributedAutoencoderKLQwenImage
 `QwenImagePipeline` use `AutoencoderKLQwenImage` for vae, so implement a distributed version:
@@ -251,7 +306,7 @@ class YourModelPipeline(nn.Module):
 +       ).to(self.device)
 ```
 
-## Encode Parallel Implementation
+## Tile/Patch Encode Implementation
 
 For models that require VAE encoding (e.g., Image-to-Video), you can also parallelize the encode operation. We use **Wan2.2** as the reference implementation.
 
@@ -449,18 +504,24 @@ Complete examples in the codebase:
 
 ---
 
-## Spatially-Sharded Decode (Wan)
+## Spatially-Sharded Decode
 
-The tile-parallel executor above assigns independent spatial **tiles** to ranks. The Wan VAE additionally supports a **spatially-sharded decode** backend, selected through `DiffusionParallelConfig.vae_parallel_mode` (`"spatial_shard_height"` or `"spatial_shard_width"`; default is `"tile"`).
+The tile-parallel executor above assigns independent spatial **tiles** to ranks.
+A spatial-shard backend instead partitions one global decoder feature map along
+height or width. Wan is the first supported implementation.
+
+`DiffusionParallelConfig.vae_parallel_mode="tile"` remains the default. For Wan,
+opt into request-shape selection with `"auto"`, or force an axis with
+`"spatial_shard_height"` or `"spatial_shard_width"`.
 
 ### How it differs from tile parallel
 
-| Aspect | Tile parallel (`"tile"`) | Spatially-sharded (`"spatial_shard_height"`/`"spatial_shard_width"`) |
-|--------|--------------------------|-----------------------------------------------|
-| Unit of work | Independent overlapping tiles | A single global feature map sharded along H or W |
-| Cross-rank communication | Gather tiles to rank 0, stitch + blend | Per-conv **halo exchange** of boundary rows/cols (P2P) |
-| Output assembly | Blend overlapping tiles | All-gather shards on rank 0, trim padding (matches `broadcast_result=False`) |
-| Scope | Decode + encode | Decode only |
+| Aspect | Tile parallel (`"tile"`) | Automatic (`"auto"`) | Spatially-sharded (`"spatial_shard_height"`/`"spatial_shard_width"`) |
+|--------|--------------------------|----------------------|-----------------------------------------------|
+| Unit of work | Independent overlapping tiles | Chooses tile or a spatial shard per request | A single global feature map sharded along H or W |
+| Cross-rank communication | Gather tiles to rank 0, stitch + blend | Depends on the selected path | Per-conv **halo exchange** of boundary rows/cols (P2P) |
+| Output assembly | Blend overlapping tiles | Depends on the selected path | All-gather shards on rank 0, trim padding (matches `broadcast_result=False`) |
+| Scope | Decode + encode | Wan decode selection; encode remains tiled | Wan decode only |
 
 Spatial-shard decode swaps the decoder's spatial convolutions/padding for halo-exchanging variants (`WanDistConv2d`, `WanDistCausalConv3d`, `WanDistZeroPad2d`) so each rank only holds a shard of the activations but still sees the correct receptive field at shard boundaries. Implementation lives in `vllm_omni/diffusion/distributed/autoencoders/wan_spatial_shard.py`.
 
@@ -476,21 +537,65 @@ serve.py (--vae-parallel-mode) / OmniEngineArgs
   -> DistributedAutoencoderKLWan.tiled_decode dispatch
 ```
 
-`DistributedAutoencoderKLWan._spatial_shard_decode_enabled()` gates the path: it requires distributed decode to be enabled, a 5D latent, and `vae_patch_parallel_size == DiT group size`. Otherwise it logs a warning and falls back to tile-parallel decode.
+`DistributedAutoencoderKLWan._spatial_shard_decode_enabled()` gates the path: it
+requires distributed decode to be enabled, a 5D latent, and
+`vae_patch_parallel_size == DiT group size`. The method is reached only after
+Diffusers has selected tiled decode for the request. In `auto` mode the longer
+axis that can cover every rank is selected, with width winning ties. If neither
+axis can cover the group, auto mode uses tile decode. This policy is cached only
+through normal kernel shape caches; it is reevaluated for every request.
 
 ### Notes
 
-- The decoder is patched **in place** the first time spatial-shard decode runs and is bound to a single split dimension for the lifetime of the VAE instance.
+- The decoder is patched **in place** the first time spatial-shard decode runs. Its wrappers consult a context-local split dimension and retain an exact direct fallback, allowing later requests to select tile, height, or width safely.
 - Numerical correctness vs. single-GPU decode is covered by `tests/diffusion/distributed/test_wan_spatial_shard.py::test_spatial_shard_decode_matches_reference` (multi-GPU, nightly `full_model` + `distributed_cuda`).
 
+## Adding Another VAE Strategy
+
+Add future strategies through the distributed autoencoder that owns the model
+semantics. Do not add model-name dispatch to the shared runner, registry, or
+`DistributedVaeExecutor`.
+
+For a new VAE family:
+
+1. **Declare capability**: document which modes, operations, shapes, dtypes, and
+   process-group layouts the adapter supports.
+2. **Implement model-local selection**: keep `auto` policy and explicit-mode
+   dispatch in the distributed autoencoder. The selector may use request shape,
+   but must be deterministic and side-effect free.
+3. **Preserve the baseline**: retain exact tile/direct fallbacks and state-dict
+   compatibility. Structural wrappers must not register duplicate parameters or
+   alter checkpoint names.
+4. **Define communication**: derive shards from the configured VAE group, keep
+   collectives symmetric, specify padding/halo ownership, and return the same
+   output placement as the tile path.
+5. **Bound request state**: make the selected axis and scratch buffers
+   request-scoped, release temporary buffers on success and failure, and support
+   alternating modes in one long-lived worker.
+6. **Validate independently**: compare each explicit strategy against the
+   single-rank reference, then test `auto`, tile fallback, unsupported topology,
+   and a mixed request sequence. Record per-rank peak memory and decode latency
+   before documenting a preferred mode.
+
+Promote a selector into shared infrastructure only after at least two VAE
+families use the same inputs, invariants, fallback behavior, and lifecycle. Until
+then, shared code owns transport and configuration while each model adapter owns
+policy.
+
 ---
 
 ## Summary
 
-Adding VAE Patch Parallel support to diffusion model:
+Adding VAE parallel support to a diffusion model:
 
-1. **Implement Distributed VAE** - Inherit from base VAE class and `DistributedVaeMixin`
-2. **Decode Parallel** - Refactor `tiled_decode` into `tile_split`/`tile_exec`/`tile_merge`
-3. **Encode Parallel** (optional) - Implement `encode_tile_split`/`encode_tile_exec`/`encode_tile_merge` for I2V models
-4. **Change VAE model in pipeline** - Use the distributed version
-5. **Test** - Verify numerical consistency with `vae_patch_parallel_size=1` vs `N`
+1. **Implement the distributed VAE** - Inherit from the base VAE class and
+   `DistributedVaeMixin`.
+2. **Keep tile as the baseline** - For tile decode, refactor `tiled_decode` into
+   `tile_split`/`tile_exec`/`tile_merge`.
+3. **Add encode parallelism when needed** - Implement
+   `encode_tile_split`/`encode_tile_exec`/`encode_tile_merge` for I2V models.
+4. **Add model-specific strategies locally** - Own selectors, topology checks,
+   communication, and fallback in the distributed autoencoder.
+5. **Wire the pipeline** - Replace the pipeline VAE with its distributed version.
+6. **Test strategy and lifecycle parity** - Compare one rank with every supported
+   strategy, including feature-off, fallback, and mixed-request sequences.

docs/design/index.md

@@ -48,7 +48,7 @@ implementation contract; it is not, by itself, a general support claim.
 - [Pipeline Parallel](feature/pipeline_parallel.md)
 - [Sequence Parallel](feature/sequence_parallel.md)
 - [Tensor Parallel](feature/tensor_parallel.md)
-- [VAE Patch Parallelism](feature/vae_parallel.md)
+- [VAE Parallel](feature/vae_parallel.md)
 
 #### Attention optimization
 

docs/design/metrics.md

@@ -1,6 +1,6 @@
 # Prometheus Metrics Design
 
-This document describes how vLLM-Omni exposes Prometheus metrics for multi-stage pipelines, the constraints that shaped the design, and how the pipeline-level metrics coexist with upstream vLLM per-engine metrics. This iteration covers the text and audio surface; diffusion / image / video families land in a follow-up.
+This document describes how vLLM-Omni exposes Prometheus metrics for multi-stage pipelines, the constraints that shaped the design, and how the pipeline-level metrics coexist with upstream vLLM per-engine metrics.
 
 ## Objectives
 

docs/usage/faq.md

@@ -10,12 +10,12 @@ A: Refer to [GPU memory calculation and configuration](../configuration/gpu_memo
 
 > Q: I encounter some bugs or CI problems, which is urgent. How can I solve it?
 
-A: At first, you can check current [issues](https://github.com/vllm-project/vllm-omni/issues) to find possible solutions. If none of these satisfy your demand and it is urgent, please find these [volunteers](https://docs.vllm.ai/projects/vllm-omni/en/latest/community/volunteers/) for help.
+A: Search the current [issues](https://github.com/vllm-project/vllm-omni/issues) first. If no existing issue matches a bug or CI failure, open a [bug report](https://github.com/vllm-project/vllm-omni/issues/new?template=400-bug-report.yml) with a minimal reproducer, environment details, and relevant logs. For technical or usage questions, use the `sig-omni` channel in [vLLM Slack](https://slack.vllm.ai/) or the [vLLM Forum](https://discuss.vllm.ai/).
 
 > Q: Does vLLM-Omni support AWQ or any other quantization?
 
-A: We plan to introduce GGUF FP8 prequantized models and online FP8 quantization in version 0.16.0. Support for other quantization types will follow in future releases. For details, please see our [Q1 quantization roadmap](https://github.com/vllm-project/vllm-omni/issues/1057).
+A: AWQ is available through the inherited vLLM quantization registry, and vLLM-Omni supports several additional methods. Actual support and validation vary by model, component, hardware, and checkpoint. See the [quantization overview](../user_guide/quantization/overview.md) for validated combinations and method-specific guides.
 
 > Q: Does vLLM-Omni support multimodal streaming input and output?
 
-A: Not yet. We already put it on the [Roadmap](https://github.com/vllm-project/vllm-omni/issues/165). Please stay tuned!
+A: Yes, for specific models and serving paths. Current capabilities include [streaming video input](../serving/video_stream_api.md), [streaming diffusion output](../user_guide/diffusion/execution_modes.md#streaming-output), and experimental full-duplex real-time audio described in the [project overview](../README.md). Check the model and API documentation because input and output support is model-specific.

docs/usage/metrics.md

@@ -1,23 +1,23 @@
 # Production Metrics
 
-vLLM-Omni exposes Prometheus metrics via the `/metrics` endpoint on the OpenAI-compatible API server. This page covers the text and audio surface; diffusion / image / video metrics are tracked in a follow-up PR.
+vLLM-Omni exposes Prometheus metrics via the `/metrics` endpoint on the OpenAI-compatible API server.
 
 ```bash
 vllm-omni serve Qwen/Qwen3-Omni-30B-A3B-Instruct --port 8000 --log-stats
 curl http://localhost:8000/metrics
 ```
 
-**`--log-stats` is required to populate metric data.** Without the flag, the endpoint still returns `200 OK` but the upstream `vllm:*` wrap is not registered at all, and the 15 `vllm:omni_*` families are registered as placeholders with no data written to them. This keeps the runtime cost essentially zero for deployments that don't need monitoring. With the flag, all ~80 families populate.
+**`--log-stats` is required to populate metric data.** Without the flag, the endpoint still returns `200 OK` but the upstream `vllm:*` wrap is not registered at all, and the 15 `vllm_omni:*` families are registered as placeholders with no data written to them. This keeps the runtime cost essentially zero for deployments that don't need monitoring. With the flag, all ~80 families populate.
 
 ## Metric Namespaces
 
 | Prefix | Source | Present when |
 |--------|--------|--------------|
-| `vllm:omni_` | vLLM-Omni orchestrator / audio modality / cross-stage transfer | Pipeline-dependent |
+| `vllm_omni:` | vLLM-Omni orchestrator / audio modality / cross-stage transfer | Pipeline-dependent |
 | `vllm:` | Upstream vLLM engine, wrapped by `OmniPrometheusStatLogger` to expose `{stage, replica}` | Pipeline includes an LLM (AR) stage |
 | `http_` / `process_` | Uvicorn / Python runtime | Always |
 
-## Pipeline-Level Metrics (`vllm:omni_`)
+## Pipeline-Level Metrics (`vllm_omni:`)
 
 Defined in `vllm_omni/metrics/prometheus.py`. Track request lifecycle across the full multi-stage pipeline.
 
@@ -25,42 +25,66 @@ Defined in `vllm_omni/metrics/prometheus.py`. Track request lifecycle across the
 
 | Metric | Type | Labels | Description |
 |--------|------|--------|-------------|
-| `vllm:omni_num_requests_running` | Gauge | `model_name` | Pipeline-global in-flight requests (dispatched to engine, not yet finalized) |
-| `vllm:omni_num_requests_waiting` | Gauge | `model_name` | Requests waiting in the Orchestrator queue |
-| `vllm:omni_requests_success_total` | Counter | `model_name`, `finished_reason` | Total requests by completion reason. `finished_reason` ∈ {`stop`, `length`, `abort`, ...} mirroring upstream `vllm:request_success_total`; aborts cover client disconnect / cancellation paths in addition to upstream `FinishReason.ABORT` |
+| `vllm_omni:num_requests_running` | Gauge | `model_name` | Pipeline-global in-flight requests (dispatched to engine, not yet finalized) |
+| `vllm_omni:num_requests_waiting` | Gauge | `model_name` | Requests waiting in the Orchestrator queue |
+| `vllm_omni:requests_success_total` | Counter | `model_name`, `finished_reason` | Total requests by completion reason. `finished_reason` ∈ {`stop`, `length`, `abort`, ...} mirroring upstream `vllm:request_success_total`; aborts cover client disconnect / cancellation paths in addition to upstream `FinishReason.ABORT` |
 
 ### Latency
 
 | Metric | Type | Labels | Description |
 |--------|------|--------|-------------|
-| `vllm:omni_e2e_request_latency_s` | Histogram | `model_name` | Pipeline-global end-to-end request latency in seconds |
+| `vllm_omni:e2e_request_latency_s` | Histogram | `model_name` | Pipeline-global end-to-end request latency in seconds |
 
-## Audio Modality Metrics (`vllm:omni_`)
+## Audio Modality Metrics (`vllm_omni:`)
 
 Emitted at request finalize, except for `audio_ttfp_s` (streaming-hook at the first audio packet) and `audio_underrun_s` / `audio_continuity_ok_total` (streaming finalize, after the chunk stream is exhausted). All carry `{model_name, stage, replica}` plus the listed extra label.
 
 | Metric | Type | Extra label | Description |
 |--------|------|-------------|-------------|
-| `vllm:omni_audio_ttfp_s` | Histogram | — | Time from request arrival to first audio packet/frame |
-| `vllm:omni_audio_duration_s` | Histogram | — | Audio content duration (`audio_frames / sample_rate`) |
-| `vllm:omni_audio_rtf` | Histogram | — | Real-time factor (`stage_gen_time_s / audio_duration_s`); streaming TTS SLO red line `< 1`; uses `RTF_BUCKETS` |
-| `vllm:omni_audio_frames_total` | Counter | — | Cumulative audio frame count; throughput via `rate()` |
-| `vllm:omni_audio_underrun_s` | Histogram | — | Per-request worst-case player deficit; `> 0` indicates listener heard silent gaps |
-| `vllm:omni_audio_continuity_ok_total` | Counter | `threshold_ms` | Incremented when the request's worst underrun stayed below `threshold_ms` |
-| `vllm:omni_audio_skipped_requests_total` | Counter | `reason` | Silent-loss counter — code2wav rejected malformed codec input and returned `200 OK` with empty audio |
+| `vllm_omni:audio_ttfp_s` | Histogram | — | Time from request arrival to first audio packet/frame |
+| `vllm_omni:audio_duration_s` | Histogram | — | Audio content duration (`audio_frames / sample_rate`) |
+| `vllm_omni:audio_rtf` | Histogram | — | Real-time factor (`stage_gen_time_s / audio_duration_s`); streaming TTS SLO red line `< 1`; uses `RTF_BUCKETS` |
+| `vllm_omni:audio_frames_total` | Counter | — | Cumulative audio frame count; throughput via `rate()` |
+| `vllm_omni:audio_underrun_s` | Histogram | — | Per-request worst-case player deficit; `> 0` indicates listener heard silent gaps |
+| `vllm_omni:audio_continuity_ok_total` | Counter | `threshold_ms` | Incremented when the request's worst underrun stayed below `threshold_ms` |
+| `vllm_omni:audio_skipped_requests_total` | Counter | `reason` | Silent-loss counter — code2wav rejected malformed codec input and returned `200 OK` with empty audio |
 
 The continuity math comes from `vllm_omni/benchmarks/audio_continuity.py::compute_continuity_stats` so the server-side observation aligns with the bench-side definition.
 
-## Cross-Stage Transfer Metrics (`vllm:omni_`)
+## Diffusion Metrics (`vllm_omni:`)
+
+Per-request timing breakdowns for diffusion (image/video) stages. Emitted at request finalize when `stage_metrics.diffusion_metrics` is present. All carry `{model_name, stage, replica}`.
+
+| Metric | Type | Description |
+|--------|------|-------------|
+| `vllm_omni:diffusion_exec_s` | Histogram | DiT forward pass execution time per request in seconds |
+| `vllm_omni:diffusion_exec_per_step_s` | Histogram | DiT forward pass execution time per denoising step in seconds (`exec_s / num_inference_steps`) |
+| `vllm_omni:diffusion_preprocess_s` | Histogram | Diffusion input preprocessing time per request in seconds |
+| `vllm_omni:diffusion_postprocess_s` | Histogram | Diffusion output postprocessing (VAE decode) time per request in seconds |
+
+`diffusion_exec_s` uses `SECONDS_BUCKETS` (50 ms → 300 s); the other three use `SECONDS_FAST_BUCKETS` (1 ms → 60 s). `diffusion_exec_per_step_s` is skipped when `num_inference_steps` is not set.
+
+### Example PromQL
+
+```promql
+# P99 DiT exec time by stage
+histogram_quantile(0.99, rate(vllm_omni:diffusion_exec_s_bucket[5m]))
+
+# Ratio of preprocess to total exec wall time
+rate(vllm_omni:diffusion_preprocess_s_sum[5m])
+  / rate(vllm_omni:diffusion_exec_s_sum[5m])
+```
+
+## Cross-Stage Transfer Metrics (`vllm_omni:`)
 
 Per-physical-transfer histograms tracking the data hop between adjacent stages. Labels `{model_name, from_stage, from_replica, to_stage, to_replica}` let dashboards attribute latency to specific replica edges. `from_replica` / `to_replica` are resolved from the orchestrator's sticky-routing binding (`stage_pool.get_bound_replica_id(request_id)`), so no extra plumbing through `TransferEdgeStats` is needed.
 
 | Metric | Type | Description |
 |--------|------|-------------|
-| `vllm:omni_transfer_size_bytes` | Histogram | Per-transfer payload size in bytes |
-| `vllm:omni_transfer_tx_s` | Histogram | Sender-side time (serialize + submit to connector) |
-| `vllm:omni_transfer_rx_s` | Histogram | Receiver-side time (recv + deserialize) |
-| `vllm:omni_transfer_in_flight_s` | Histogram | Network in-flight time (TX done → RX recv start) |
+| `vllm_omni:transfer_size_bytes` | Histogram | Per-transfer payload size in bytes |
+| `vllm_omni:transfer_tx_s` | Histogram | Sender-side time (serialize + submit to connector) |
+| `vllm_omni:transfer_rx_s` | Histogram | Receiver-side time (recv + deserialize) |
+| `vllm_omni:transfer_in_flight_s` | Histogram | Network in-flight time (TX done → RX recv start) |
 
 ## vLLM Engine Metrics (`vllm:`)
 
@@ -83,9 +107,9 @@ For the full list of upstream metrics, see [the vLLM docs](https://github.com/vl
 
 | Metric group | Multi-stage LLM (Qwen3-Omni) |
 |---|---|
-| `vllm:omni_` request tracking + latency | With `--log-stats` |
-| `vllm:omni_` audio modality | With `--log-stats`, if pipeline has a talker stage |
-| `vllm:omni_` transfer | With `--log-stats`, if pipeline has ≥ 2 stages |
+| `vllm_omni:` request tracking + latency | With `--log-stats` |
+| `vllm_omni:` audio modality | With `--log-stats`, if pipeline has a talker stage |
+| `vllm_omni:` transfer | With `--log-stats`, if pipeline has ≥ 2 stages |
 | `vllm:` engine metrics (per `(stage, replica)`) | With `--log-stats` |
 | `vllm:` MFU metrics | With `--log-stats --enable-mfu-metrics` |
 
@@ -94,5 +118,5 @@ For the full list of upstream metrics, see [the vLLM docs](https://github.com/vl
 - All time-bearing metrics use the `_s` suffix (values in seconds). Buckets are `SECONDS_BUCKETS` for e2e / generation-style values and `SECONDS_FAST_BUCKETS` (1 ms → 60 s) for the fine-grained transfer and audio-underrun values.
 - Counters use the `_total` suffix (auto-appended by `prometheus_client`).
 - Sizes use the `_bytes` suffix.
-- All omni-specific families are prefixed `vllm:omni_`. The upstream `unregister_vllm_metrics()` function is monkey-patched (see `vllm_omni/patch.py`) to a scoped version that still strips upstream `vllm:*` collectors so multi-engine init within one process does not crash on duplicate registration, but preserves anything prefixed `vllm:omni_` / `vllm_omni`.
-- Text and audio first-output use distinct families (`vllm:time_to_first_token_seconds` reused from upstream for text; `vllm:omni_audio_ttfp_s` for audio) rather than a single metric with a `modality` label.
+- All omni-specific families are prefixed `vllm_omni:`. The upstream `unregister_vllm_metrics()` function is monkey-patched (see `vllm_omni/patch.py`) to a scoped version that still strips upstream `vllm:*` collectors so multi-engine init within one process does not crash on duplicate registration, but preserves anything prefixed `vllm_omni:`.
+- Text and audio first-output use distinct families (`vllm:time_to_first_token_seconds` reused from upstream for text; `vllm_omni:audio_ttfp_s` for audio) rather than a single metric with a `modality` label.

docs/user_guide/diffusion/parallelism/overview.md

@@ -10,7 +10,7 @@ This guide covers the parallelism methods in vLLM-Omni for speeding up diffusion
 | **[Sequence Parallelism](sequence_parallel.md)**   | Splits sequence dimension across GPUs (Ulysses-SP, Ring-Attention, or hybrid) for high-resolution images and videos |
 | **[CFG-Parallel](cfg_parallel.md)**                | Runs CFG positive/negative branches on separate GPUs for ~1.8x speedup on guided generation                         |
 | **[Pipeline Parallelism](pipeline_parallel.md)**   | Splits the denoising transformer block-wise across sequential GPU stages to reduce per-GPU model memory             |
-| **[VAE Parallelism](vae_parallelism.md)** | Distributes VAE decode spatially across GPUs to reduce peak VAE memory                                              |
+| **[VAE Parallel](vae_parallel.md)** | Distributes VAE decode spatially across GPUs to reduce peak VAE memory                                              |
 | **[HSDP](hsdp.md)**                                | Shards full model weights via PyTorch FSDP2 to enable large-model inference on memory-constrained GPUs              |
 | **[Expert Parallelism](expert_parallel.md)**       | Shards MoE expert blocks across GPUs for MoE models (e.g. HunyuanImage3.0)                                          |
 

docs/user_guide/diffusion/parallelism/vae_parallel.md

@@ -1,4 +1,4 @@
-# VAE Parallelism Guide
+# VAE Parallel Guide
 
 
 ## Table of Content
@@ -113,7 +113,7 @@ In `DiffusionParallelConfig`:
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
 | `vae_patch_parallel_size` | int | 1 | Number of GPUs for VAE patch/tile parallelism. Set to 2 or higher to enable. Should typically match `tensor_parallel_size` as they share the same process group. |
-| `vae_parallel_mode` | str | `"tile"` | VAE parallel decode strategy: `"tile"` (default tile/patch parallel decode), `"spatial_shard_height"`, or `"spatial_shard_width"` (spatially-sharded decode, Wan only). See [Spatially-Sharded Decode](#spatially-sharded-decode-wan). |
+| `vae_parallel_mode` | str | `"tile"` | VAE parallel decode strategy: `"tile"`, `"auto"` (Wan request-shape selection), `"spatial_shard_height"`, or `"spatial_shard_width"`. Spatial sharding is currently Wan-only. See [Spatially-Sharded Decode](#spatially-sharded-decode-wan). |
 
 Additional requirements:
 
@@ -128,7 +128,7 @@ Additional requirements:
 
 ## Spatially-Sharded Decode (Wan)
 
-The default `vae_parallel_mode="tile"` distributes whole tiles across ranks. For the **Wan** VAE there is an alternative decode strategy, **spatially-sharded decode**, selected via `vae_parallel_mode="spatial_shard_height"` or `vae_parallel_mode="spatial_shard_width"`.
+The default `vae_parallel_mode="tile"` keeps patch/tile parallel decode unchanged. For Wan, set `vae_parallel_mode="auto"` to select spatial-shard decode only after a request is already large enough to enter Diffusers' tiled-decode path and the VAE uses the full DiT decode group. Auto mode splits the longer eligible spatial axis to reduce halo traffic. Direct-decode requests and partial groups fall back to tile decode. Use `"spatial_shard_height"` or `"spatial_shard_width"` to force an axis.
 
 Instead of assigning independent tiles to ranks, spatial-shard decode shards the decoder feature maps along the height (`spatial_shard_height`) or width (`spatial_shard_width`) dimension and exchanges halo rows/columns between neighboring ranks around the spatial convolutions. This keeps the receptive field correct across shard boundaries, so the result matches the single-GPU decode within numerical tolerance.
 
@@ -141,7 +141,7 @@ omni = Omni(
     parallel_config=DiffusionParallelConfig(
         tensor_parallel_size=2,
         vae_patch_parallel_size=2,               # must match the DiT group size
-        vae_parallel_mode="spatial_shard_width", # or "spatial_shard_height"
+        vae_parallel_mode="auto",                # opt in to Wan request-shape selection
     ),
 )
 ```
@@ -152,14 +152,14 @@ Or from the CLI / serving entrypoint:
 vllm serve Wan-AI/Wan2.1-T2V-1.3B-Diffusers --omni \
     --tensor-parallel-size 2 \
     --vae-patch-parallel-size 2 \
-    --vae-parallel-mode spatial_shard_width
+    --vae-parallel-mode auto
 ```
 
 **Constraints and behavior:**
 
-- Spatial-shard decode is **decode-only** and currently implemented for the **Wan** VAE. Other models ignore `spatial_shard_*` modes.
-- It requires `vae_patch_parallel_size` to **match the DiT process group size**. If it does not, the VAE logs a warning and **falls back to tile-parallel decode** at runtime.
-- `spatial_shard_height` and `spatial_shard_width` are mutually exclusive for a given VAE instance (the decoder is patched in place for a single split dimension).
+- Spatial-shard decode is **decode-only** and currently implemented for the **Wan** VAE. Other VAE families retain their existing mode handling and may reject explicit Wan spatial modes.
+- It requires `vae_patch_parallel_size` to **match the DiT process group size**. `auto` quietly falls back to tile decode for a partial group; an explicitly forced spatial mode logs a warning.
+- The decoder's context-aware wrappers retain the direct path, so one process can alternate small/large and portrait/landscape requests safely.
 
 For end-to-end latency/throughput, launch serving with the desired `vae_parallel_mode` and use the existing diffusion serving benchmark:
 

docs/user_guide/diffusion_features.md

@@ -61,7 +61,7 @@ Memory optimization methods help reduce GPU memory usage, enabling inference on
 |--------|-------------|----------|
 | **[CPU Offload](diffusion/cpu_offload.md)** | Offloads model components to CPU memory | Limited VRAM, large models on consumer GPUs |
 | **[Quantization](quantization/overview.md)** | Reduces transformer stages from BF16 to FP8/INT8/etc. | Limited VRAM, minimal accuracy loss    |
-| **[VAE Parallelism](diffusion/parallelism/vae_parallelism.md)** | Distributes VAE decode work across GPUs | High-resolution generation with reduced VAE memory peak |
+| **[VAE Parallel](diffusion/parallelism/vae_parallel.md)** | Distributes VAE decode work across GPUs | High-resolution generation with reduced VAE memory peak |
 
 ### Extensions
 

docs/user_guide/examples/offline_inference/text_to_video.md

@@ -12,6 +12,7 @@ For backend selection and SageAttention usage, see the [Diffusion Attention Back
 | Model | Default Resolution | Default Frames | Default Steps | Guidance | VRAM (BF16) |
 |---|---|---|---|---|---|
 | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 720x1280 | 81 | 40 | 4.0 | ~60 GiB |
+| `robbyant/lingbot-video-dense-1.3b` / `robbyant/lingbot-video-moe-30b-a3b` | 192x320 | 9 | 2 | 3.0 | ~68 GiB (MoE smoke) |
 | `Lightricks/LTX-2` | 512x768 | 121 | 40 | video 3.0 / audio 7.0 | Model-dependent |
 | `hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v` | 480x832 | 121 | 50 | 6.0 | 1×A100 80GB |
 | `hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_t2v` | 720x1280 | 121 | 50 | 6.0 | FP8 + VAE tiling required |
@@ -35,6 +36,25 @@ python text_to_video.py \
   --output t2v_out.mp4
 ```
 
+### LingBot-Video
+
+The shared runner recognizes both official LingBot checkpoint IDs and selects
+the native `LingBotVideoPipeline` automatically:
+
+```bash
+python text_to_video.py \
+  --model robbyant/lingbot-video-dense-1.3b \
+  --prompt "a robotic arm picks up a red block" \
+  --height 192 --width 320 --num-frames 9 --num-inference-steps 2 \
+  --guidance-scale 3.0 --flow-shift 3.0 --fps 24 \
+  --output lingbot_t2v.mp4
+```
+
+Use `robbyant/lingbot-video-moe-30b-a3b` for the MoE checkpoint. For local
+paths without `lingbot` in the directory name, add
+`--model-class-name LingBotVideoPipeline`. Model-only options use
+`--extra-body`, for example `'{"batch_cfg": true, "output_type": "np"}'`.
+
 ### LTX-2
 
 ```bash

docs/user_guide/quantization/overview.md

@@ -12,20 +12,20 @@ For the internal architecture and backend extension points, see the
 
 | Mode | Guide | Description | Methods |
 |------|-------|-------------|---------|
-| Online quantization | [Online Quantization](online.md) | vLLM-Omni computes quantized weights and scales while loading the model. | FP8 W8A8, Int8 W8A8, MXFP8 W8A8, MXFP4 W4A4 |
+| Online quantization | [Online Quantization](online.md) | vLLM-Omni computes quantized weights and scales while loading the model. | FP8 W8A8, Int8 W8A8, BitsAndBytes W4, MXFP8 W8A8, MXFP4 W4A4 |
 | Runtime attention quantization | [Quantized KV Cache](quantized_kvcache.md) | vLLM-Omni dynamically quantizes eligible diffusion Flash Attention tensors during inference. | FP8 FA |
-| Pre-quantized checkpoints | Method-specific guides | The checkpoint or an offline quantizer provides quantized weights and scales before serving. | ModelOpt, GGUF, AutoRound, msModelSlim, serialized Int8, offline MXFP8, offline MXFP4 DualScale |
+| Pre-quantized checkpoints | Method-specific guides | The checkpoint or an offline quantizer provides quantized weights and scales before serving. | ModelOpt, AutoRound, msModelSlim, serialized Int8, offline MXFP8, offline MXFP4 DualScale |
 
 ## Hardware Support
 
-| Device | FP8 W8A8 | Int8 W8A8 | ModelOpt | MXFP8 W8A8 | MXFP4 W4A4 | GGUF | AutoRound | msModelSlim |
-|--------|----------|-----------|----------|------------|------------|------|-----------|-------------|
-| NVIDIA Blackwell GPU (SM 100+) | ✅ | ✅ | ✅ | ⭕ | ⭕ | ✅ | ✅ | ❌ |
-| NVIDIA Ada/Hopper GPU (SM 89+) | ✅ | ✅ | ✅ | ⭕ | ⭕ | ✅ | ✅ | ❌ |
-| NVIDIA Ampere GPU (SM 80+) | ✅ | ✅ | ⭕ | ⭕ | ⭕ | ✅ | ✅ | ❌ |
-| AMD ROCm | ⭕ | ⭕ | ⭕ | ⭕ | ✅ | ⭕ | ⭕ | ❌ |
-| Intel XPU | ⭕ | ⭕ | ⭕ | ⭕ | ⭕ | ⭕ | ✅ | ❌ |
-| Ascend NPU | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
+| Device | FP8 W8A8 | Int8 W8A8 | BitsAndBytes W4 | ModelOpt | MXFP8 W8A8 | MXFP4 W4A4 | AutoRound | msModelSlim |
+|--------|----------|-----------|------------------|----------|------------|------------|-----------|-------------|
+| NVIDIA Blackwell GPU (SM 100+) | ✅ | ✅ | ✅ | ✅ | ⭕ | ⭕ | ✅ | ❌ |
+| NVIDIA Ada/Hopper GPU (SM 89+) | ✅ | ✅ | ✅ | ✅ | ⭕ | ⭕ | ✅ | ❌ |
+| NVIDIA Ampere GPU (SM 80+) | ✅ | ✅ | ✅ | ⭕ | ⭕ | ⭕ | ✅ | ❌ |
+| AMD ROCm | ⭕ | ⭕ | ❌ | ⭕ | ⭕ | ✅ | ⭕ | ❌ |
+| Intel XPU | ⭕ | ⭕ | ❌ | ⭕ | ⭕ | ⭕ | ✅ | ❌ |
+| Ascend NPU | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
 
 Legend: `✅` supported, `❌` unsupported, `⭕` not verified in this
 guide. FP8 on Ampere may use a weight-only path where available.
@@ -43,10 +43,10 @@ otherwise.
 |--------|-------|------|----------------|--------|
 | FP8 W8A8 | [FP8](fp8.md) | Online W8A8 or checkpoint FP8 | Qwen-Image; Wan2.2 is not validated | Validated for Qwen-Image family and other DiT models |
 | Int8 W8A8 | [Int8](int8.md) | Online or serialized W8A8 | Qwen-Image; Wan2.2 is not validated | Validated for Qwen-Image and Z-Image |
+| BitsAndBytes W4 | [BitsAndBytes](bitsandbytes.md) | Online W4 weight-only | Z-Image-Turbo; Qwen-Image/Wan2.2 not validated | Validated for Z-Image-Turbo |
 | ModelOpt | [ModelOpt](modelopt.md) | Pre-quantized FP8 checkpoints | Qwen-Image, Z-Image, FLUX.2, HunyuanImage-3.0 | Validated for ModelOpt FP8 diffusion checkpoints |
 | MXFP8 W8A8 | [MXFP8](mxfp8.md) | Online W8A8 or offline pre-quantized | Wan2.2-T2V-A14B, I2V-A14B, TI2V-5B | Ascend NPU only; validated for Wan2.2 |
 | MXFP4 W4A4 | [MXFP4](mxfp4.md) | `mxfp4`: online single-scale only; `mxfp4_dualscale`: online or offline dual-scale (offline recommended) | Wan2.2-T2V-A14B, I2V-A14B | Ascend NPU only; validated for Wan2.2 A14B cascade models; TI2V-5B not supported; offline `mxfp4_dualscale` uses calibrated `mul_scale` for best accuracy |
-| GGUF | [GGUF](gguf.md) | Pre-quantized transformer weights | Qwen-Image | Validated where a model-specific GGUF adapter exists |
 | AutoRound | [AutoRound](autoround.md) | Pre-quantized W4A16 checkpoints | FLUX.1-dev; Qwen-Image/Wan2.2 not validated | Checkpoint-driven |
 | msModelSlim | [msModelSlim](msmodelslim.md) | Pre-quantized Ascend checkpoints | Wan2.2 recipe; HunyuanImage-3.0 inference target | Ascend/NPU path |
 
@@ -61,9 +61,9 @@ in BF16 unless the model guide explicitly adds support.
 |--------|-------|-------|----------------|--------|
 | ModelOpt | [ModelOpt](modelopt.md) | Thinker or language-model checkpoint config | Qwen3-Omni thinker | ModelOpt checkpoint path |
 | Int8 | [Int8](int8.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
+| BitsAndBytes | [BitsAndBytes](bitsandbytes.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
 | MXFP8 | [MXFP8](mxfp8.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
 | MXFP4 | [MXFP4](mxfp4.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
-| GGUF | [GGUF](gguf.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
 | AutoRound | [AutoRound](autoround.md) | Thinker or language-model checkpoint config | Qwen2.5-Omni, Qwen3-Omni | Supported through AutoRound checkpoints |
 | msModelSlim | [msModelSlim](msmodelslim.md) | Not currently validated for omni/TTS stages | Qwen3-Omni, Qwen3-TTS | Not validated |
 
@@ -76,10 +76,10 @@ attached to the intended stage rather than applied globally.
 |--------|-------|-------|----------------|--------|
 | FP8 | [FP8](fp8.md) | Stage-specific DiT or transformer module | BAGEL, GLM-Image | Requires model-specific validation |
 | Int8 | [Int8](int8.md) | Stage-specific DiT or transformer module | BAGEL, GLM-Image | Requires model-specific validation |
+| BitsAndBytes | [BitsAndBytes](bitsandbytes.md) | Stage-specific transformer or DiT module | BAGEL, GLM-Image | Not validated |
 | ModelOpt | [ModelOpt](modelopt.md) | Checkpoint-defined diffusion stage | BAGEL, GLM-Image | Requires model-specific validation |
 | MXFP8 | [MXFP8](mxfp8.md) | Stage-specific DiT or transformer module | BAGEL, GLM-Image | Not validated |
 | MXFP4 | [MXFP4](mxfp4.md) | Stage-specific DiT or transformer module | BAGEL, GLM-Image | Not validated |
-| GGUF | [GGUF](gguf.md) | Stage-specific transformer weights | BAGEL, GLM-Image | No validated adapter listed |
 | AutoRound | [AutoRound](autoround.md) | Checkpoint-defined stage | BAGEL, GLM-Image | No validated checkpoint listed |
 | msModelSlim | [msModelSlim](msmodelslim.md) | Ascend-generated stage weights | GLM-Image | Requires model-specific adaptation |
 
@@ -106,7 +106,7 @@ config = build_quant_config({
 
 | Component | Default quantized? | Notes |
 |-----------|--------------------|-------|
-| Diffusion transformer | Yes | Primary target for FP8, Int8, ModelOpt, MXFP8, MXFP4, GGUF, AutoRound, and msModelSlim |
+| Diffusion transformer | Yes | Primary target for FP8, Int8, BitsAndBytes, ModelOpt, MXFP8, MXFP4, AutoRound, and msModelSlim |
 | Text encoder | No | Keep BF16 unless a method-specific guide documents support |
 | VAE | No | Keep BF16; storage-only paths are method-specific |
 | Scheduler/tokenizer | No | Loaded from the base model repository |
@@ -139,8 +139,8 @@ from vllm_omni.quantization import build_quant_config
 
 build_quant_config("fp8")
 build_quant_config({"method": "fp8", "activation_scheme": "static"})
+build_quant_config("bitsandbytes")
 build_quant_config("auto-round", bits=4, group_size=128)
-build_quant_config({"method": "gguf", "gguf_model": "/path/to/model.gguf"})
 build_quant_config({"transformer": {"method": "fp8"}, "vae": None})
 build_quant_config(None)
 ```

examples/offline_inference/aura_omni/end2end.py

@@ -131,7 +131,7 @@ def build_sampling_params() -> list[SamplingParams]:
 
 
 def save_stage_output(stage_outputs, output_dir: str) -> None:
-    output = stage_outputs.request_output
+    output = stage_outputs
     request_id = output.request_id
     if stage_outputs.final_output_type == "text":
         text_output = output.outputs[0].text

examples/offline_inference/covo_audio/end2end.py

@@ -59,7 +59,7 @@ def main(args):
 
     omni = Omni(
         model=args.model_name,
-        stage_configs_path=args.stage_configs_path,
+        deploy_config=args.deploy_config,
         log_stats=args.log_stats,
         stage_init_timeout=args.stage_init_timeout,
         batch_timeout=args.batch_timeout,
@@ -104,7 +104,7 @@ def main(args):
     os.makedirs(output_dir, exist_ok=True)
 
     for stage_outputs in omni_outputs:
-        output = stage_outputs.request_output
+        output = stage_outputs
         if stage_outputs.final_output_type == "text":
             request_id = output.request_id
             text_output = output.outputs[0].text
@@ -167,10 +167,10 @@ def parse_args():
         help="Sampling rate for audio loading (default: 16000).",
     )
     parser.add_argument(
-        "--stage-configs-path",
+        "--deploy-config",
         type=str,
         default=None,
-        help="Path to stage configs YAML file.",
+        help="Path to a deploy config YAML file.",
     )
     parser.add_argument(
         "--log-stats",

examples/offline_inference/custom_pipeline/image_to_image/image_edit.py

@@ -228,7 +228,7 @@ async def main():
     if not outputs:
         raise ValueError("No output produced from omni.generate()")
 
-    first_out = outputs[0].request_output
+    first_out = outputs[0]
     req_out: OmniRequestOutput = first_out
 
     # Verify trajectory data (from custom pipeline)

examples/offline_inference/dynin_omni/end2end.py

@@ -741,7 +741,7 @@ def make_mmu_prompt(
 
 def iter_mm_outputs(outputs: list[Any]):
     for omni_out in outputs:
-        req_out = getattr(omni_out, "request_output", None)
+        req_out = omni_out
         req_list = req_out if isinstance(req_out, list) else [req_out]
         for item in req_list:
             if item is None:
@@ -876,7 +876,7 @@ def parse_args(repo_root: Path) -> argparse.Namespace:
     parser.add_argument("--task", type=str, required=True, choices=TASK_CHOICES)
     parser.add_argument("--model", type=str, required=True, help="HF repo id or local model directory.")
     parser.add_argument(
-        "--stage-config-path",
+        "--deploy-config",
         type=str,
         default=str(repo_root / "vllm_omni/deploy/dynin_omni.yaml"),
         help="Path to deploy config yaml.",
@@ -1394,8 +1394,8 @@ def main() -> None:
 
     from vllm_omni.entrypoints.omni import Omni
 
-    stage_config_path = str(Path(args.stage_config_path).expanduser())
-    omni = Omni(model=model_source, stage_configs_path=stage_config_path, dtype=args.dtype)
+    deploy_config_path = str(Path(args.deploy_config).expanduser())
+    omni = Omni(model=model_source, deploy_config=deploy_config_path, dtype=args.dtype)
     sampling_params_list = [
         SamplingParams(max_tokens=int(args.max_tokens_per_stage), temperature=0.0, top_p=1.0, detokenize=False)
         for _ in range(omni.num_stages)

examples/offline_inference/hunyuan_image3/end2end.py

@@ -73,7 +73,6 @@ def parse_args():
     )
     parser.add_argument("--sys-type", type=str, default=None, help="Override system prompt type.")
     parser.add_argument("--deploy-config", type=str, default=None, help="Custom deploy YAML path.")
-    parser.add_argument("--stage-configs-path", type=str, default=None, help="Custom legacy stage config YAML path.")
     parser.add_argument(
         "--stream",
         action="store_true",
@@ -146,13 +145,7 @@ def main():
     else:
         bot_task = args.bot_task
 
-    if args.deploy_config is not None and args.stage_configs_path is not None:
-        raise ValueError("--deploy-config and --stage-configs-path are mutually exclusive.")
-
-    deploy_config = args.deploy_config
-    stage_configs_path = args.stage_configs_path
-    if deploy_config is None and stage_configs_path is None:
-        deploy_config = _MODALITY_DEFAULT_DEPLOY_CONFIG[args.modality]
+    deploy_config = args.deploy_config or _MODALITY_DEFAULT_DEPLOY_CONFIG[args.modality]
 
     omni_kwargs = {
         "model": args.model,
@@ -168,10 +161,7 @@ def main():
 
     if additional_config is not None:
         omni_kwargs["additional_config"] = additional_config
-    if deploy_config is not None:
-        omni_kwargs["deploy_config"] = deploy_config
-    else:
-        omni_kwargs["stage_configs_path"] = stage_configs_path
+    omni_kwargs["deploy_config"] = deploy_config
 
     omni = Omni(**omni_kwargs)
 
@@ -272,10 +262,7 @@ def main():
     print(f"  Modality: {args.modality}")
     print(f"  Prompt task: {task}")
     print(f"  Bot task: {bot_task}")
-    if deploy_config is not None:
-        print(f"  Deploy config: {deploy_config}")
-    else:
-        print(f"  Stage config: {stage_configs_path}")
+    print(f"  Deploy config: {deploy_config}")
     print(f"  Num stages: {omni.num_stages}")
     if args.modality in ("text2img", "img2img"):
         print(f"  Inference steps: {args.steps}")
@@ -303,7 +290,7 @@ def main():
     )
     img_idx = 0
     for req_output in omni_outputs:
-        ro = getattr(req_output, "request_output", None)
+        ro = req_output
         stage_id = getattr(req_output, "stage_id", None)
 
         # AR stage text — each CompletionOutput.text is already a delta when

examples/offline_inference/image_to_image/image_edit.py

@@ -621,7 +621,7 @@ def main():
         images = getattr(output, "images", None)
         if images:
             break
-        req_out = getattr(output, "request_output", None)
+        req_out = output
         images = getattr(req_out, "images", None) if req_out is not None else None
         if images:
             break

examples/offline_inference/image_to_video/image_to_video.py

@@ -646,8 +646,8 @@ def main():
         if frames.multimodal_output and "audio" in frames.multimodal_output:
             audio = frames.multimodal_output["audio"]
             audio_sample_rate = frames.multimodal_output.get("audio_sample_rate", audio_sample_rate)
-        if frames.is_pipeline_output and frames.request_output is not None:
-            inner_output = frames.request_output
+        if frames.is_pipeline_output and frames is not None:
+            inner_output = frames
             if isinstance(inner_output, OmniRequestOutput):
                 if inner_output.multimodal_output and "audio" in inner_output.multimodal_output:
                     audio = inner_output.multimodal_output["audio"]

examples/offline_inference/mammothmodal2_preview/run_mammothmoda2_image_summarize.py

@@ -0,0 +1,117 @@
+"""
+Offline inference example: MammothModa2 image summarization (single AR stage).
+
+Example:
+  uv run python examples/offline_inference/mammothmodal2_preview/run_mammothmoda2_image_summarize.py \
+    --model path/to/MammothModa2-Preview \
+    --deploy-config vllm_omni/deploy/mammoth_moda2_ar.yaml \
+    --image /path/to/input.jpg \
+    --question "Please summarize the content of this image."
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+
+from PIL import Image
+from vllm import SamplingParams
+from vllm.multimodal.image import convert_image_mode
+
+from vllm_omni import Omni
+
+DEFAULT_SYSTEM = "You are a helpful assistant."
+DEFAULT_QUESTION = "Please summarize the content of this image."
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="MammothModa2 image summarization (offline, AR only).")
+    parser.add_argument("--model", type=str, required=True, help="Path to model directory or model id.")
+    parser.add_argument("--deploy-config", type=str, required=True, help="Path to deploy config yaml.")
+    parser.add_argument("--image", type=str, required=True, help="Path to input image.")
+    parser.add_argument("--question", type=str, default=DEFAULT_QUESTION, help="Question/instruction for the model.")
+    parser.add_argument("--system", type=str, default=DEFAULT_SYSTEM, help="System prompt.")
+    parser.add_argument(
+        "--max-tokens",
+        type=int,
+        default=512,
+        help="Max new tokens to generate.",
+    )
+    parser.add_argument("--temperature", type=float, default=0.2)
+    parser.add_argument("--top-p", type=float, default=0.9)
+    parser.add_argument("--seed", type=int, default=42)
+    parser.add_argument("--trust-remote-code", action="store_true")
+    parser.add_argument(
+        "--enable-diffusion-pipeline-profiler",
+        action="store_true",
+        help="Enable diffusion pipeline profiler to display stage durations.",
+    )
+    return parser.parse_args()
+
+
+def build_prompt(system: str, question: str) -> str:
+    return (
+        f"<|im_start|>system\n{system}<|im_end|>\n"
+        "<|im_start|>user\n"
+        "<|vision_start|><|image_pad|><|vision_end|>"
+        f"{question}<|im_end|>\n"
+        "<|im_start|>assistant\n"
+    )
+
+
+def main() -> None:
+    args = parse_args()
+
+    if not os.path.exists(args.image):
+        raise FileNotFoundError(f"Image file not found: {args.image}")
+
+    pil_image = Image.open(args.image)
+    image_data = convert_image_mode(pil_image, "RGB")
+    prompt = build_prompt(args.system, args.question)
+
+    omni = Omni(
+        model=args.model,
+        deploy_config=args.deploy_config,
+        trust_remote_code=args.trust_remote_code,
+        enable_diffusion_pipeline_profiler=args.enable_diffusion_pipeline_profiler,
+    )
+    try:
+        sp = SamplingParams(
+            temperature=float(args.temperature),
+            top_p=float(args.top_p),
+            top_k=-1,
+            max_tokens=int(args.max_tokens),
+            seed=int(args.seed),
+            detokenize=True,
+        )
+        # NOTE: omni.generate() returns a Generator[OmniRequestOutput, None, None].
+        # Consume it inside the try block so the worker isn't closed early.
+        outputs = list(
+            omni.generate(
+                [
+                    {
+                        "prompt": prompt,
+                        "multi_modal_data": {"image": image_data},
+                        "additional_information": {"omni_task": ["chat"]},
+                    }
+                ],
+                [sp],
+            )
+        )
+    finally:
+        omni.close()
+
+    lines: list[str] = []
+    for stage_outputs in outputs:
+        ro = stage_outputs
+        text = ro.outputs[0].text if getattr(ro, "outputs", None) else str(ro)
+        lines.append(f"request_id: {getattr(ro, 'request_id', 'unknown')}\n")
+        lines.append("answer:\n")
+        lines.append(text.strip() + "\n")
+        lines.append("\n")
+
+    print("\n".join(lines))
+
+
+if __name__ == "__main__":
+    main()

examples/offline_inference/mimo_audio/end2end.py

@@ -297,7 +297,7 @@ def main(args):
     os.makedirs(output_dir, exist_ok=True)
 
     for stage_outputs in omni_outputs:
-        output = stage_outputs.request_output
+        output = stage_outputs
         if stage_outputs.final_output_type == "text":
             request_id = output.request_id
             text_output = output.outputs[0].text

examples/offline_inference/ming_flash_omni/end2end.py

@@ -347,7 +347,7 @@ def main(args):
         omni.start_profile(stages=args.profiler_stages)
 
     for stage_outputs in omni.generate(prompts, sampling_params_list):
-        output = stage_outputs.request_output
+        output = stage_outputs
         if stage_outputs.final_output_type == "text":
             request_id = output.request_id
             text_output = output.outputs[0].text

examples/offline_inference/minicpmo/end2end.py

@@ -452,12 +452,6 @@ def parse_args():
         default=None,
         help="Path to a .txt file with one prompt per line (preferred).",
     )
-    parser.add_argument(
-        "--stage-configs-path",
-        type=str,
-        default=None,
-        help="Path to a stage configs file (deprecated; prefer --deploy-config).",
-    )
     parser.add_argument(
         "--deploy-config",
         type=str,

examples/offline_inference/nemotron_voicechat/end2end.py

@@ -0,0 +1,229 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""Offline speech-to-speech with NVIDIA-NemotronLabs-VoiceChat-11B.
+
+Feeds a 16 kHz user utterance (plus a spoken-style system prompt) through the
+3-stage nemotron_voicechat pipeline and writes the agent's reply as text and a
+22.05 kHz WAV.
+
+Usage:
+    python end2end.py \
+        --checkpoint /path/to/NVIDIA-NemotronLabs-VoiceChat-11B \
+        --wav /path/to/user_question_16k.wav \
+        --output-dir results/nemotron_voicechat
+
+The system prompt defaults to the NeMo reference default. The text tokenizer
+resolves from the ``nvidia/NVIDIA-Nemotron-Nano-9B-v2`` HF id (or the
+``NEMOTRON_VOICECHAT_LLM_PATH`` env var for air-gapped runs).
+
+Sizing the input WAV: the timeline is frame-locked, so the reply budget IS the
+input duration — the acoustic channel trails the text channel, and a reply
+that has not finished when the frames run out is truncated SILENTLY (no error;
+ASR of the WAV will just disagree with the emitted text). Leave generous
+trailing silence after the question (a question ending at ~4.5 s truncated in
+an 8 s WAV but completed cleanly in a 16 s one). The script warns when the
+text channel is still emitting non-PAD tokens near the last frame.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+from pathlib import Path
+
+import numpy as np
+import soundfile as sf
+
+# Warn if the text channel is still speaking within this many final frames
+# (12 frames = ~1 s) — the frame-locked reply likely got cut off.
+TRUNCATION_GUARD_FRAMES = 12
+
+# The NeMo reference offline script's default system prompt.
+DEFAULT_SYSTEM_PROMPT = (
+    "You are an AI voice assistant developed by NVIDIA. "
+    "Your name is NVIDIA Voice Chat. "
+    "Answer in a spoken, conversational style rather than a written one. "
+    "Do not repeat the same sentence over and over again. "
+    "Start the conversation by greeting the user."
+)
+
+
+def load_wav_16k_mono(path: str) -> np.ndarray:
+    wav, sr = sf.read(path, dtype="float32")
+    if wav.ndim > 1:
+        wav = wav.mean(axis=1)
+    if sr != 16000:
+        from vllm.multimodal.audio import resample_audio_scipy
+
+        wav = resample_audio_scipy(wav, orig_sr=sr, target_sr=16000)
+    return wav.astype(np.float32)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--checkpoint", required=True, help="NVIDIA-NemotronLabs-VoiceChat-11B directory")
+    parser.add_argument("--wav", required=True, help="user speech (any sr; resampled to 16 kHz mono)")
+    parser.add_argument("--output-dir", default="results/nemotron_voicechat")
+    parser.add_argument("--system-prompt", default=DEFAULT_SYSTEM_PROMPT)
+    parser.add_argument(
+        "--stage-configs-path",
+        default=None,
+        help=(
+            "Deploy yaml override. Pass vllm_omni/deploy/nemotron_labs_voicechat_streaming.yaml "
+            "for the async-chunk streaming pipeline (stages overlap; audio starts decoding "
+            "before the thinker finishes). Default: the bundled non-streaming yaml (bit-exact "
+            "parity path)."
+        ),
+    )
+    args = parser.parse_args()
+
+    if not Path(args.wav).is_file():
+        raise FileNotFoundError(f"Input WAV not found: {args.wav}")
+    ckpt = Path(args.checkpoint)
+    raw_cfg = json.loads((ckpt / "config.json").read_text())
+    stt_cfg = raw_cfg["model"]["stt"]["model"]
+
+    from transformers import AutoTokenizer
+    from vllm.sampling_params import SamplingParams
+
+    from vllm_omni.entrypoints.omni import Omni
+    from vllm_omni.model_executor.models.nemotron_voicechat.nemotron_voicechat_thinker import (
+        compute_acoustic_frame_count,
+    )
+
+    tok_ref = os.environ.get("NEMOTRON_VOICECHAT_LLM_PATH") or stt_cfg.get(
+        "pretrained_llm", "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
+    )
+    tokenizer = AutoTokenizer.from_pretrained(tok_ref, trust_remote_code=False)
+    bos_id = tokenizer.convert_tokens_to_ids(stt_cfg.get("bos_token", "<s>"))
+    eos_id = tokenizer.convert_tokens_to_ids(stt_cfg.get("eos_token", "</s>"))
+    pad_id = tokenizer.convert_tokens_to_ids(stt_cfg.get("pad_token", "<SPECIAL_12>"))
+    # NeMo encode_system_prompt: [bos] + tokens + [eos].
+    prompt_ids = [bos_id] + tokenizer.encode(args.system_prompt, add_special_tokens=False) + [eos_id]
+
+    wav = load_wav_16k_mono(args.wav)
+    n_frames = compute_acoustic_frame_count(stt_cfg, int(wav.shape[0]))
+    print(f"input: {wav.shape[0]} samples @16kHz -> {n_frames} acoustic frames; prompt {len(prompt_ids)} tokens")
+
+    inputs = {
+        # vLLM prompt = system prompt + one placeholder position (acoustic
+        # frame 0); the thinker replaces every position with fused embeddings.
+        "prompt_token_ids": prompt_ids + [pad_id],
+        "additional_information": {
+            # Plain float list: raw ndarrays do not survive the engine-core
+            # message serialization.
+            "nvc_audio": wav.astype(np.float32).tolist(),
+            "nvc_sr": 16000,
+            "nvc_prompt_token_ids": prompt_ids,
+            "nvc_expected_frames": n_frames,
+        },
+    }
+    sampling_params_list = [
+        # thinker: greedy, frame-locked generation length. NO min_tokens: the
+        # tokenizer's eos_token is <SPECIAL_12> — the frame-locked PAD/silence
+        # token — and vLLM's min_tokens masks "EOS" to -inf, which would forbid
+        # silence and force the agent to babble through the whole utterance.
+        SamplingParams(temperature=0.0, max_tokens=n_frames, ignore_eos=True, seed=0),
+        # talker: placeholder loop, stops on the stage's stop token. 16383:
+        # the 1-token placeholder prompt + max_tokens must fit the stage's
+        # max_model_len (16384).
+        SamplingParams(temperature=0.0, max_tokens=16383, stop_token_ids=[1], detokenize=False, seed=0),
+        # code2wav: single decode step.
+        SamplingParams(temperature=0.0, max_tokens=1, detokenize=False, seed=0),
+    ]
+
+    omni_kwargs: dict = {}
+    if args.stage_configs_path:
+        omni_kwargs["stage_configs_path"] = args.stage_configs_path
+    omni = Omni(model=str(ckpt), trust_remote_code=True, **omni_kwargs)
+    outputs = omni.generate(inputs, sampling_params_list)
+
+    out_dir = Path(args.output_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+    stem = Path(args.wav).stem
+
+    agent_text = None
+    audio = None
+    sr = 22050
+    for stage_outputs in outputs:
+        # One entry per final stage (text from the thinker, audio from code2wav);
+        # mirrors the qwen2_5_omni offline example's collection loop.
+        output = getattr(stage_outputs, "request_output", stage_outputs)
+        final_type = getattr(stage_outputs, "final_output_type", None)
+        completions = getattr(output, "outputs", None) or []
+        if not completions:
+            continue
+        completion = completions[0]
+        if final_type in (None, "text"):
+            text = getattr(completion, "text", None)
+            if text:
+                agent_text = text
+            token_ids = getattr(completion, "token_ids", None)
+            if token_ids:
+                (out_dir / f"{stem}_text_tokens.json").write_text(json.dumps(list(map(int, token_ids))))
+                # Frame lock: the reply budget IS the input length. If the text
+                # channel is still mid-sentence when the frames run out, the
+                # spoken answer is silently truncated — warn instead.
+                tail = list(map(int, token_ids))[-TRUNCATION_GUARD_FRAMES:]
+                if any(t != pad_id for t in tail):
+                    print(
+                        f"WARNING: the text channel emits non-PAD tokens within the last "
+                        f"{TRUNCATION_GUARD_FRAMES} frames ({TRUNCATION_GUARD_FRAMES * 80} ms) — the reply "
+                        "likely ran out of acoustic frames and is cut off mid-sentence. "
+                        "Pad the input WAV with more trailing silence and rerun."
+                    )
+            # Best-effort: dump the function-channel debug timeline if the
+            # engine surfaced the request info on this output.
+            info = getattr(output, "additional_information", None)
+            func_timeline = info.get("nvc_function_tokens") if isinstance(info, dict) else None
+            if func_timeline is not None:
+                values = func_timeline.tolist() if hasattr(func_timeline, "tolist") else list(func_timeline)
+                (out_dir / f"{stem}_function_tokens.json").write_text(json.dumps([int(v) for v in values]))
+                print(f"function-channel timeline: {len(values)} entries")
+        if final_type in (None, "audio"):
+            # multimodal_output is a MultimodalPayload (Mapping) — duck-type it.
+            mm = getattr(completion, "multimodal_output", None)
+            get = getattr(mm, "get", None)
+            if not callable(get):
+                continue
+            candidate = None
+            for key in ("audio", "model_outputs", "wav", "waveform"):
+                value = get(key)
+                if value is not None:
+                    candidate = value
+                    break
+            if isinstance(candidate, list):
+                candidate = candidate[0] if candidate else None
+            if candidate is not None:
+                arr = np.asarray(
+                    candidate.float().cpu() if hasattr(candidate, "cpu") else candidate, dtype=np.float32
+                ).reshape(-1)
+                if arr.size > 0:
+                    audio = arr
+                sr_val = get("sr")
+                if sr_val is not None:
+                    sr_item = sr_val[0] if isinstance(sr_val, list) else sr_val
+                    sr = int(sr_item.item() if hasattr(sr_item, "item") else sr_item)
+
+    if agent_text is not None:
+        # The frame-locked text channel emits PAD tokens on silent frames;
+        # strip them (NeMo's post-inference detokenizer does the same).
+        pad_token = stt_cfg.get("pad_token", "<SPECIAL_12>")
+        agent_text = " ".join(agent_text.replace(pad_token, " ").split())
+        text_path = out_dir / f"{stem}_output.txt"
+        text_path.write_text(agent_text + "\n")
+        print(f"agent text -> {text_path}\n{agent_text}")
+    else:
+        print("WARNING: no agent text in outputs")
+
+    if audio is not None and audio.size > 0:
+        wav_path = out_dir / f"{stem}_output.wav"
+        sf.write(wav_path, audio, sr)
+        print(f"agent audio -> {wav_path} ({audio.size / sr:.2f}s @ {sr} Hz)")
+    else:
+        raise RuntimeError("No audio produced by the code2wav stage.")
+
+
+if __name__ == "__main__":
+    main()

examples/offline_inference/qwen2_5_omni/end2end.py

@@ -392,7 +392,7 @@ def main(args):
     total_requests = len(prompts)
     processed_count = 0
     for stage_outputs in omni_generator:
-        output = stage_outputs.request_output
+        output = stage_outputs
         if stage_outputs.final_output_type == "text":
             request_id = output.request_id
             text_output = output.outputs[0].text

examples/offline_inference/qwen3_omni/end2end.py

@@ -367,7 +367,7 @@ def main(args):
     print(f"query type: {args.query_type}")
 
     for stage_outputs in omni_generator:
-        output = stage_outputs.request_output
+        output = stage_outputs
         if stage_outputs.final_output_type == "text":
             request_id = output.request_id
             text_output = output.outputs[0].text
@@ -484,10 +484,10 @@ def parse_args():
         help="Path to a .txt file with one prompt per line (preferred).",
     )
     parser.add_argument(
-        "--stage-configs-path",
+        "--deploy-config",
         type=str,
         default=None,
-        help="Path to a stage configs file.",
+        help="Path to a deploy config YAML.",
     )
     parser.add_argument(
         "--video-path",

examples/offline_inference/qwen3_omni/end2end_async_chunk.py

@@ -237,7 +237,7 @@ async def run_single_request(
             sampling_params_list=sampling_params_list,
             output_modalities=output_modalities,
         ):
-            output = omni_output.request_output
+            output = omni_output
             if omni_output.final_output_type == "text":
                 if stage_0_first_output_ts is None:
                     stage_0_first_output_ts = time.perf_counter()

examples/offline_inference/speech_to_video/speech_to_video.py

@@ -26,7 +26,6 @@ import torch
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform
 
 
@@ -352,8 +351,8 @@ def main():
         else:
             print("[Profiler] No valid profiling data returned.")
 
-    # Extract output from result
-    output = OmniRequestOutput.unwrap_result(result)
+    # omni.generate() returns a list for sync calls; unwrap single-result list.
+    output = result[0] if isinstance(result, list) else result
 
     if not output.images:
         raise ValueError("No video frames found in OmniRequestOutput.")

examples/offline_inference/step_audio2/end2end.py

@@ -232,15 +232,8 @@ def main(args):
         raise ValueError(f"Unknown query type: {args.query_type}")
 
     # Initialize vLLM-Omni with Step-Audio2
-    # Resolve deploy config path. Keep explicit stage_configs_path for legacy
-    # custom configs, but use bundled deploy configs by default.
-    deploy_config_path = None
-    stage_config_path = None
-    if args.stage_configs_path and args.deploy_config:
-        raise ValueError("--stage-configs-path and --deploy-config are mutually exclusive")
-    if args.stage_configs_path:
-        stage_config_path = args.stage_configs_path
-    elif args.deploy_config:
+    # Resolve an explicit deploy config or use the bundled task default.
+    if args.deploy_config:
         deploy_config_path = args.deploy_config
     else:
         configs_dir = Path(__file__).parent.parent.parent.parent / "vllm_omni/deploy"
@@ -262,10 +255,7 @@ def main(args):
         "ray_address": args.ray_address,
         "trust_remote_code": True,
     }
-    if stage_config_path is not None:
-        omni_kwargs["stage_configs_path"] = stage_config_path
-    else:
-        omni_kwargs["deploy_config"] = deploy_config_path
+    omni_kwargs["deploy_config"] = deploy_config_path
 
     omni_llm = Omni(**omni_kwargs)
 
@@ -345,7 +335,7 @@ def main(args):
     # Process outputs from each stage
     for stage_idx, stage_outputs in enumerate(omni_outputs):
         # request_output may be a single object or a list
-        raw_output = stage_outputs.request_output
+        raw_output = stage_outputs
         outputs_list = raw_output if isinstance(raw_output, list) else [raw_output]
 
         if stage_outputs.final_output_type == "text":
@@ -418,12 +408,6 @@ def parse_args():
             "  - HuggingFace ID: stepfun-ai/Step-Audio-2-mini"
         ),
     )
-    parser.add_argument(
-        "--stage-configs-path",
-        type=str,
-        default=None,
-        help="Path to a legacy stage config YAML file (default: use bundled deploy config)",
-    )
     parser.add_argument(
         "--deploy-config",
         type=str,

examples/offline_inference/text_to_audio/text_to_audio.py

@@ -43,6 +43,7 @@ from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.model_extras import get_extra_body_params, get_model_class_name
+from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform
 
 AUDIOX_TASKS = ("t2a", "t2m", "v2a", "v2m", "tv2a", "tv2m")
@@ -388,9 +389,9 @@ def main():
         raise ValueError("No output generated from omni.generate()")
 
     output = outputs[0]
-    if not hasattr(output, "request_output") or not output.request_output:
+    if not isinstance(output, OmniRequestOutput) or not output:
         raise ValueError("No request_output found in OmniRequestOutput")
-    request_output = output.request_output
+    request_output = output
     if not hasattr(request_output, "multimodal_output"):
         raise ValueError("No multimodal_output found in request_output")
 

examples/offline_inference/text_to_image/gradio_demo.py

@@ -115,7 +115,7 @@ def build_demo(args: argparse.Namespace) -> gr.Blocks:
         )
         images_outputs = []
         for output in outputs:
-            req_out = output.request_output
+            req_out = output
             if not isinstance(req_out, OmniRequestOutput) or not hasattr(req_out, "images"):
                 raise ValueError("Invalid request_output structure or missing 'images' key")
             images = req_out.images

examples/offline_inference/text_to_image/text_to_image.py

@@ -92,12 +92,6 @@ def parse_args() -> argparse.Namespace:
         "meituan-longcat/LongCat-Image, OvisAI/Ovis-Image, "
         "stabilityai/stable-diffusion-3.5-medium, Tongyi-MAI/Z-Image-Turbo and etc.",
     )
-    parser.add_argument(
-        "--stage-configs-path",
-        type=str,
-        default=None,
-        help="[Deprecated] Path to a legacy stage_args-format YAML. Prefer --deploy-config.",
-    )
     parser.add_argument(
         "--deploy-config",
         type=str,
@@ -475,8 +469,6 @@ def main():
         omni_kwargs["tensor_parallel_size"] = args.tensor_parallel_size
     if args.enforce_eager is not None:
         omni_kwargs["enforce_eager"] = args.enforce_eager
-    if args.stage_configs_path:
-        omni_kwargs["stage_configs_path"] = args.stage_configs_path
     if args.deploy_config:
         omni_kwargs["deploy_config"] = args.deploy_config
     if use_nextstep:
@@ -515,8 +507,8 @@ def main():
     print(f"  Image size: {args.width}x{args.height}")
     if args.lora_path:
         print(f"  LoRA: scale={args.lora_scale}")
-    if args.stage_configs_path:
-        print(f"  stage-configs-path: {args.stage_configs_path}")
+    if args.deploy_config:
+        print(f"  deploy-config: {args.deploy_config}")
     print(f"{'=' * 60}\n")
 
     # Build LoRA request when --lora-path is set
@@ -656,7 +648,7 @@ def main():
         images = getattr(output, "images", None)
         if images:
             break
-        req_out = getattr(output, "request_output", None)
+        req_out = output
         images = getattr(req_out, "images", None) if req_out is not None else None
         if images:
             break

examples/offline_inference/text_to_speech/README.md

@@ -372,18 +372,10 @@ python examples/offline_inference/text_to_speech/qwen3_tts/end2end.py \
 Streaming requires `async_chunk: true` in the stage config.
 
 ### Word Timestamps
-Generate a WAV offline and a JSON sidecar with word-level timestamps from
-`Qwen/Qwen3-ForcedAligner-0.6B`:
-```bash
-python examples/offline_inference/text_to_speech/qwen3_tts/word_timestamps.py \
-    --model Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice \
-    --forced-aligner Qwen/Qwen3-ForcedAligner-0.6B \
-    --text "Hello world." \
-    --output-dir /tmp/qwen3_tts_timestamps
-```
-The script writes `qwen3_tts_word_timestamps.wav` and
-`qwen3_tts_word_timestamps.json`. On machines without a local CUDA toolkit,
-set `VLLM_USE_FLASHINFER_SAMPLER=0` to avoid FlashInfer sampler JIT.
+Word-level timestamps are currently a serving-path feature: launch
+`vllm-omni serve` with `--forced-aligner` and request `word_timestamps`
+(see `examples/online_serving/text_to_speech/README.md`). An offline
+example is not provided in this release.
 
 ### Batched decoding
 The Code2Wav stage supports batched decoding through the SpeechTokenizer. Pass multiple prompts via `--txt-prompts` and set `--batch-size` accordingly. To raise `max_num_seqs` on either stage, point `--stage-configs-path` at a stage configs YAML with the desired values (see `vllm_omni/model_executor/stage_configs/` for templates):

examples/offline_inference/text_to_speech/cosyvoice3/end2end.py

@@ -168,7 +168,7 @@ def run_e2e():
     print(f"Received {len(outputs)} outputs.")
     for i, output in enumerate(outputs):
         try:
-            ro = output.request_output
+            ro = output
             if ro is None:
                 print("No request_output found.")
                 continue

examples/offline_inference/text_to_speech/fish_speech/end2end.py

@@ -38,7 +38,7 @@ from vllm_omni.utils.tracking_parser import TrackingArgumentParser
 logger = logging.getLogger(__name__)
 
 DEFAULT_MODEL = "fishaudio/s2-pro"
-DEFAULT_STAGE_CONFIG = os.path.join(
+DEFAULT_DEPLOY_CONFIG = os.path.join(
     os.path.dirname(__file__),
     "..",
     "..",
@@ -152,10 +152,7 @@ def _build_omni_kwargs(args, model_name: str) -> dict:
         log_stats=args.log_stats,
         stage_init_timeout=args.stage_init_timeout,
     )
-    if args.deploy_config:
-        omni_kwargs["deploy_config"] = args.deploy_config
-    else:
-        omni_kwargs["stage_configs_path"] = args.stage_configs_path or DEFAULT_STAGE_CONFIG
+    omni_kwargs["deploy_config"] = args.deploy_config or DEFAULT_DEPLOY_CONFIG
     return omni_kwargs
 
 
@@ -179,7 +176,7 @@ def main(args):
 
     t_start = time.perf_counter()
     for stage_outputs in omni.generate(inputs):
-        request_output = stage_outputs.request_output
+        request_output = stage_outputs
         if request_output is None or not request_output.outputs:
             continue
         _save_wav(
@@ -215,7 +212,7 @@ async def main_streaming(args):
     sample_rate = None
 
     async for stage_output in omni.generate(prompt, request_id=request_id):
-        mm = stage_output.request_output.outputs[0].multimodal_output
+        mm = stage_output.outputs[0].multimodal_output
         if not stage_output.finished:
             t_now = time.perf_counter()
             sample_rate, n = _collect_audio_chunk(mm, all_audio_chunks, sample_rate)
@@ -278,12 +275,6 @@ def parse_args():
         default=None,
         help="Deploy config YAML path (under vllm_omni/deploy/).",
     )
-    parser.add_argument(
-        "--stage-configs-path",
-        type=str,
-        default=None,
-        help="Path to stage configs YAML (legacy, prefer --deploy-config).",
-    )
     parser.add_argument(
         "--output-dir",
         default="output_audio",

examples/offline_inference/text_to_speech/glm_tts/end2end.py

@@ -145,7 +145,7 @@ def main(args):
 
     omni = Omni(
         model=args.model,
-        stage_configs_path=deploy_config_path,
+        deploy_config=deploy_config_path,
         log_stats=args.log_stats,
         stage_init_timeout=args.stage_init_timeout,
     )

examples/offline_inference/text_to_speech/moss_tts_nano/end2end.py

@@ -137,7 +137,7 @@ def main(args) -> None:
     params_list = sampling_params
 
     for stage_outputs in omni.generate(inputs, params_list):
-        for i, req_output in enumerate(stage_outputs.request_output):
+        for i, req_output in enumerate(stage_outputs):
             for j, out in enumerate(req_output.outputs):
                 mm = out.multimodal_output
                 if mm is None:

examples/offline_inference/text_to_speech/omnivoice/end2end.py

@@ -135,7 +135,7 @@ def run_e2e():
     print(f"Received {len(outputs)} outputs.")
     for i, output in enumerate(outputs):
         try:
-            ro = output.request_output
+            ro = output
             if ro is None:
                 print("No request_output found.")
                 continue

examples/offline_inference/text_to_speech/qwen3_tts/end2end.py

@@ -391,7 +391,7 @@ def main(args):
     for batch_start in range(0, len(inputs), batch_size):
         batch = inputs[batch_start : batch_start + batch_size]
         for stage_outputs in omni.generate(batch):
-            output = stage_outputs.request_output
+            output = stage_outputs
             _save_wav(output_dir, output.request_id, output.outputs[0].multimodal_output)
 
 
@@ -412,7 +412,7 @@ async def main_streaming(args):
         t_prev = t_start
         chunk_idx = 0
         async for stage_output in omni.generate(prompt, request_id=request_id):
-            mm = stage_output.request_output.outputs[0].multimodal_output
+            mm = stage_output.outputs[0].multimodal_output
             if not stage_output.finished:
                 t_now = time.perf_counter()
                 audio = mm.get("audio")
@@ -490,10 +490,10 @@ def parse_args():
         help="Path to a .txt file with one prompt per line (preferred).",
     )
     parser.add_argument(
-        "--stage-configs-path",
+        "--deploy-config",
         type=str,
         default=None,
-        help="Path to a stage configs file.",
+        help="Path to a deploy config YAML.",
     )
     parser.add_argument(
         "--audio-path",

examples/offline_inference/text_to_speech/qwen3_tts/word_timestamps.py

@@ -116,7 +116,7 @@ def main(args: Any) -> None:
     prompt = _build_custom_voice_input(args)
     final_output = None
     for stage_outputs in omni.generate([prompt]):
-        final_output = stage_outputs.request_output
+        final_output = stage_outputs
     if final_output is None:
         raise RuntimeError("Qwen3-TTS did not produce an output.")
 

examples/offline_inference/text_to_speech/soulxsinger/benchmark.py

@@ -38,7 +38,7 @@ def _stage_durations_from_output(outputs) -> dict[str, float]:
     durations = getattr(omni_out, "stage_durations", None) or {}
     if durations:
         return dict(durations)
-    ro = getattr(omni_out, "request_output", None)
+    ro = omni_out
     if ro is not None:
         return dict(getattr(ro, "stage_durations", None) or {})
     return {}

examples/offline_inference/text_to_speech/soulxsinger/end2end.py

@@ -99,7 +99,7 @@ def extract_audio(outputs) -> tuple[np.ndarray, int]:
     out = outputs[0]
     mm = getattr(out, "multimodal_output", None)
     if mm is None:
-        ro = getattr(out, "request_output", None)
+        ro = out
         if ro is not None:
             mm = getattr(ro, "multimodal_output", None)
             if mm is None and ro.outputs:

examples/offline_inference/text_to_video/text_to_video.md

@@ -8,6 +8,7 @@ A unified script for text-to-video generation. Supports multiple models with mod
 |---|---|---|---|---|---|
 | `Wan-AI/Wan2.1-VACE-1.3B-diffusers` | 480x832 | 81 | 30 | 5.0 | ~20 GiB (RTX 5090, VAE tiling) |
 | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 720x1280 | 81 | 40 | 4.0 | ~60 GiB |
+| `robbyant/lingbot-video-dense-1.3b` / `robbyant/lingbot-video-moe-30b-a3b` | 192x320 | 9 | 2 | 3.0 | ~68 GiB (MoE smoke) |
 | `Lightricks/LTX-2` | 512x768 | 121 | 40 | video 3.0 / audio 7.0 | Model-dependent |
 | `hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v` | 480x832 | 121 | 50 | 6.0 | 1×A100 80GB |
 | `hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_t2v` | 720x1280 | 121 | 50 | 6.0 | FP8 + VAE tiling required |
@@ -57,18 +58,31 @@ python text_to_video.py \
 
 ### LingBot-Video
 
-The dedicated LingBot example explicitly selects the video output modality:
+The shared runner recognizes the official dense and MoE checkpoint IDs, selects
+`LingBotVideoPipeline`, and builds the canonical video request envelope:
 
 ```bash
-python examples/offline_inference/text_to_video/text_to_video_lingbot.py \
+python text_to_video.py \
   --model robbyant/lingbot-video-dense-1.3b \
+  --prompt "a robotic arm picks up a red block" \
   --height 192 --width 320 --num-frames 9 --num-inference-steps 2 \
+  --guidance-scale 3.0 --flow-shift 3.0 --fps 24 \
   --output lingbot_t2v.mp4
 ```
 
 Use the same command with `robbyant/lingbot-video-moe-30b-a3b` for the MoE
 checkpoint. Requested frame counts are rounded upward to the causal
-VAE's `4n+1` grid.
+VAE's `4n+1` grid. For a local checkpoint path whose name does not contain
+`lingbot`, pass `--model-class-name LingBotVideoPipeline`.
+
+LingBot-only options such as `batch_cfg` and `output_type` use the shared
+model-extra channel, for example:
+
+```bash
+python text_to_video.py \
+  --model robbyant/lingbot-video-dense-1.3b \
+  --extra-body '{"batch_cfg": true, "output_type": "np"}'
+```
 
 ### LTX-2
 

examples/offline_inference/text_to_video/text_to_video.py

@@ -16,7 +16,7 @@ from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.lora.request import LoRARequest
 from vllm_omni.lora.utils import stable_lora_int_id
-from vllm_omni.model_extras import get_extra_body_params, get_model_class_name
+from vllm_omni.model_extras import get_extra_body_params, get_model_class_name, get_output_tensor_range
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.platforms import current_omni_platform
 
@@ -80,6 +80,17 @@ _MODEL_PRESETS = {
         "fps": 16,
         "output": "helios_output.mp4",
     },
+    "lingbot": {
+        "model_class_name": "LingBotVideoPipeline",
+        "height": 192,
+        "width": 320,
+        "num_frames": 9,
+        "num_inference_steps": 2,
+        "guidance_scale": 3.0,
+        "fps": 24,
+        "flow_shift": 3.0,
+        "output": "lingbot_video_output.mp4",
+    },
     "ltx2": {
         "height": 512,
         "width": 768,
@@ -110,27 +121,51 @@ _MODEL_PRESETS = {
 def _detect_preset(model: str, model_class_name: str | None = None) -> dict:
     model_lower = model.lower()
     class_lower = (model_class_name or "").lower()
-    if "distilled" in class_lower or "distilled" in model_lower:
-        return _MODEL_PRESETS["ltx2_distilled"]
-    if "ltx23" in class_lower or "ltx-2.3" in model_lower or "ltx_2.3" in model_lower:
-        return _MODEL_PRESETS["ltx23"]
-    if "ltx2" in class_lower or "ltx-2" in model_lower or "ltx_2" in model_lower:
+    if "lingbot" in model_lower or "lingbotvideo" in class_lower:
+        return _MODEL_PRESETS["lingbot"]
+    if "ltx" in class_lower or "ltx" in model_lower:
+        if "distilled" in class_lower or "distilled" in model_lower:
+            return _MODEL_PRESETS["ltx2_distilled"]
+        if "ltx23" in class_lower or "ltx-2.3" in model_lower or "ltx_2.3" in model_lower:
+            return _MODEL_PRESETS["ltx23"]
         return _MODEL_PRESETS["ltx2"]
-    if "vace" in model_lower:
+    if "vace" in model_lower or "vace" in class_lower:
         return _MODEL_PRESETS["vace"]
     # Edge must be matched before the generic cosmos branch (its "cosmos" substring would
     # otherwise pick up the Nano/Super 720p / gs 6.0 / flow_shift 10.0 preset).
     if ("cosmos" in model_lower or "cosmos" in class_lower) and ("edge" in model_lower or "edge" in class_lower):
         return _MODEL_PRESETS["cosmos3_edge"]
-    if "cosmos" in model_lower:
+    if "cosmos" in model_lower or "cosmos" in class_lower:
         return _MODEL_PRESETS["cosmos"]
-    if "hunyuan" in model_lower:
+    if "hunyuan" in model_lower or "hunyuan" in class_lower:
         return _MODEL_PRESETS["hunyuan"]
-    if "helios" in model_lower:
+    if "helios" in model_lower or "helios" in class_lower:
         return _MODEL_PRESETS["helios"]
     return _MODEL_PRESETS["wan"]
 
 
+def build_text_to_video_prompt(prompt: str, negative_prompt: str | None) -> dict[str, Any]:
+    """Build the canonical request envelope for the shared T2V example."""
+    result: dict[str, Any] = {
+        "prompt": prompt,
+        "modalities": ["video"],
+    }
+    if negative_prompt is not None:
+        result["negative_prompt"] = negative_prompt
+    return result
+
+
+def _normalize_float_tensor(tensor: torch.Tensor, source_range: str) -> torch.Tensor:
+    """Normalize decoded frames according to the pipeline's output contract."""
+    if not tensor.is_floating_point():
+        return tensor
+    if source_range == "negative_one_to_one":
+        return tensor.clamp(-1, 1) * 0.5 + 0.5
+    if source_range == "zero_to_one":
+        return tensor.clamp(0, 1)
+    raise ValueError(f"Unsupported floating-point tensor range: {source_range!r}")
+
+
 def parse_profiler_config(value: str) -> dict[str, Any]:
     try:
         config = json.loads(value)
@@ -160,7 +195,7 @@ def parse_extra_body(value: str) -> dict[str, Any]:
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(
         description="Generate a video from a text prompt. "
-        "Supports Wan2.2, HunyuanVideo-1.5, Helios, and other text-to-video models."
+        "Supports Wan2.2, HunyuanVideo-1.5, Helios, LingBot-Video, and other text-to-video models."
     )
     parser.add_argument(
         "--model",
@@ -390,7 +425,7 @@ def _extract_peak_memory_mb(result: Any) -> float:
         return 0.0
     val = getattr(result, "peak_memory_mb", 0.0)
     if not val:
-        inner = getattr(result, "request_output", None)
+        inner = result
         if isinstance(inner, list):
             inner = inner[0] if inner else None
         val = getattr(inner, "peak_memory_mb", 0.0)
@@ -408,6 +443,7 @@ def main():
     for key, default_val in preset.items():
         if getattr(args, key.replace("-", "_"), None) is None:
             setattr(args, key.replace("-", "_"), default_val)
+    model_class_name = args.model_class_name
 
     generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(args.seed)
     # Cache-dit config (Wan2.2 only)
@@ -480,6 +516,7 @@ def main():
     omni = Omni(**omni_kwargs)
     model_class_name = get_model_class_name(omni) or model_class_name
     declared_extra_body_params = get_extra_body_params(model_class_name)
+    output_tensor_range = get_output_tensor_range(model_class_name)
 
     if profiler_enabled:
         print("[Profiler] Starting profiling...")
@@ -513,12 +550,11 @@ def main():
             lora_path=lora_path,
         )
 
-    prompt_dict = {"prompt": args.prompt}
-    if args.negative_prompt is not None:
-        prompt_dict["negative_prompt"] = args.negative_prompt
-    elif preset not in (_MODEL_PRESETS["ltx2"], _MODEL_PRESETS["ltx23"]):
+    negative_prompt = args.negative_prompt
+    if negative_prompt is None and all(preset is not _MODEL_PRESETS[name] for name in ("lingbot", "ltx2", "ltx23")):
         # Preserve the historical empty-prompt behavior for non-LTX examples.
-        prompt_dict["negative_prompt"] = ""
+        negative_prompt = ""
+    prompt_dict = build_text_to_video_prompt(args.prompt, negative_prompt)
 
     extra_args = {}
     if lora_request:
@@ -575,8 +611,8 @@ def main():
         if frames.multimodal_output and "audio" in frames.multimodal_output:
             audio = frames.multimodal_output["audio"]
             audio_sample_rate = frames.multimodal_output.get("audio_sample_rate", audio_sample_rate)
-        if frames.is_pipeline_output and frames.request_output is not None:
-            inner_output = frames.request_output
+        if frames.is_pipeline_output and frames is not None:
+            inner_output = frames
             if isinstance(inner_output, OmniRequestOutput):
                 if inner_output.multimodal_output and "audio" in inner_output.multimodal_output:
                     audio = inner_output.multimodal_output["audio"]
@@ -630,8 +666,7 @@ def main():
                 frame_tensor = frame_tensor[0]
             if frame_tensor.dim() == 3 and frame_tensor.shape[0] in (3, 4):
                 frame_tensor = frame_tensor.permute(1, 2, 0)
-            if frame_tensor.is_floating_point():
-                frame_tensor = frame_tensor.clamp(-1, 1) * 0.5 + 0.5
+            frame_tensor = _normalize_float_tensor(frame_tensor, output_tensor_range)
             return frame_tensor.float().numpy()
         if isinstance(frame, np.ndarray):
             frame_array = frame
@@ -683,8 +718,7 @@ def main():
                 video_tensor = video_tensor[0]
         elif video_tensor.dim() == 4 and video_tensor.shape[0] in (3, 4):
             video_tensor = video_tensor.permute(1, 2, 3, 0)
-        if video_tensor.is_floating_point():
-            video_tensor = video_tensor.clamp(-1, 1) * 0.5 + 0.5
+        video_tensor = _normalize_float_tensor(video_tensor, output_tensor_range)
         video_array = video_tensor.float().numpy()
     elif isinstance(frames, np.ndarray):
         video_array = frames

examples/offline_inference/text_to_video/text_to_video_lingbot.py

@@ -1,131 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from __future__ import annotations
-
-import argparse
-import time
-from pathlib import Path
-from typing import Any
-
-import numpy as np
-import torch
-
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.outputs import OmniRequestOutput
-from vllm_omni.platforms import current_omni_platform
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Generate text-to-video with a LingBot-Video checkpoint.")
-    parser.add_argument("--model", default="robbyant/lingbot-video-dense-1.3b")
-    parser.add_argument("--prompt", default="a robotic arm picks up a red block")
-    parser.add_argument("--negative-prompt", default=None)
-    parser.add_argument("--output", default="lingbot_video_output.mp4")
-    parser.add_argument("--height", type=int, default=192)
-    parser.add_argument("--width", type=int, default=320)
-    parser.add_argument("--num-frames", type=int, default=9)
-    parser.add_argument("--num-inference-steps", type=int, default=2)
-    parser.add_argument("--guidance-scale", type=float, default=3.0)
-    parser.add_argument("--flow-shift", type=float, default=3.0)
-    parser.add_argument("--seed", type=int, default=42)
-    parser.add_argument("--fps", type=int, default=24)
-    parser.add_argument("--output-type", choices=["pt", "np"], default="pt")
-    parser.add_argument("--batch-cfg", action="store_true")
-    return parser.parse_args()
-
-
-def _peak_memory_mb(output: Any) -> float:
-    result = output[0] if isinstance(output, list) and output else output
-    return float(getattr(result, "peak_memory_mb", 0.0) or 0.0)
-
-
-def _extract_video(output: Any) -> Any:
-    if isinstance(output, list):
-        output = output[0] if output else None
-    if isinstance(output, OmniRequestOutput):
-        if output.final_output_type != "image":
-            raise ValueError(f"Unexpected output type {output.final_output_type!r}.")
-        if not output.images:
-            raise ValueError("No video frames found in OmniRequestOutput.")
-        return output.images[0] if len(output.images) == 1 else output.images
-    return output
-
-
-def _to_frame_array(video: Any) -> np.ndarray:
-    if isinstance(video, torch.Tensor):
-        tensor = video.detach().cpu()
-        if tensor.dim() == 5:
-            tensor = tensor[0]
-        if tensor.dim() == 4 and tensor.shape[-1] not in (3, 4):
-            if tensor.shape[0] in (3, 4):
-                tensor = tensor.permute(1, 2, 3, 0)
-            elif tensor.shape[1] in (3, 4):
-                tensor = tensor.permute(0, 2, 3, 1)
-        if tensor.is_floating_point():
-            if float(tensor.min()) < 0.0:
-                tensor = tensor.clamp(-1, 1) * 0.5 + 0.5
-            else:
-                tensor = tensor.clamp(0, 1)
-        return tensor.float().numpy()
-    if isinstance(video, np.ndarray):
-        array = video[0] if video.ndim == 5 else video
-        if np.issubdtype(array.dtype, np.integer):
-            return array.astype(np.float32) / 255.0
-        return array.astype(np.float32, copy=False)
-    if isinstance(video, list):
-        return np.stack([_to_frame_array(frame) for frame in video], axis=0)
-    return np.asarray(video, dtype=np.float32)
-
-
-def main() -> None:
-    args = parse_args()
-    generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(args.seed)
-    omni = Omni(
-        model=args.model,
-        model_class_name="LingBotVideoPipeline",
-        flow_shift=args.flow_shift,
-        parallel_config=DiffusionParallelConfig(),
-    )
-
-    prompt: dict[str, Any] = {
-        "prompt": args.prompt,
-        "modalities": ["video"],
-        "num_frames": args.num_frames,
-    }
-    if args.negative_prompt is not None:
-        prompt["negative_prompt"] = args.negative_prompt
-
-    sampling_params = OmniDiffusionSamplingParams(
-        height=args.height,
-        width=args.width,
-        num_frames=args.num_frames,
-        num_inference_steps=args.num_inference_steps,
-        guidance_scale=args.guidance_scale,
-        generator=generator,
-        output_type=args.output_type,
-        extra_args={"batch_cfg": args.batch_cfg, "shift": args.flow_shift},
-    )
-
-    start = time.perf_counter()
-    output = omni.generate(prompt, sampling_params)
-    elapsed = time.perf_counter() - start
-    print(f"Total generation time: {elapsed:.4f} seconds ({elapsed * 1000:.2f} ms)")
-    peak_mb = _peak_memory_mb(output)
-    if peak_mb:
-        print(f"Worker peak GPU memory (reserved): {peak_mb:.2f} MiB ({peak_mb / 1024:.2f} GiB)")
-
-    video = _to_frame_array(_extract_video(output))
-    output_path = Path(args.output)
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    from diffusers.utils import export_to_video
-
-    export_to_video(list(video), str(output_path), fps=args.fps)
-    print(f"Saved generated video to {output_path}")
-    omni.close()
-
-
-if __name__ == "__main__":
-    main()

examples/offline_inference/x_to_video_audio/x_to_video_audio.py

@@ -138,7 +138,7 @@ def _extract_peak_memory_mb(result: Any) -> float:
         return 0.0
     val = getattr(result, "peak_memory_mb", 0.0)
     if not val:
-        inner = getattr(result, "request_output", None)
+        inner = result
         if isinstance(inner, list):
             inner = inner[0] if inner else None
         val = getattr(inner, "peak_memory_mb", 0.0)

examples/online_serving/text_to_speech/README.md

@@ -605,6 +605,13 @@ python qwen3_tts/streaming_speech_client.py \
 The client writes one PCM file per sentence and a matching
 `sentence_XXX_timestamps.json` sidecar.
 
+Non-streaming requests can also ask for timestamps: pass
+`"word_timestamps": true` to `POST /v1/audio/speech` and read the
+`X-Word-Timestamps` response header (JSON list of `{word, start_ms, end_ms}`,
+ASCII-escaped). Past 4 KB the header is replaced by
+`X-Word-Timestamps-Omitted: oversize; bytes=<n>; limit=4096` and the audio
+still returns — use the WebSocket path for long transcripts.
+
 To *see* the alignment instead of reading a JSON sidecar, run the
 word-timestamp Gradio demo (server must be launched with `--forced-aligner`):
 ```bash

mkdocs.yml

@@ -73,6 +73,8 @@ plugins:
   - awesome-nav
   - redirects:
       redirect_maps:
+        community/meetups.md: community/meetings.md
+        community/volunteers.md: community/contact_us.md
         user_guide/diffusion/request_batching.md: user_guide/diffusion/execution_modes.md
         user_guide/diffusion/step_execution.md: user_guide/diffusion/execution_modes.md
         user_guide/diffusion/cpu_offload_diffusion.md: user_guide/diffusion/cpu_offload.md
@@ -111,6 +113,7 @@ plugins:
         - "vllm_omni.entrypoints.openai"  # avoid importing vllm in mkdocs building
         - "vllm_omni.entrypoints.openai.protocol"  # avoid importing vllm in mkdocs building
         - "vllm_omni.experimental"  # avoid importing optional serving deps (openai, fastapi, uvicorn, httpx) in mkdocs building
+        - "vllm_omni.model_executor.models.nemotron_voicechat.nemo_vendored"  # vendored NeMo code; upstream docstrings trip griffe strict warnings
   - mkdocstrings:
       handlers:
         python:

pyproject.toml

@@ -206,6 +206,12 @@ ignore = [
 # These files keep upstream conventions (star imports, long lines, math-symbol
 # argument names) so they stay easy to diff against upstream releases.
 "vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py" = ["E402", "E501", "F403", "F405", "N803"]
+# NemotronVoiceChat inference modules vendored from NVIDIA-NeMo/Speech (branch
+# nemotron-labs-voicechat). Numerics-bearing code is kept byte-close to
+# upstream to stay diffable and parity-safe, so upstream style (long lines,
+# legacy typing, whitespace, unused-import shims) is exempted. TID251 and the
+# F-correctness rules still apply.
+"vllm_omni/model_executor/models/nemotron_voicechat/nemo_vendored/**" = ["E501", "E721", "E741", "F401", "F841", "I001", "N803", "N814", "UP006", "UP007", "UP008", "UP011", "UP032", "UP035", "W291", "W293"]
 "vllm_omni/experimental/fullduplex/joyvl/decision/prompts.py" = ["E501"]  # System/summarizer prompts are long prose lines
 
 [tool.ruff.lint.flake8-tidy-imports.banned-api]
@@ -314,6 +320,10 @@ extend-ignore-identifiers-re = [
     ".*[Nn][Dd]3.*",
     ".*updator.*",
     "byt5.*",
+    # NemotronVoiceChat vendored triton kernel: dout/DOUT abbreviate d_out
+    # (output feature dim), not "doubt".
+    ".*dout.*",
+    ".*DOUT.*",
 ]
 extend-ignore-re = [
     "\\.\\.\\.n,d->\\.\\.\\.nd",
@@ -334,6 +344,9 @@ byt5 = "byt5"
 # "seeked" is the standard HTML media element DOM event name (fires when a
 # seek completes), used by the word-timestamps Gradio demo. Not a typo.
 seeked = "seeked"
+# Ryan Leary: author credited in the NemotronVoiceChat vendored NeMo
+# features.py copyright header. A surname, not a typo of "leery".
+Leary = "Leary"
 # IndexTTS 2.5 language codes and speech-emotion-recognition event token.
 fo = "fo"
 ba = "ba"

recipes/NVIDIA/NemotronLabs-VoiceChat.md

@@ -0,0 +1,113 @@
+# NemotronLabs VoiceChat 11B
+
+> Offline speech-to-speech: a 16 kHz user utterance in, the agent's spoken reply
+> (text + 22.05 kHz WAV) out, on a 3-stage vLLM-Omni pipeline.
+
+## Summary
+
+- Vendor: NVIDIA
+- Model: [`nvidia/NVIDIA-NemotronLabs-VoiceChat-11B`](https://huggingface.co/nvidia/NVIDIA-NemotronLabs-VoiceChat-11B)
+- Task: Speech-to-speech voice chat. The model runs a frame-locked 12.5 Hz
+  timeline: a Conformer + NemotronH hybrid-Mamba thinker emits one text token
+  per acoustic frame, a Gemma3-1B EAR-TTS talker turns the text timeline into
+  31-quantizer RVQ code stacks, and an RVQ-VAE codec decodes them to audio.
+- Mode: Offline single-turn inference (batch=1). Online/duplex serving,
+  function calling, and batch>1 are documented follow-ups.
+- Maintainer: [`@yuekaizhang`](https://github.com/yuekaizhang)
+
+## When to use this recipe
+
+Use this recipe to run a single-turn voice-chat exchange with
+`NVIDIA-NemotronLabs-VoiceChat-11B` on one GPU: you provide a user utterance as
+a WAV file (any sample rate; it is resampled to 16 kHz mono) plus an optional
+spoken-style system prompt, and get back the agent's reply as text and a
+22.05 kHz WAV. The integration is NeMo-free at runtime — the perception
+Conformer, EAR-TTS talker, and RVQ-VAE codec are vendored, dependency-stripped
+NeMo modules (`nemo_vendored/`), so no `nemo_toolkit` install is needed.
+
+## References
+
+- Offline example:
+  [`examples/offline_inference/nemotron_voicechat/end2end.py`](../../examples/offline_inference/nemotron_voicechat/end2end.py)
+- Model modules (thinker / talker / code2wav / vendored NeMo):
+  [`vllm_omni/model_executor/models/nemotron_voicechat/`](../../vllm_omni/model_executor/models/nemotron_voicechat/)
+- Staged pipeline config:
+  [`vllm_omni/deploy/nemotron_labs_voicechat.yaml`](../../vllm_omni/deploy/nemotron_labs_voicechat.yaml)
+- Upstream model card:
+  [`nvidia/NVIDIA-NemotronLabs-VoiceChat-11B`](https://huggingface.co/nvidia/NVIDIA-NemotronLabs-VoiceChat-11B)
+- Reference implementation: NVIDIA-NeMo/Speech, branch `nemotron-labs-voicechat`
+
+## Pipeline
+
+| stage | arch | dtype | role |
+|---|---|---|---|
+| 0 thinker | `NemotronVoiceChatThinkerForConditionalGeneration` (LLM_AR) | fp32 | WAV + system prompt -> frame-locked text-token timeline (+ function channel) |
+| 1 talker | `NemotronVoiceChatTalker` (LLM_AR) | fp32 | text timeline -> 31-quantizer RVQ code stacks (one per 80 ms frame) |
+| 2 code2wav | `NemotronVoiceChatCode2Wav` (LLM_GENERATION) | fp32 | RVQ-VAE decode -> 22.05 kHz PCM |
+
+Stages 0/1 default to fp32 for exact parity with the NeMo reference
+implementation (greedy decoding matches it token for token on the acceptance
+fixture). The deploy yaml documents a ~2x-faster bf16 thinker option whose
+output stayed within one word of the reference in testing.
+
+## Hardware Support
+
+## GPU
+
+### 1x H100 80GB
+
+#### Environment
+
+- OS: Linux
+- Python: 3.12
+- vLLM version: 0.26.0
+- vLLM-Omni version or commit: this PR / current `main`
+
+#### Command
+
+```bash
+# Tokenizer: the checkpoint ships no HF tokenizer; it resolves from the
+# nvidia/NVIDIA-Nemotron-Nano-9B-v2 HF id automatically. For air-gapped runs,
+# point NEMOTRON_VOICECHAT_LLM_PATH at a local snapshot of that repo instead.
+python examples/offline_inference/nemotron_voicechat/end2end.py \
+    --checkpoint /path/to/NVIDIA-NemotronLabs-VoiceChat-11B \
+    --wav /path/to/user_question.wav \
+    --output-dir results/nemotron_voicechat
+```
+
+#### Verification
+
+```bash
+ls results/nemotron_voicechat
+# <stem>_output.txt          the agent reply as text
+# <stem>_output.wav          the agent reply as 22.05 kHz audio
+# <stem>_text_tokens.json    the frame-locked text-token timeline
+```
+
+The reply text should read as a coherent spoken-style answer to the question in
+the input WAV, and the WAV should transcribe to (approximately) the same text
+with any ASR model.
+
+#### Notes
+
+- Memory usage: the shipped yaml runs all three stages on one GPU
+  (`gpu_memory_utilization` 0.62 / 0.12 / 0.06); peak usage is dominated by the
+  fp32 thinker. The fp32 default has a hard floor of roughly 43 GB of thinker
+  weights alone (9B backbone + 587M `embed_tokens` + 587M `function_head` +
+  0.6B Conformer), so 48 GB cards cannot run it — use the bf16 thinker option
+  documented in the deploy yaml on anything smaller than an 80 GB part.
+- Input sizing: the timeline is frame-locked, so the reply budget IS the input
+  duration. The acoustic channel trails the text channel; if the WAV does not
+  carry enough trailing silence for the reply to finish, the spoken answer is
+  truncated silently. Leave generous trailing silence (a question ending at
+  ~4.5 s truncated in an 8 s WAV but completed cleanly in 16 s); the offline
+  example warns when the text channel is still speaking near the last frame.
+- Key flags: sampling is greedy end to end. The thinker is frame-locked —
+  `max_tokens` equals the acoustic frame count with `ignore_eos=True`. Do NOT
+  set `min_tokens` on the thinker: the tokenizer's EOS token is also the
+  frame-locked PAD/silence token, so masking it forces the model to babble
+  instead of pausing.
+- The talker's `max_tokens` is 16383 (its stage prompt is one placeholder
+  token, and the stage context is 16384).
+- Known limitations: batch=1 only; offline single turn (no online/duplex
+  serving yet); the function-call channel is decoded but not yet acted on.

recipes/README.md

@@ -66,6 +66,7 @@ recipes/
 | [`OpenBMB/VoxCPM2.md`](./OpenBMB/VoxCPM2.md) | Online + offline TTS with native AR pipeline (48 kHz, 30+ languages) | 1x RTX 4090 24GB |
 | [`OpenMOSS/MOSS-TTS.md`](./OpenMOSS/MOSS-TTS.md) | Online + offline multilingual TTS (MOSS-TTS family, 8B) | 1x H100 80GB |
 | [`NVIDIA/Nemotron-Labs-Audex.md`](./NVIDIA/Nemotron-Labs-Audex.md) | TTS / text-to-audio / audio understanding / cascaded S2S (Audex 2B + 30B-A3B) | 1x H100 80GB |
+| [`NVIDIA/NemotronLabs-VoiceChat.md`](./NVIDIA/NemotronLabs-VoiceChat.md) | Offline speech-to-speech voice chat (11B, frame-locked 12.5 Hz, 3-stage thinker/talker/code2wav) | 1x H100 80GB |
 | [`Qwen/Qwen-Image.md`](./Qwen/Qwen-Image.md) | Text-to-image serving with step-wise continuous batching replay and ModelOpt mixed FP8/NVFP4 | 1x A100 80GB / 2x B200 |
 | [`Qwen/Qwen-Image-2512.md`](./Qwen/Qwen-Image-2512.md) | Text-to-image serving with step-wise continuous batching replay and ModelOpt FP8 / mixed FP8/NVFP4 | 1x A800 80GB / 2x B200 |
 | [`Qwen/Qwen-Image-Edit.md`](./Qwen/Qwen-Image-Edit.md) | Text-guided single-image editing | 1x or 2x H200 141GB |

recipes/Robbyant/LingBot-Video.md

@@ -27,7 +27,7 @@ parallel, cache, quantized, or expert-kernel backends.
 - Dense checkpoint: <https://huggingface.co/robbyant/lingbot-video-dense-1.3b>
 - MoE checkpoint: <https://huggingface.co/robbyant/lingbot-video-moe-30b-a3b>
 - Upstream project: <https://github.com/Robbyant/lingbot-video>
-- Related offline example: [`examples/offline_inference/text_to_video/text_to_video_lingbot.py`](../../examples/offline_inference/text_to_video/text_to_video_lingbot.py)
+- Related offline example: [`examples/offline_inference/text_to_video/text_to_video.py`](../../examples/offline_inference/text_to_video/text_to_video.py)
 - Related online video API docs: [`docs/serving/videos_api.md`](../../docs/serving/videos_api.md)
 
 ## Hardware Support
@@ -52,7 +52,7 @@ The MoE path is validated with BF16 expert weights.
 
 ```bash
 CUDA_VISIBLE_DEVICES=0 \
-python examples/offline_inference/text_to_video/text_to_video_lingbot.py \
+python examples/offline_inference/text_to_video/text_to_video.py \
   --model robbyant/lingbot-video-dense-1.3b \
   --prompt "a robotic arm picks up a red block" \
   --output lingbot_t2v.mp4 \
@@ -70,7 +70,7 @@ python examples/offline_inference/text_to_video/text_to_video_lingbot.py \
 
 ```bash
 CUDA_VISIBLE_DEVICES=0 \
-python examples/offline_inference/text_to_video/text_to_video_lingbot.py \
+python examples/offline_inference/text_to_video/text_to_video.py \
   --model robbyant/lingbot-video-moe-30b-a3b \
   --prompt "a robotic arm picks up a red block" \
   --output lingbot_moe_t2v.mp4 \

recipes/Tencent/HunyuanImage-3.0-Instruct.md

@@ -82,7 +82,7 @@ example and the DiT sections below.
 - Optional environment variables:
 
 ```bash
-export CACHE_DIT_VERSION=1.3.0
+export CACHE_DIT_VERSION=1.5.0
 ```
 
 HunyuanImage-3.0 sets the diffusion attention backend to `TORCH_SDPA`

requirements/common.txt

@@ -16,7 +16,7 @@ diffusers==0.38.0
 safetensors>=0.8.0
 accelerate==1.12.0
 soundfile>=0.13.1
-cache-dit==1.3.0
+cache-dit==1.5.0
 tqdm>=4.66.0
 torchsde>=0.2.6
 openai-whisper>=20250625
@@ -34,3 +34,5 @@ pydantic-settings>=2.0.0
 # base image does not ship it, so declare it explicitly. The loader also guards
 # this import so a missing gguf no longer breaks all diffusion model loading.
 gguf>=0.10.0
+# Required for running the Cosmos3 model with guardrails enabled (default behavior).
+cosmos-guardrail>=0.3.1

tests/benchmarks/test_daily_omni_local_source.py

@@ -0,0 +1,92 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Offline Daily-Omni sources: local mirrors must not fall back to the Hub id."""
+
+import json
+import os
+import tarfile
+from pathlib import Path
+
+import pytest
+
+from vllm_omni.benchmarks.data_modules.daily_omni_dataset import (
+    daily_omni_local_qa_json,
+    daily_omni_local_videos_dir,
+    resolve_daily_omni_local_root,
+)
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.benchmark]
+
+
+def _write_qa_json(root: Path) -> Path:
+    qa = root / "qa.json"
+    qa.write_text(json.dumps([{"Question": "q", "Answer": "A", "video_id": "vid0"}]), encoding="utf-8")
+    return qa
+
+
+def test_resolve_local_root_returns_none_for_hub_id() -> None:
+    assert resolve_daily_omni_local_root("liarliar/Daily-Omni") is None
+    assert resolve_daily_omni_local_root("") is None
+    assert resolve_daily_omni_local_root(None) is None
+
+
+def test_plain_local_mirror(tmp_path: Path) -> None:
+    _write_qa_json(tmp_path)
+    (tmp_path / "Videos" / "vid0").mkdir(parents=True)
+
+    root = resolve_daily_omni_local_root(str(tmp_path))
+    assert root == tmp_path.resolve()
+    assert daily_omni_local_qa_json(root) == (tmp_path / "qa.json").resolve()
+    assert daily_omni_local_videos_dir(root) == root / "Videos"
+
+
+def test_hf_cache_dir_resolves_to_snapshot(tmp_path: Path) -> None:
+    cache_dir = tmp_path / "datasets--liarliar--Daily-Omni"
+    snapshot = cache_dir / "snapshots" / "deadbeef"
+    snapshot.mkdir(parents=True)
+    _write_qa_json(snapshot)
+    (cache_dir / "refs").mkdir()
+    (cache_dir / "refs" / "main").write_text("deadbeef", encoding="utf-8")
+
+    root = resolve_daily_omni_local_root(str(cache_dir))
+    assert root == snapshot.resolve()
+    assert daily_omni_local_qa_json(root) == (snapshot / "qa.json").resolve()
+
+
+def test_hf_cache_dir_without_refs_picks_newest_snapshot(tmp_path: Path) -> None:
+    cache_dir = tmp_path / "datasets--liarliar--Daily-Omni"
+    older = cache_dir / "snapshots" / "aaa"
+    newer = cache_dir / "snapshots" / "bbb"
+    older.mkdir(parents=True)
+    newer.mkdir(parents=True)
+    os.utime(older, (1, 1))
+    os.utime(newer, (10, 10))
+
+    assert resolve_daily_omni_local_root(str(cache_dir)) == newer.resolve()
+
+
+def test_videos_tar_is_extracted(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("HF_HOME", str(tmp_path / "hf_home"))
+    root = tmp_path / "mirror"
+    (root / "Videos" / "vid0").mkdir(parents=True)
+    (root / "Videos" / "vid0" / "vid0_video.mp4").write_bytes(b"fake")
+    with tarfile.open(root / "Videos.tar", "w") as tf:
+        tf.add(root / "Videos", arcname="Videos")
+    # Only the tarball is left in the mirror.
+    for child in (root / "Videos").rglob("*"):
+        if child.is_file():
+            child.unlink()
+    (root / "Videos" / "vid0").rmdir()
+    (root / "Videos").rmdir()
+
+    videos_dir = daily_omni_local_videos_dir(root)
+    assert videos_dir is not None
+    assert (videos_dir / "vid0" / "vid0_video.mp4").is_file()
+
+    # Second call reuses the extraction instead of redoing it.
+    assert daily_omni_local_videos_dir(root) == videos_dir
+
+
+def test_missing_qa_json_reports_none(tmp_path: Path) -> None:
+    root = resolve_daily_omni_local_root(str(tmp_path))
+    assert daily_omni_local_qa_json(root) is None
+    assert daily_omni_local_videos_dir(root) is None

tests/benchmarks/test_daily_omni_pack_modes.py

@@ -2,9 +2,8 @@
 
 The ``minicpm-interleave`` mode is an accuracy-critical protocol: MiniCPM-o only reaches
 OpenBMB's reported Daily-Omni score when frames and 1s audio segments arrive strictly
-interleaved, with ``max_slice_nums=1`` / ``use_image_id=False`` and an empty MiniCPM
-system message (not the Qwen system prompt). These tests pin that contract so a refactor
-cannot silently regress the score.
+interleaved, with ``max_slice_nums=1`` / ``use_image_id=False`` and no Qwen system prompt.
+These tests pin that contract so a refactor cannot silently regress the score.
 
 vllm stubs are installed by tests/benchmarks/conftest.py before collection.
 """
@@ -99,22 +98,13 @@ def _stub_extract(monkeypatch, dataset: DailyOmniDataset, counter: list[int]) ->
 # ---------------------------------------------------------------------------
 
 
-@pytest.mark.parametrize(("n", "k"), [(10, 4), (100, 64), (1000, 64), (7, 7), (3, 5), (1, 1)])
-def test_uniform_sample_indices_matches_openbmb_midpoint(n: int, k: int) -> None:
-    """Pin OpenBMB midpoint-bin sampling (not np.linspace endpoints)."""
+@pytest.mark.parametrize(("n", "k"), [(10, 4), (100, 64), (7, 7), (3, 5), (1, 1)])
+def test_uniform_sample_indices_matches_numpy_linspace(n: int, k: int) -> None:
     got = _uniform_sample_indices(n, k)
     if n <= k:
         assert got == list(range(n))
     else:
-        gap = n / k
-        expected = [int(i * gap + gap / 2) for i in range(k)]
-        assert got == expected
-        # Endpoints differ from linspace for typical Daily-Omni downsample sizes.
-        linspace = [int(i) for i in np.linspace(0, n - 1, k, dtype=int)]
-        if n == 1000 and k == 64:
-            assert got[0] == 7 and got[-1] == 992
-            assert linspace[0] == 0 and linspace[-1] == 999
-            assert got != linspace
+        assert got == [int(i) for i in np.linspace(0, n - 1, k, dtype=int)]
 
 
 def test_uniform_sample_indices_non_positive_k() -> None:
@@ -223,16 +213,14 @@ def test_minicpm_interleave_missing_local_video_returns_none(qa_json, tmp_path)
 # ---------------------------------------------------------------------------
 
 
-def test_minicpm_interleave_keeps_empty_system_message(monkeypatch, qa_json, video_dir) -> None:
-    """Official MiniCPM omni mode still sends role=system with empty content."""
+def test_minicpm_interleave_omits_qwen_system_message(monkeypatch, qa_json, video_dir) -> None:
     ds = _make_dataset(qa_json, video_dir, pack_mode="minicpm-interleave", input_mode="all")
     _stub_extract(monkeypatch, ds, [])
     parts, _extra, _position = ds._compose_daily_omni_multimodal(_VIDEO_ID, None)
 
     messages = ds._build_daily_omni_openai_messages(parts, "q?", {"A": "a"})
 
-    assert [m["role"] for m in messages] == ["system", "user"]
-    assert messages[0]["content"][0]["text"] == ""
+    assert [m["role"] for m in messages] == ["user"]
 
 
 def test_qwen_pack_mode_keeps_system_message(qa_json, video_dir) -> None:

tests/config/test_config_factory.py

@@ -6,7 +6,7 @@ Unit tests for StageConfigFactory and related classes.
 
 import importlib
 import warnings
-from dataclasses import dataclass, fields
+from dataclasses import dataclass
 from pathlib import Path
 from unittest.mock import patch
 
@@ -34,9 +34,8 @@ from vllm_omni.config.stage_config import (
     load_deploy_config,
     merge_pipeline_deploy,
     pipeline_cfg_resolver,
-    strip_parent_engine_args,
 )
-from vllm_omni.engine.arg_utils import SHARED_FIELDS, EngineArgs, internal_blacklist_keys
+from vllm_omni.engine.arg_utils import SHARED_FIELDS, internal_blacklist_keys
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
@@ -475,7 +474,7 @@ class TestStageConfigFactory:
         cli_overrides = {
             "gpu_memory_utilization": 0.9,
             "model": "some_model",  # Internal
-            "stage_configs_path": "/path",  # Internal
+            "deploy_config": "/path",  # Internal
             "batch_timeout": 10,  # Internal
         }
 
@@ -483,7 +482,7 @@ class TestStageConfigFactory:
 
         assert overrides["gpu_memory_utilization"] == 0.9
         assert "model" not in overrides
-        assert "stage_configs_path" not in overrides
+        assert "deploy_config" not in overrides
         assert "batch_timeout" not in overrides
 
     def test_per_stage_override_excludes_internal_keys(self):
@@ -525,49 +524,6 @@ class TestStageResolutionHelpers:
         assert "model" not in overrides
         assert "parallel_config" not in overrides
 
-    def test_strip_parent_engine_args_reports_only_surprising_parent_overrides(self):
-        parent_fields = {f.name: f for f in fields(EngineArgs)}
-        filtered, overridden = strip_parent_engine_args(
-            {
-                "model": "some/model",
-                "stage_configs_path": "/tmp/stages.yaml",
-                "tensor_parallel_size": 4,
-                "worker_extension_cls": "some.Extension",
-                "custom_pipeline_args": {"pipeline_class": "demo.Pipeline"},
-            },
-            parent_fields=parent_fields,
-            keep_keys={"worker_extension_cls"},
-            strip_keys={"stage_configs_path"},
-            no_warn_keys={"model"},
-        )
-
-        assert filtered == {
-            "worker_extension_cls": "some.Extension",
-            "custom_pipeline_args": {"pipeline_class": "demo.Pipeline"},
-        }
-        assert overridden == ["tensor_parallel_size"]
-
-    def test_strip_parent_engine_args_keeps_allowed_media_access_controls(self):
-        parent_fields = {f.name: f for f in fields(EngineArgs)}
-        filtered, overridden = strip_parent_engine_args(
-            {
-                "model": "some/model",
-                "stage_configs_path": "/tmp/stages.yaml",
-                "allowed_local_media_path": "/data/qwentts",
-                "allowed_media_domains": ["example.com"],
-            },
-            parent_fields=parent_fields,
-            keep_keys={"allowed_local_media_path", "allowed_media_domains"},
-            strip_keys={"stage_configs_path"},
-            no_warn_keys={"model"},
-        )
-
-        assert filtered == {
-            "allowed_local_media_path": "/data/qwentts",
-            "allowed_media_domains": ["example.com"],
-        }
-        assert overridden == []
-
 
 class TestPipelineDiscovery:
     """Tests for the central pipeline registry (``OMNI_PIPELINES``)."""

tests/config/test_forced_aligner_injection.py

@@ -0,0 +1,71 @@
+import pytest
+
+import vllm_omni.config.pipeline_registry  # noqa: F401  (populate registry)
+from vllm_omni.config.pipeline_registry import OMNI_PIPELINES as _PIPELINE_REGISTRY
+from vllm_omni.config.stage_config import (
+    DeployConfig,
+    StageExecutionType,
+)
+from vllm_omni.model_executor.stage_input_processors.forced_aligner import (
+    POOLING_OUTPUT_DECODER_PATH,
+    TIMESTAMPS_MODALITY,
+)
+from vllm_omni.utils.forced_aligner import _DEFAULT_CONFIG_PATH, inject_forced_aligner_stage
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+def test_inject_appends_pooling_aligner_stage():
+    pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
+    deploy = DeployConfig()
+    n0 = len(pipeline.stages)
+
+    ext_pipeline, ext_deploy = inject_forced_aligner_stage(
+        pipeline, deploy, {"forced_aligner": "/models/Qwen3-ForcedAligner-0.6B"}
+    )
+
+    # A new pooling stage is appended; the original pipeline is unchanged.
+    assert len(pipeline.stages) == n0
+    assert len(ext_pipeline.stages) == n0 + 1
+    # The original deploy config is not mutated either (double-injection safe).
+    assert len(deploy.stages) == 0
+    assert len(ext_deploy.stages) == 1
+    aligner = ext_pipeline.stages[-1]
+    # Pooling stage reuses LLM_AR (+ runner=pooling), no dedicated LLM_POOLING.
+    assert aligner.execution_type == StageExecutionType.LLM_AR
+    assert aligner.final_output is True
+    assert aligner.final_output_type == TIMESTAMPS_MODALITY
+    assert aligner.input_sources == (n0 - 1,)
+    assert aligner.requires_multimodal_data is True
+    assert aligner.owns_tokenizer is True
+
+    # The deploy stage carries the separate aligner checkpoint + decoder hook.
+    ds = ext_deploy.stages[-1]
+    assert ds.engine_extras["model"] == "/models/Qwen3-ForcedAligner-0.6B"
+    assert ds.engine_extras["runner"] == "pooling"
+    assert ds.engine_extras["pooling_output_decoder"] == POOLING_OUTPUT_DECODER_PATH
+
+
+def test_inject_from_config_only():
+    # --forced-aligner-config alone (model in the YAML) must also inject.
+    pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
+    deploy = DeployConfig()
+    n0 = len(pipeline.stages)
+
+    ext_pipeline, ext_deploy = inject_forced_aligner_stage(
+        pipeline, deploy, {"forced_aligner": None, "forced_aligner_config": str(_DEFAULT_CONFIG_PATH)}
+    )
+
+    assert len(ext_pipeline.stages) == n0 + 1
+    assert ext_pipeline.stages[-1].final_output_type == TIMESTAMPS_MODALITY
+    assert ext_deploy.stages[-1].engine_extras["model"] == "Qwen/Qwen3-ForcedAligner-0.6B"
+
+
+def test_inject_noop_without_forced_aligner():
+    pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
+    deploy = DeployConfig()
+
+    ext_pipeline, ext_deploy = inject_forced_aligner_stage(pipeline, deploy, {})
+
+    assert ext_pipeline is pipeline
+    assert len(ext_deploy.stages) == 0

tests/config/test_omni_config.py

@@ -513,6 +513,7 @@ def test_sub_config_fields_match_structured_scopes():
         "enable_prefix_caching",
         "disable_hybrid_kv_cache_manager",
         "mm_processor_cache_gb",
+        "mamba_ssm_cache_dtype",
     }
     assert {f.name for f in fields(OmniStageSchedulerConfig)} == {
         "max_num_seqs",
@@ -581,8 +582,12 @@ def test_diffusion_parallel_config_keeps_current_diffusion_parallel_surface():
     assert cfg.data_parallel_size == 3
     assert cfg.cfg_parallel_size == 3
     assert cfg.mask_sp_padding is True
+    assert cfg.vae_parallel_mode == "tile"
     assert cfg.world_size == 72
 
+    auto_cfg = OmniStageDiffusionParallelConfig(vae_parallel_mode="auto")
+    assert auto_cfg.vae_parallel_mode == "auto"
+
 
 def test_parallel_config_derived_fields_are_not_init_inputs():
     with pytest.raises(ValidationError):

tests/core/sched/test_omni_ar_scheduler_logprobs.py

@@ -27,6 +27,7 @@ _MIXIN_UPDATE_HELPERS = (
     "_attach_finished_request_sets",
     "_attach_scheduler_stats",
     "_capture_omni_connector_output",
+    "_maybe_decode_pooling_output",
 )
 
 
@@ -138,6 +139,7 @@ def _make_scheduler_stub(requests: list[_Request]) -> SimpleNamespace:
         pending_stop_after_extraction=set(),
         waiting_for_transfer_free=set(),
         _new_prompt_len_snapshot={},
+        _pooling_output_decoder=None,
         finished_req_ids=set(),
         finished_req_ids_dict=defaultdict(set),
         kv_cache_manager=SimpleNamespace(take_events=lambda: None),
@@ -267,3 +269,72 @@ def test_invalid_logprobs_finish_only_the_affected_scheduler_request() -> None:
     assert output_by_id["bad"].new_token_ids == []
     assert output_by_id["good"].new_token_ids == [8]
     np.testing.assert_array_equal(output_by_id["good"].new_logprobs.logprob_token_ids[:, 0], [8])
+
+
+def _pooling_model_runner_output(pooler_tensor) -> SimpleNamespace:
+    return SimpleNamespace(
+        sampled_token_ids=[[]],
+        logprobs=None,
+        prompt_logprobs_dict={},
+        pooler_output=[pooler_tensor],
+        num_nans_in_logits=None,
+        kv_connector_output=None,
+        cudagraph_stats=None,
+        req_id_to_index={"req": 0},
+        routed_experts=None,
+    )
+
+
+def _make_pooling_stub(decoder) -> tuple[_Request, SimpleNamespace]:
+    request = _Request("req")
+    request.sampling_params = SimpleNamespace(num_logprobs=None)
+    request.pooling_params = SimpleNamespace()
+    scheduler = _make_scheduler_stub([request])
+    scheduler._pooling_output_decoder = decoder
+    scheduler.vllm_config = SimpleNamespace(model_config=SimpleNamespace(hf_config=None))
+    _bind_request_lifecycle(scheduler, update_request=lambda req, tokens: (tokens, False))
+    return request, scheduler
+
+
+def test_pooling_decode_success_emits_decoded_payload() -> None:
+    import torch
+
+    decoded = torch.tensor([[1, 2]], dtype=torch.int32)
+    request, scheduler = _make_pooling_stub(lambda logits, req, hf: decoded)
+
+    scheduler_output = SimpleNamespace(
+        num_scheduled_tokens={"req": 1},
+        scheduled_spec_decode_tokens={},
+        num_invalid_spec_tokens=0,
+    )
+    outputs = OmniARScheduler.update_from_output(
+        scheduler, scheduler_output, _pooling_model_runner_output(torch.zeros(2, 4))
+    )
+    (output,) = outputs[0].outputs
+
+    assert output.finish_reason is FinishReason.STOP
+    assert output.pooling_output is decoded
+
+
+def test_pooling_decode_failure_finishes_request_with_error() -> None:
+    import torch
+
+    def raising_decoder(logits, req, hf):
+        raise ValueError("boom")
+
+    request, scheduler = _make_pooling_stub(raising_decoder)
+
+    scheduler_output = SimpleNamespace(
+        num_scheduled_tokens={"req": 1},
+        scheduled_spec_decode_tokens={},
+        num_invalid_spec_tokens=0,
+    )
+    outputs = OmniARScheduler.update_from_output(
+        scheduler, scheduler_output, _pooling_model_runner_output(torch.zeros(2, 4))
+    )
+    (output,) = outputs[0].outputs
+
+    # Decoder failure is a request error (500), not an empty success.
+    assert output.finish_reason is FinishReason.ERROR
+    assert output.pooling_output is None
+    assert "pooling output decode failed" in str(request.stop_reason)

tests/core/sched/test_omni_scheduling_coordinator.py

@@ -120,6 +120,9 @@ class TestFullPayloadCoordinatorSelection(unittest.TestCase):
             ("MiMoAudioModel", "code2wav"),
             ("Qwen3TTSCode2Wav", "code2wav"),
             ("MiniCPMO45Code2Wav", "code2wav"),
+            # nemotron_voicechat: sync-path RVQ-VAE decode waits for the
+            # talker's full [frames, 31] code-stack payload.
+            ("NemotronVoiceChatCode2Wav", "code2wav"),
             ("CosyVoice3Model", "cosyvoice3_code2wav"),
             # audex: sync-path TTS decoder (async_chunk: false deploys) and
             # the TTA XCodec1 stage (always sync full-payload).

tests/dfx/perf/tests/test_minicpmo_4_5_ready.json

@@ -0,0 +1,80 @@
+[
+  {
+    "test_name": "test_minicpmo_4_5_ready",
+    "mark": [
+      {
+        "hardware_marks": {
+          "res": {
+            "cuda": "H100",
+            "npu": "A3"
+          },
+          "num_cards": 1
+        }
+      },
+      "core_model",
+      "omni"
+    ],
+    "server_params": {
+      "model": "openbmb/MiniCPM-o-4_5",
+      "extra_cli_args": [
+        "--deploy-config",
+        "vllm_omni/deploy/minicpmo_4_5.yaml",
+        "--trust-remote-code"
+      ]
+    },
+    "benchmark_params": [
+      {
+        "dataset_name": "seed-tts",
+        "dataset_path": "zhaochenyang20/seed-tts-eval",
+        "backend": "openai-chat-omni",
+        "endpoint": "/v1/chat/completions",
+        "num_prompts": [
+          10,
+          40
+        ],
+        "max_concurrency": [
+          1,
+          4
+        ],
+        "no_oversample": true,
+        "trust_remote_code": true,
+        "seed_tts_locale": "en",
+        "extra_body": {
+          "modalities": [
+            "text",
+            "audio"
+          ],
+          "chat_template_kwargs": {
+            "enable_thinking": false,
+            "use_tts_template": true
+          }
+        },
+        "percentile-metrics": "ttft,e2el,audio_ttfp,audio_rtf,audio_duration",
+        "baseline": {
+          "H100": {
+            "request_throughput": [
+              0.7803,
+              1.1008
+            ],
+            "mean_ttft_ms": [
+              166.9764,
+              307.7133
+            ],
+            "mean_e2el_ms": [
+              1361.1682,
+              3800.6332
+            ],
+            "mean_audio_ttfp_ms": [
+              635.8099,
+              1615.6591
+            ],
+            "mean_audio_rtf": [
+              0.3151,
+              0.8568
+            ]
+          }
+        }
+      }
+    ]
+  }
+]

tests/diffusion/batching/test_diffusion_batching.py

@@ -125,7 +125,7 @@ def _extract_images(output: OmniRequestOutput) -> list:
         return output.images
     # When the output comes from the orchestrator pipeline, images may be
     # nested inside request_output.
-    inner = getattr(output, "request_output", None)
+    inner = output
     if inner is not None and hasattr(inner, "images") and inner.images:
         return inner.images
     return []

tests/diffusion/cache/test_teacache.py

@@ -56,10 +56,10 @@ def test_teacache(model_name: str):
         # Extract images from request_output['images']
         first_output = outputs[0]
         assert first_output.final_output_type == "image"
-        if not hasattr(first_output, "request_output") or not first_output.request_output:
+        if not isinstance(first_output, OmniRequestOutput) or not first_output:
             raise ValueError("No request_output found in OmniRequestOutput")
 
-        req_out = first_output.request_output
+        req_out = first_output
         if not isinstance(req_out, OmniRequestOutput) or not hasattr(req_out, "images"):
             raise ValueError("Invalid request_output structure or missing 'images' key")
 

tests/diffusion/distributed/test_sequence_parallel.py

@@ -131,7 +131,7 @@ def _run_inference(
             elapsed_ms = (time.time() - start) * 1000
 
             return InferenceResult(
-                images=outputs[0].request_output.images,
+                images=outputs[0].images,
                 elapsed_ms=elapsed_ms,
             )
     finally:

tests/diffusion/distributed/test_tensor_parallel.py

@@ -49,10 +49,10 @@ def _diff_metrics(a: Image.Image, b: Image.Image) -> tuple[float, float]:
 def _extract_single_image(outputs) -> Image.Image:
     first_output = outputs[0]
     assert first_output.final_output_type == "image"
-    if not hasattr(first_output, "request_output") or not first_output.request_output:
+    if not isinstance(first_output, OmniRequestOutput) or not first_output:
         raise ValueError("No request_output found in OmniRequestOutput")
 
-    req_out = first_output.request_output
+    req_out = first_output
     if not isinstance(req_out, OmniRequestOutput) or not hasattr(req_out, "images"):
         raise ValueError("Invalid request_output structure or missing 'images' key")
 

tests/diffusion/distributed/test_wan_spatial_shard.py

@@ -1,4 +1,5 @@
 import os
+from contextlib import contextmanager
 from types import SimpleNamespace
 
 import pytest
@@ -10,6 +11,7 @@ from tests.helpers.mark import hardware_test
 from vllm_omni.diffusion.distributed.autoencoders import wan_spatial_shard
 from vllm_omni.diffusion.distributed.autoencoders.autoencoder_kl_wan import (
     DistributedAutoencoderKLWan,
+    select_auto_spatial_shard_split_dim,
 )
 from vllm_omni.platforms import current_omni_platform
 
@@ -17,6 +19,23 @@ from vllm_omni.platforms import current_omni_platform
 # the end uses hardware_test (H100 x2) plus full_model / diffusion / parallel.
 
 
+@contextmanager
+def _active_spatial_shard(split_dim: str):
+    token = wan_spatial_shard._SPATIAL_SHARD_CONTEXT.set(
+        wan_spatial_shard.SpatialShardContext(
+            input_extent=2,
+            local_input_extent=2,
+            split_dim=split_dim,
+            rank=0,
+            world_size=3,
+        )
+    )
+    try:
+        yield
+    finally:
+        wan_spatial_shard._SPATIAL_SHARD_CONTEXT.reset(token)
+
+
 @pytest.mark.core_model
 @pytest.mark.cpu
 def test_split_for_parallel_decode_pads_uneven_height():
@@ -231,6 +250,128 @@ def test_reshard_from_trimmed_width_pads_invalid_columns(monkeypatch: pytest.Mon
     assert torch.equal(out.flatten(), torch.tensor([4.0, 0.0]))
 
 
+@pytest.mark.core_model
+@pytest.mark.cpu
+@pytest.mark.parametrize("split_dim", ["height", "width"])
+@pytest.mark.parametrize(
+    ("full_extent", "world_size"),
+    [
+        pytest.param(5, 4, id="four-ranks-empty-tail"),
+        pytest.param(20, 8, id="production-eight-ranks-empty-tail"),
+        pytest.param(39, 2, id="two-ranks-partial-tail"),
+    ],
+)
+def test_reshard_from_trimmed_extent_clamps_and_reconstructs(
+    monkeypatch: pytest.MonkeyPatch,
+    split_dim: str,
+    full_extent: int,
+    world_size: int,
+):
+    local_extent = (full_extent + world_size - 1) // world_size
+    shape = (1, 1, 1, full_extent, 1) if split_dim == "height" else (1, 1, 1, 1, full_extent)
+    x = torch.arange(full_extent, dtype=torch.float32).reshape(shape)
+    outputs = []
+
+    for rank in range(world_size):
+        monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group, rank=rank: (rank, world_size))
+        token = wan_spatial_shard._SPATIAL_SHARD_CONTEXT.set(
+            wan_spatial_shard.SpatialShardContext(
+                input_extent=full_extent,
+                local_input_extent=local_extent,
+                split_dim=split_dim,
+                rank=rank,
+                world_size=world_size,
+            )
+        )
+        try:
+            out = wan_spatial_shard.reshard_from_trimmed_extent(
+                x,
+                local_extent=local_extent,
+                split_dim=split_dim,
+                group=object(),
+            )
+        finally:
+            wan_spatial_shard._SPATIAL_SHARD_CONTEXT.reset(token)
+
+        start = rank * local_extent
+        valid_extent = min(local_extent, max(0, full_extent - start))
+        expected = torch.zeros(local_extent, dtype=torch.float32)
+        if valid_extent:
+            expected[:valid_extent] = torch.arange(start, start + valid_extent, dtype=torch.float32)
+        assert out.shape[_spatial_dim_for_test(split_dim)] == local_extent
+        assert torch.equal(out.flatten(), expected)
+        outputs.append(out)
+
+    reconstructed = torch.cat(outputs, dim=_spatial_dim_for_test(split_dim))
+    reconstructed = reconstructed.narrow(_spatial_dim_for_test(split_dim), 0, full_extent)
+    assert torch.equal(reconstructed, x)
+
+
+def _spatial_dim_for_test(split_dim: str) -> int:
+    return -2 if split_dim == "height" else -1
+
+
+@pytest.mark.core_model
+@pytest.mark.cpu
+def test_reshard_from_trimmed_extent_rejects_rank_mismatch(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (1, 2))
+    x = torch.zeros((1, 1, 1, 1, 4), dtype=torch.float32)
+    token = wan_spatial_shard._SPATIAL_SHARD_CONTEXT.set(
+        wan_spatial_shard.SpatialShardContext(
+            input_extent=4,
+            local_input_extent=2,
+            split_dim="width",
+            rank=0,
+            world_size=2,
+        )
+    )
+    try:
+        with pytest.raises(RuntimeError, match="group_rank=1, context_rank=0"):
+            wan_spatial_shard.reshard_from_trimmed_extent(
+                x,
+                local_extent=2,
+                split_dim="width",
+                group=object(),
+            )
+    finally:
+        wan_spatial_shard._SPATIAL_SHARD_CONTEXT.reset(token)
+
+
+@pytest.mark.core_model
+@pytest.mark.cpu
+def test_attention_wrapper_rejects_spatial_extent_change(monkeypatch: pytest.MonkeyPatch):
+    class ShrinkingWanAttentionBlock(torch.nn.Module):
+        def forward(self, x):
+            return x[..., :-1]
+
+    monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (0, 2))
+    monkeypatch.setattr(
+        wan_spatial_shard,
+        "all_gather_along_dim",
+        lambda x, *, group, dim, dst=None: torch.cat([x, x], dim=dim),
+    )
+    module = ShrinkingWanAttentionBlock()
+    wan_spatial_shard._patch_attention_block(module, group=object())
+    x = torch.zeros((1, 1, 1, 1, 2), dtype=torch.float32)
+    token = wan_spatial_shard._SPATIAL_SHARD_CONTEXT.set(
+        wan_spatial_shard.SpatialShardContext(
+            input_extent=4,
+            local_input_extent=2,
+            split_dim="width",
+            rank=0,
+            world_size=2,
+        )
+    )
+    try:
+        with pytest.raises(
+            RuntimeError,
+            match=r"local_extent=2, gathered_extent=4, trimmed_extent=4, output_extent=3",
+        ):
+            module(x)
+    finally:
+        wan_spatial_shard._SPATIAL_SHARD_CONTEXT.reset(token)
+
+
 @pytest.mark.core_model
 @pytest.mark.cpu
 def test_halo_exchange_single_rank_noop(monkeypatch: pytest.MonkeyPatch):
@@ -255,12 +396,14 @@ def test_dist_zero_pad_only_applies_global_height_edges(monkeypatch: pytest.Monk
 
     monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (1, 3))
     mid_rank_pad = wan_spatial_shard.WanDistZeroPad2d((0, 1, 1, 1), group=object())
-    mid = mid_rank_pad(x)
+    with _active_spatial_shard("height"):
+        mid = mid_rank_pad(x)
     assert mid.shape == (1, 1, 2, 3)
 
     monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (2, 3))
     last_rank_pad = wan_spatial_shard.WanDistZeroPad2d((0, 1, 1, 1), group=object())
-    last = last_rank_pad(x)
+    with _active_spatial_shard("height"):
+        last = last_rank_pad(x)
     assert last.shape == (1, 1, 3, 3)
 
 
@@ -270,13 +413,15 @@ def test_dist_zero_pad_only_applies_global_width_edges(monkeypatch: pytest.Monke
     x = torch.ones((1, 1, 2, 2))
 
     monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (1, 3))
-    mid_rank_pad = wan_spatial_shard.WanDistZeroPad2d((1, 1, 0, 0), group=object(), split_dim="width")
-    mid = mid_rank_pad(x)
+    mid_rank_pad = wan_spatial_shard.WanDistZeroPad2d((1, 1, 0, 0), group=object())
+    with _active_spatial_shard("width"):
+        mid = mid_rank_pad(x)
     assert mid.shape == (1, 1, 2, 2)
 
     monkeypatch.setattr(wan_spatial_shard, "_rank_world", lambda group: (2, 3))
-    last_rank_pad = wan_spatial_shard.WanDistZeroPad2d((1, 1, 0, 0), group=object(), split_dim="width")
-    last = last_rank_pad(x)
+    last_rank_pad = wan_spatial_shard.WanDistZeroPad2d((1, 1, 0, 0), group=object())
+    with _active_spatial_shard("width"):
+        last = last_rank_pad(x)
     assert last.shape == (1, 1, 2, 3)
 
 
@@ -299,6 +444,111 @@ def test_spatial_shard_height_gate_falls_back_for_partial_group(monkeypatch: pyt
     assert vae._spatial_shard_decode_enabled(z) is False
 
 
+@pytest.mark.core_model
+@pytest.mark.cpu
+@pytest.mark.parametrize(
+    "z_shape,world_size,expected",
+    [
+        ((1, 16, 1, 60, 104), 4, "width"),
+        ((1, 16, 5, 60, 104), 4, "width"),
+        ((1, 16, 5, 104, 60), 4, "height"),
+        ((1, 16, 5, 2, 2), 4, None),
+        ((1, 16, 5, 64, 64), 4, "width"),
+        ((1, 16, 64, 64), 4, None),
+        ((1, 16, 5, 64, 64), 1, None),
+    ],
+)
+def test_auto_spatial_shard_selector_is_shape_aware(z_shape, world_size, expected):
+    assert select_auto_spatial_shard_split_dim(z_shape, world_size) == expected
+
+
+@pytest.mark.core_model
+@pytest.mark.cpu
+def test_auto_spatial_shard_gate_requires_full_group(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(
+        "vllm_omni.diffusion.distributed.autoencoders.autoencoder_kl_wan.dist.get_world_size",
+        lambda group=None: 4,
+    )
+
+    vae = DistributedAutoencoderKLWan.__new__(DistributedAutoencoderKLWan)
+    vae.use_tiling = True
+    vae.distributed_executor = SimpleNamespace(group=object(), parallel_size=4, parallel_mode="auto")
+    vae.is_distributed_enabled = lambda: True
+    large = torch.zeros((1, 16, 5, 60, 104))
+    unshardable = torch.zeros((1, 16, 1, 2, 2))
+    unsupported = torch.zeros((1, 16, 60, 104))
+
+    assert vae._spatial_shard_decode_split_dim(large, 4) == "width"
+    assert vae._spatial_shard_decode_enabled(large) is True
+    assert vae._spatial_shard_decode_enabled(unshardable) is False
+    assert vae._spatial_shard_decode_enabled(unsupported) is False
+
+    vae.distributed_executor.parallel_size = 2
+    assert vae._spatial_shard_decode_enabled(large) is False
+
+
+@pytest.mark.core_model
+@pytest.mark.cpu
+def test_set_parallel_size_prepares_spatial_wrappers_only_for_a_full_group(monkeypatch: pytest.MonkeyPatch):
+    vae = DistributedAutoencoderKLWan.__new__(DistributedAutoencoderKLWan)
+    executor = SimpleNamespace(group=object(), parallel_size=1, parallel_mode="tile")
+
+    def set_parallel_size(parallel_size: int, mode: str = "tile") -> None:
+        executor.parallel_size = parallel_size
+        executor.parallel_mode = mode
+
+    executor.set_parallel_size = set_parallel_size
+    object.__setattr__(vae, "distributed_executor", executor)
+    installed = []
+
+    monkeypatch.setattr(
+        "vllm_omni.diffusion.distributed.autoencoders.autoencoder_kl_wan.dist.get_world_size",
+        lambda group=None: 4,
+    )
+    monkeypatch.setattr(
+        wan_spatial_shard,
+        "install_wan_spatial_shard_decode",
+        lambda candidate, group, split_dim: installed.append((candidate, group, split_dim)),
+    )
+
+    vae.set_parallel_size(2, mode="auto")
+    assert installed == []
+
+    vae.set_parallel_size(4, mode="auto")
+    assert installed == [(vae, vae.distributed_executor.group, "height")]
+
+
+@pytest.mark.core_model
+@pytest.mark.cpu
+@torch.no_grad()
+def test_installed_spatial_decoder_keeps_direct_fallback_and_can_change_axis():
+    vae = DistributedAutoencoderKLWan(
+        base_dim=8,
+        decoder_base_dim=8,
+        z_dim=4,
+        dim_mult=[1, 2],
+        num_res_blocks=1,
+        attn_scales=[],
+        temperal_downsample=[False],
+        latents_mean=[0.0] * 4,
+        latents_std=[1.0] * 4,
+        scale_factor_temporal=1,
+        scale_factor_spatial=2,
+    ).eval()
+    latents = torch.randn(1, 4, 1, 4, 5)
+    parameter_names = {name for name, _ in vae.named_parameters()}
+    state_dict = {name: value.clone() for name, value in vae.state_dict().items()}
+    reference = vae.decode(latents, return_dict=False)[0]
+
+    wan_spatial_shard.install_wan_spatial_shard_decode(vae, group=object(), split_dim="height")
+    assert any(getattr(module, "_vllm_omni_dynamic_spatial_shard_conv", False) for module in vae.decoder.modules())
+    wan_spatial_shard.install_wan_spatial_shard_decode(vae, group=object(), split_dim="width")
+
+    assert {name for name, _ in vae.named_parameters()} == parameter_names
+    vae.load_state_dict(state_dict, strict=True)
+    assert torch.equal(vae.decode(latents, return_dict=False)[0], reference)
+
+
 @pytest.mark.core_model
 @pytest.mark.cpu
 def test_spatial_shard_width_gate_selects_width(monkeypatch: pytest.MonkeyPatch):
@@ -444,3 +694,99 @@ def test_spatial_shard_decode_matches_reference(split_dim: str):
     assert mean_abs_diff <= _SPATIAL_SHARD_TOLERANCE, (
         f"spatial_shard_{split_dim} mean_abs_diff {mean_abs_diff} exceeds {_SPATIAL_SHARD_TOLERANCE}"
     )
+
+
+def _auto_spatial_shard_decode_worker(rank: int, return_dict, master_port: str) -> None:
+    from vllm_omni.diffusion.distributed.parallel_state import (
+        destroy_model_parallel,
+        init_distributed_environment,
+        initialize_model_parallel,
+    )
+
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
+    os.environ["MASTER_PORT"] = master_port
+    device = current_omni_platform.get_torch_device(rank)
+    current_omni_platform.set_device(device)
+    backend = current_omni_platform.dist_backend
+    init_distributed_environment(world_size=2, rank=rank, local_rank=rank, backend=backend)
+    initialize_model_parallel(sequence_parallel_size=2, ulysses_degree=2, backend=backend)
+
+    try:
+        vae = DistributedAutoencoderKLWan.from_pretrained(
+            _SPATIAL_SHARD_MODEL,
+            subfolder=_SPATIAL_SHARD_SUBFOLDER,
+            torch_dtype=torch.float32,
+        ).to(device=device, dtype=torch.float32)
+        vae.eval()
+        generator = torch.Generator(device=device).manual_seed(1)
+        small = torch.randn((1, vae.config.z_dim, 1, 60, 104), generator=generator, device=device)
+        landscape = torch.randn((1, vae.config.z_dim, 2, 60, 104), generator=generator, device=device)
+        portrait = torch.randn((1, vae.config.z_dim, 2, 104, 60), generator=generator, device=device)
+
+        with torch.inference_mode():
+            vae.use_tiling = False
+            vae.set_parallel_size(1, mode="tile")
+            small_reference = vae.decode(small, return_dict=False)[0]
+
+            vae.use_tiling = True
+            vae.set_parallel_size(2, mode="tile")
+            tile_before_install = vae.decode(small, return_dict=False)[0]
+
+            vae.set_parallel_size(2, mode="auto")
+            assert vae._spatial_shard_decode_split_dim(small, 2) == "width"
+            small_sharded = vae.decode(small, return_dict=False)[0]
+            assert vae._spatial_shard_decode_split_dim(landscape, 2) == "width"
+            vae.decode(landscape, return_dict=False)
+
+            vae.set_parallel_size(2, mode="tile")
+            tile_after_install = vae.decode(small, return_dict=False)[0]
+
+            vae.use_tiling = False
+            vae.set_parallel_size(1, mode="tile")
+            portrait_reference = vae.decode(portrait, return_dict=False)[0]
+
+            vae.use_tiling = True
+            vae.set_parallel_size(2, mode="auto")
+            assert vae._spatial_shard_decode_split_dim(portrait, 2) == "height"
+            portrait_sharded = vae.decode(portrait, return_dict=False)[0]
+
+        assert all(
+            buffer is None
+            for module in vae.decoder.modules()
+            for name, buffer in module._buffers.items()
+            if name in {"_halo_recv_top_buf", "_halo_recv_bottom_buf"}
+        )
+
+        if rank == 0:
+            small_diff = (small_sharded.float() - small_reference.float()).abs()
+            portrait_diff = (portrait_sharded.float() - portrait_reference.float()).abs()
+            return_dict["tile_fallback_exact"] = torch.equal(tile_after_install, tile_before_install)
+            return_dict["small_max_abs_diff"] = small_diff.max().item()
+            return_dict["small_mean_abs_diff"] = small_diff.mean().item()
+            return_dict["portrait_max_abs_diff"] = portrait_diff.max().item()
+            return_dict["portrait_mean_abs_diff"] = portrait_diff.mean().item()
+    finally:
+        destroy_model_parallel()
+        if dist.is_initialized():
+            dist.destroy_process_group()
+
+
+@pytest.mark.full_model
+@pytest.mark.diffusion
+@pytest.mark.parallel
+@hardware_test(res={"cuda": "H100"}, num_cards=2)
+def test_auto_spatial_shard_switches_shape_paths_safely():
+    manager = mp.get_context("spawn").Manager()
+    return_dict = manager.dict()
+    mp.spawn(
+        _auto_spatial_shard_decode_worker,
+        args=(return_dict, "29502"),
+        nprocs=2,
+        join=True,
+    )
+
+    assert return_dict.get("tile_fallback_exact") is True
+    assert return_dict["small_max_abs_diff"] <= _SPATIAL_SHARD_TOLERANCE
+    assert return_dict["small_mean_abs_diff"] <= _SPATIAL_SHARD_TOLERANCE
+    assert return_dict["portrait_max_abs_diff"] <= _SPATIAL_SHARD_TOLERANCE
+    assert return_dict["portrait_mean_abs_diff"] <= _SPATIAL_SHARD_TOLERANCE

tests/diffusion/lora/test_diffusion_lora.py

@@ -29,10 +29,10 @@ def test_diffusion_model(tmp_path: Path, omni_runner):
             raise ValueError("Empty outputs from Omni.generate()")
         first_output = outputs[0]
         assert first_output.final_output_type == "image"
-        if not hasattr(first_output, "request_output") or not first_output.request_output:
+        if not isinstance(first_output, OmniRequestOutput) or not first_output:
             raise ValueError("No request_output found in OmniRequestOutput")
 
-        req_out = first_output.request_output
+        req_out = first_output
         if not isinstance(req_out, OmniRequestOutput) or not hasattr(req_out, "images"):
             raise ValueError("Invalid request_output structure or missing 'images' key")
         return req_out.images

tests/diffusion/models/minimax_h3/test_minimax_h3_rmsnorm.py

@@ -0,0 +1,26 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import torch
+
+from vllm_omni.diffusion.layers.norm import RMSNorm
+from vllm_omni.diffusion.models.minimax_h3.encoder import MiniMaxH3Qwen3VLRMSNorm
+
+
+def test_qwen3_vl_rmsnorm_uses_common_fused_rmsnorm_contract() -> None:
+    """Qwen keeps BF16 gamma while native fallback accumulates in FP32."""
+    eps = 1e-6
+    norm = MiniMaxH3Qwen3VLRMSNorm(hidden_size=4, eps=eps, dtype=torch.bfloat16)
+    norm.weight.data.copy_(torch.tensor([1.0, 0.5, 1.5, 2.0], dtype=torch.bfloat16))
+    x = torch.tensor(
+        [[1.0, -2.0, 3.0, -4.0], [-5.0, 6.0, -7.0, 8.0]],
+        dtype=torch.bfloat16,
+    )
+
+    x_fp32 = x.float()
+    expected = (x_fp32 * torch.rsqrt(x_fp32.square().mean(-1, keepdim=True) + eps) * norm.weight.float()).to(x.dtype)
+
+    assert isinstance(norm, RMSNorm)
+    assert norm.weight.dtype == torch.bfloat16
+    assert set(norm.state_dict()) == {"weight"}
+    torch.testing.assert_close(norm.forward_native(x), expected, atol=0, rtol=0)

tests/diffusion/offloader/test_diffusion_cpu_offload.py

@@ -122,8 +122,8 @@ def test_cpu_offload_diffusion_model(model_name: str):
     print(f"No offload peak memory: {no_offload_peak_memory} MB")
 
     if model_name == "stabilityai/stable-audio-open-1.0":
-        audio_offload = output_offload[0].request_output.multimodal_output.get("audio")
-        audio_no_offload = output_no_offload[0].request_output.multimodal_output.get("audio")
+        audio_offload = output_offload[0].multimodal_output.get("audio")
+        audio_no_offload = output_no_offload[0].multimodal_output.get("audio")
         check_audio_determinism(audio_offload, audio_no_offload, atol=1e-2)
 
     # Set platform-specific VRAM saving thresholds to account

tests/diffusion/offloader/test_diffusion_layerwise_offload.py

@@ -20,18 +20,10 @@ IMAGE_VIDEO_MODELS = {
 
 MODELS = {**AUDIO_MODEL, **IMAGE_VIDEO_MODELS}
 
-LAYERWISE_MODEL_CASES = [
-    pytest.param(
-        "stabilityai/stable-audio-open-1.0",
-        marks=[pytest.mark.full_model, pytest.mark.diffusion],
-        id="stable_audio",
-    ),
-    pytest.param(
-        "riverclouds/qwen_image_random",
-        marks=[pytest.mark.core_model, pytest.mark.diffusion],
-        id="qwen_image_random",
-    ),
-]
+MODEL_MARKS = {
+    "riverclouds/qwen_image_random": pytest.mark.core_model,
+    "stabilityai/stable-audio-open-1.0": pytest.mark.full_model,
+}
 
 AUDIO_MODEL_PARAMS = {
     "runner_params": {},
@@ -101,8 +93,9 @@ def run_inference(
     return peak, output
 
 
+@pytest.mark.diffusion
 @hardware_test(res={"cuda": "L4", "rocm": "MI325"})
-@pytest.mark.parametrize("model_name", LAYERWISE_MODEL_CASES)
+@pytest.mark.parametrize("model_name", list(MODELS.keys()))
 def test_layerwise_offload_diffusion_model(model_name: str):
     """Test that layerwise offloading reduces GPU memory usage.
 
@@ -134,8 +127,8 @@ def test_layerwise_offload_diffusion_model(model_name: str):
     print(f"No offload peak memory: {no_offload_peak_memory} MB")
 
     if model_name == "stabilityai/stable-audio-open-1.0":
-        audio_offload = output_offload[0].request_output.multimodal_output.get("audio")
-        audio_no_offload = output_no_offload[0].request_output.multimodal_output.get("audio")
+        audio_offload = output_offload[0].multimodal_output.get("audio")
+        audio_no_offload = output_no_offload[0].multimodal_output.get("audio")
         # Match the sibling cpu-offload test's tolerance: layerwise offload moves
         # blocks across the PCIe bus on a side stream, which can perturb cuBLAS
         # algorithm selection and produce ~ULP-level drift larger than 1e-3.

tests/diffusion/quantization/test_gguf_diffusion.py

@@ -94,8 +94,8 @@ def _generate_single_stage_image(
     if hasattr(first_output, "images") and first_output.images:
         images = first_output.images
     else:
-        assert hasattr(first_output, "request_output") and first_output.request_output
-        request_output = first_output.request_output
+        assert isinstance(first_output, OmniRequestOutput) and first_output
+        request_output = first_output
         if isinstance(request_output, list):
             req_out = request_output[0]
         else:

tests/diffusion/quantization/test_quantization_fp8.py

@@ -88,8 +88,8 @@ def _generate_single_stage_image(
         if hasattr(first_output, "images") and first_output.images:
             images = first_output.images
         else:
-            assert hasattr(first_output, "request_output") and first_output.request_output
-            request_output = first_output.request_output
+            assert isinstance(first_output, OmniRequestOutput) and first_output
+            request_output = first_output
             if isinstance(request_output, list):
                 req_out = request_output[0]
             else:
@@ -147,10 +147,10 @@ def _generate_single_stage_video(
 
         first = outputs[0]
 
-        # Unwrap pipeline-style outputs (multi-stage / OmniRequestOutput.request_output).
+        # Unwrap pipeline-style outputs (multi-stage / OmniRequestOutput).
         frames: Any = None
-        if hasattr(first, "request_output") and isinstance(first.request_output, list):
-            inner = first.request_output[0]
+        if isinstance(first, OmniRequestOutput) and isinstance(first, list):
+            inner = first[0]
             if isinstance(inner, OmniRequestOutput) and inner.images:
                 frames = inner.images[0]
         if frames is None and hasattr(first, "images") and first.images:
@@ -189,7 +189,7 @@ def _generate_single_stage_video(
 
 
 def _generate_bagel_image(
-    quantization_config: str | None = None,
+    diffusion_quantization_config: str | None = None,
     num_inference_steps: int = 15,
 ) -> tuple[Any, float]:
     """Generate an image with BAGEL (multi-stage: LLM + Diffusion).
@@ -199,11 +199,11 @@ def _generate_bagel_image(
     config_path = get_deploy_config_path("ci/bagel.yaml")
     omni_kwargs: dict[str, Any] = {
         "model": "ByteDance-Seed/BAGEL-7B-MoT",
-        "stage_configs_path": config_path,
+        "deploy_config": config_path,
         "stage_init_timeout": 300,
     }
-    if quantization_config:
-        omni_kwargs["quantization_config"] = quantization_config
+    if diffusion_quantization_config:
+        omni_kwargs["diffusion_quantization_config"] = diffusion_quantization_config
 
     model_name = omni_kwargs.pop("model")
     with OmniRunner(model_name, **omni_kwargs) as runner:
@@ -234,8 +234,8 @@ def _generate_bagel_image(
             if images := getattr(req_output, "images", None):
                 generated_image = images[0]
                 break
-            if hasattr(req_output, "request_output") and req_output.request_output:
-                stage_outputs = req_output.request_output
+            if isinstance(req_output, OmniRequestOutput) and req_output:
+                stage_outputs = req_output
                 if not isinstance(stage_outputs, list):
                     stage_outputs = [stage_outputs]
                 for stage_out in stage_outputs:
@@ -250,8 +250,8 @@ def _generate_bagel_image(
 
         # Check LLM stage output — should have finish_reason=stop (not length)
         for req_output in omni_outputs:
-            if hasattr(req_output, "request_output") and req_output.request_output:
-                stage_outputs = req_output.request_output
+            if isinstance(req_output, OmniRequestOutput) and req_output:
+                stage_outputs = req_output
                 if not isinstance(stage_outputs, list):
                     stage_outputs = [stage_outputs]
                 for stage_out in stage_outputs:
@@ -380,21 +380,21 @@ def test_single_stage_ltx2_fp8_uses_less_memory():
 
 @hardware_test(res={"cuda": "H100"})
 def test_bagel_fp8_generates_image():
-    """BAGEL with FP8 quantization_config generates a valid image.
+    """BAGEL with diffusion-stage FP8 generates a valid image.
 
     FP8 should only apply to the diffusion stage (Stage-1), not the
     LLM stage (Stage-0). We verify this by checking:
       1. Image is generated successfully
       2. LLM stage finish_reason is 'stop' (not 'length' from garbled output)
     """
-    image, _ = _generate_bagel_image(quantization_config="fp8")
+    image, _ = _generate_bagel_image(diffusion_quantization_config="fp8")
     image.save("test_bagel_fp8.png")
 
 
 @hardware_test(res={"cuda": "H100"})
 def test_bagel_bf16_generates_image():
     """BAGEL without quantization generates a valid image (baseline)."""
-    image, _ = _generate_bagel_image(quantization_config=None)
+    image, _ = _generate_bagel_image(diffusion_quantization_config=None)
     image.save("test_bagel_bf16.png")
 
 

tests/diffusion/quantization/test_quantization_quality.py

@@ -310,7 +310,7 @@ def _generate_image(omni, config: QualityTestConfig):
     first = outputs[0]
     if hasattr(first, "images") and first.images:
         return first.images[0], peak_mem
-    inner = first.request_output
+    inner = first
     if inner is not None and hasattr(inner, "images") and inner.images:
         return inner.images[0], peak_mem
     raise ValueError("Could not extract image from output.")
@@ -341,15 +341,15 @@ def _generate_video(omni, config: QualityTestConfig):
     )
 
     peak_mem = torch.accelerator.max_memory_allocated() / (1024**3)
-    first = outputs[0]
-    if hasattr(first, "request_output") and isinstance(first.request_output, list):
-        inner = first.request_output[0]
-        if isinstance(inner, OmniRequestOutput) and hasattr(inner, "images"):
-            frames = inner.images[0] if inner.images else None
-        else:
-            frames = inner
-    elif hasattr(first, "images") and first.images:
-        frames = first.images[0]
+    if isinstance(outputs, list) and isinstance(outputs[0], OmniRequestOutput):
+        first = outputs[0]
+    elif isinstance(outputs, OmniRequestOutput):
+        first = outputs
+    else:
+        raise ValueError("Could not extract video frames from output.")
+
+    if hasattr(first, "images"):
+        frames = first.images[0] if first.images else None
     else:
         raise ValueError("Could not extract video frames from output.")
 
@@ -444,11 +444,11 @@ def test_benchmark_generate_image_unwraps_nested_omni_request_output(monkeypatch
 
     image = Image.new("RGB", (2, 2))
     inner = OmniRequestOutput.from_diffusion(request_id="req", images=[image])
-    outer = OmniRequestOutput(
+    outer = OmniRequestOutput.from_stage_output(
+        inner,
         request_id="req",
         stage_id=0,
         final_output_type="image",
-        request_output=inner,
         finished=True,
     )
 

tests/diffusion/quantization/test_trajectory_similarity_tool.py

@@ -73,11 +73,9 @@ def test_run_summary_reports_worker_peak_memory():
     assert summary["per_run_peak_memory_mb"] == [100.0, 150.0]
 
 
-def test_request_peak_memory_prefers_inner_diffusion_output():
-    inner = type("InnerOutput", (), {"peak_memory_mb": 321.0})()
-    outer = type("OuterOutput", (), {"request_output": inner, "peak_memory_mb": 123.0})()
-
-    assert _request_peak_memory_mb(outer) == 321.0
+def test_request_peak_memory_reads_peak_memory_mb_directly():
+    result = type("Output", (), {"peak_memory_mb": 123.0})()
+    assert _request_peak_memory_mb(result) == 123.0
 
 
 def test_metric_guidance_describes_thresholds():