fix

Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>

.github/ISSUE_TEMPLATE/1-bug-report.yml

@@ -8,7 +8,7 @@ body:
   attributes:
     label: Environment
     description: |
-      Please share your environment with us. You can run the command **python fastvideo/utils/env_utils.py** and copy-paste its output below.
+      Please share your environment with us. You can run the command **python fastvideo/utils/collect_env.py** and copy-paste its output below.
     placeholder: FastVideo version, platform, python version, cuda version...
   validations:
     required: true

.github/workflows/build-image-template.yml

@@ -0,0 +1,106 @@
+name: Build Image Template
+
+on:
+  workflow_call:
+    inputs:
+      python_version:
+        required: true
+        type: string
+      dockerfile_path:
+        required: true
+        type: string
+      tag_suffix:
+        required: true
+        type: string
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+    
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+      
+      - name: Free up disk space
+        run: |
+          # Display initial space
+          echo "Initial disk space:"
+          df -h
+          
+          # Remove large directories directly
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /usr/local/lib/android
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /usr/local/share/boost
+          sudo rm -rf /usr/share/swift
+          sudo rm -rf /usr/local/lib/node_modules
+          sudo rm -rf /usr/local/share/powershell
+          sudo rm -rf /usr/share/rust
+          sudo rm -rf /usr/local/.ghcup
+          
+          # Remove cached files
+          sudo rm -rf /var/lib/apt/lists/*
+          sudo rm -rf /var/cache/apt/archives/*
+          
+          # Clean Docker
+          docker system prune -af --volumes
+          
+          # Display available space after cleanup
+          echo "Disk space after cleanup:"
+          df -h
+      
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      
+      - name: Login to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      
+      - name: Prepare tags
+        id: prepare-tags
+        run: |
+          SHORT_SHA=$(echo ${{ github.sha }} | cut -c1-7)
+          
+          TAGS="type=raw,value=${{ inputs.tag_suffix }}-latest"
+          TAGS="${TAGS}\ntype=raw,value=${{ inputs.tag_suffix }}-sha-${SHORT_SHA}"
+          
+          # Set Python 3.10 as the default image
+          if [[ "${{ inputs.python_version }}" == "3.10" ]]; then
+            TAGS="${TAGS}\ntype=raw,value=latest"
+          fi
+          
+          {
+            echo "tags<<EOF"
+            echo -e "$TAGS"
+            echo "EOF" 
+          } >> $GITHUB_OUTPUT
+      
+      - name: Extract metadata for Docker
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ghcr.io/${{ github.repository }}/fastvideo-dev
+          tags: ${{ steps.prepare-tags.outputs.tags }}
+      
+      - name: Build and push Docker image
+        id: build-push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ${{ inputs.dockerfile_path }}
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+      
+      - name: Success message
+        run: |
+          echo "✅ Python ${{ inputs.python_version }} image successfully built and pushed to ghcr.io/${{ github.repository }}/fastvideo-dev:${{ inputs.tag_suffix }}-latest"
+          echo "To run tests with this image, manually trigger the 'Run Tests' workflow." 

.github/workflows/build-image.yml

@@ -1,78 +1,52 @@
-name: Build and Push Docker Image
+name: Build and Push Docker Images
 
 on:
-  workflow_dispatch:  # Only manual triggers
+  workflow_dispatch:
+    inputs:
+      python_3_10:
+        description: 'Build Python 3.10 image'
+        required: false
+        default: false
+        type: boolean
+      python_3_11:
+        description: 'Build Python 3.11 image'
+        required: false
+        default: false
+        type: boolean
+      python_3_12:
+        description: 'Build Python 3.12 image'
+        required: false
+        default: false
+        type: boolean
+
+permissions:
+  contents: read
+  packages: write
 
 jobs:
-  build-and-push:
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      packages: write
-    
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-      
-      - name: Free up disk space
-        run: |
-          # Display initial space
-          echo "Initial disk space:"
-          df -h
-          
-          # Remove large directories directly
-          sudo rm -rf /usr/share/dotnet
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /opt/ghc
-          sudo rm -rf /usr/local/share/boost
-          sudo rm -rf /usr/share/swift
-          sudo rm -rf /usr/local/lib/node_modules
-          sudo rm -rf /usr/local/share/powershell
-          sudo rm -rf /usr/share/rust
-          sudo rm -rf /usr/local/.ghcup
-          
-          # Remove cached files
-          sudo rm -rf /var/lib/apt/lists/*
-          sudo rm -rf /var/cache/apt/archives/*
-          
-          # Clean Docker
-          docker system prune -af --volumes
-          
-          # Display available space after cleanup
-          echo "Disk space after cleanup:"
-          df -h
-      
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      
-      - name: Login to GitHub Container Registry
-        uses: docker/login-action@v3
-        with:
-          registry: ghcr.io
-          username: ${{ github.repository_owner }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-      
-      - name: Extract metadata for Docker
-        id: meta
-        uses: docker/metadata-action@v5
-        with:
-          images: ghcr.io/${{ github.repository }}/fastvideo-dev
-          tags: |
-            type=raw,value=latest
-            type=sha,format=short
-      
-      - name: Build and push Docker image
-        id: build-push
-        uses: docker/build-push-action@v6
-        with:
-          context: .
-          push: true
-          tags: ${{ steps.meta.outputs.tags }}
-          labels: ${{ steps.meta.outputs.labels }}
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
-      
-      - name: Success message
-        run: |
-          echo "✅ Image successfully built and pushed to ghcr.io/${{ github.repository }}/fastvideo-dev:latest"
-          echo "To run tests with this image, manually trigger the 'Run Tests' workflow." 
+  build-python-3-10:
+    if: ${{ github.event.inputs.python_3_10 == 'true' }}
+    uses: ./.github/workflows/build-image-template.yml
+    with:
+      python_version: '3.10'
+      dockerfile_path: docker/Dockerfile.python3.10
+      tag_suffix: py3.10
+    secrets: inherit
+
+  build-python-3-11:
+    if: ${{ github.event.inputs.python_3_11 == 'true' }}
+    uses: ./.github/workflows/build-image-template.yml
+    with:
+      python_version: '3.11'
+      dockerfile_path: docker/Dockerfile.python3.11
+      tag_suffix: py3.11
+    secrets: inherit
+
+  build-python-3-12:
+    if: ${{ github.event.inputs.python_3_12 == 'true' }}
+    uses: ./.github/workflows/build-image-template.yml
+    with:
+      python_version: '3.12'
+      dockerfile_path: docker/Dockerfile.python3.12
+      tag_suffix: py3.12
+    secrets: inherit

.github/workflows/pr-test.yml

@@ -77,199 +77,89 @@ jobs:
               - 'fastvideo/v1/models/dits/**'
               - 'fastvideo/v1/models/loaders/**'
               - 'fastvideo/v1/tests/transformers/**'
+              - 'fastvideo/v1/layers/**'
+              - 'fastvideo/v1/attention/**'
 
   encoder-test:
     needs: change-filter
     if: >-
       (github.event_name != 'workflow_dispatch' && needs.change-filter.outputs.encoder-test == 'true') || 
       (github.event_name == 'workflow_dispatch' && github.event.inputs.run_encoder_test == 'true')
-    runs-on: ubuntu-latest
-    environment: runpod-runners
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.10"
-
-      - name: Set up SSH key
-        run: |
-          mkdir -p ~/.ssh
-          echo "${{ secrets.RUNPOD_PRIVATE_KEY }}" > ~/.ssh/id_rsa
-          chmod 600 ~/.ssh/id_rsa
-          ssh-keygen -y -f ~/.ssh/id_rsa > ~/.ssh/id_rsa.pub
-
-      - name: Install dependencies
-        run: pip install requests
-
-      - name: Run tests on RunPod
-        env:
-          JOB_ID: "encoder-test"
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-        timeout-minutes: 30
-        run: >-
-          python .github/scripts/runpod_api.py 
-          --gpu-type "NVIDIA A40"
-          --gpu-count 1
-          --volume-size 100
-          --image "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
-          --test-command "pip install -e .[test] && pytest ./fastvideo/v1/tests/encoders -s"
-
-      - name: Terminate RunPod Instances
-        if: ${{ always() }}
-        env:
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-          JOB_ID: "encoder-test"
-        run: python .github/scripts/runpod_cleanup.py
+    uses: ./.github/workflows/runpod-test.yml
+    with:
+      job_id: "encoder-test"
+      gpu_type: "NVIDIA A40"
+      gpu_count: 1
+      volume_size: 100
+      image: "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
+      test_command: "pip install -e .[test] && pytest ./fastvideo/v1/tests/encoders -s"
+      timeout_minutes: 30
+    secrets:
+      RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+      RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
 
   vae-test:
     needs: change-filter
     if: >-
       (github.event_name != 'workflow_dispatch' && needs.change-filter.outputs.vae-test == 'true') || 
       (github.event_name == 'workflow_dispatch' && github.event.inputs.run_vae_test == 'true')
-    runs-on: ubuntu-latest
-    environment: runpod-runners
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.10"
-
-      - name: Set up SSH key
-        run: |
-          mkdir -p ~/.ssh
-          echo "${{ secrets.RUNPOD_PRIVATE_KEY }}" > ~/.ssh/id_rsa
-          chmod 600 ~/.ssh/id_rsa
-          ssh-keygen -y -f ~/.ssh/id_rsa > ~/.ssh/id_rsa.pub
-
-      - name: Install dependencies
-        run: pip install requests
-
-      - name: Run tests on RunPod
-        env:
-          JOB_ID: "vae-test"
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-        timeout-minutes: 30
-        run: >-
-          python .github/scripts/runpod_api.py 
-          --gpu-type "NVIDIA A40"
-          --gpu-count 1
-          --volume-size 100
-          --image "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
-          --test-command "pip install -e .[test] && pytest ./fastvideo/v1/tests/vaes -s"
-
-      - name: Terminate RunPod Instances
-        if: ${{ always() }}
-        env:
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-          JOB_ID: "vae-test"
-        run: python .github/scripts/runpod_cleanup.py
+    uses: ./.github/workflows/runpod-test.yml
+    with:
+      job_id: "vae-test"
+      gpu_type: "NVIDIA A40"
+      gpu_count: 1
+      volume_size: 100
+      image: "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
+      test_command: "pip install -e .[test] && pytest ./fastvideo/v1/tests/vaes -s"
+      timeout_minutes: 30
+    secrets:
+      RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+      RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
 
   transformer-test:
     needs: change-filter
     if: >-
       (github.event_name != 'workflow_dispatch' && needs.change-filter.outputs.transformer-test == 'true') || 
       (github.event_name == 'workflow_dispatch' && github.event.inputs.run_transformer_test == 'true')
-    runs-on: ubuntu-latest
-    environment: runpod-runners
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.10"
-
-      - name: Set up SSH key
-        run: |
-          mkdir -p ~/.ssh
-          echo "${{ secrets.RUNPOD_PRIVATE_KEY }}" > ~/.ssh/id_rsa
-          chmod 600 ~/.ssh/id_rsa
-          ssh-keygen -y -f ~/.ssh/id_rsa > ~/.ssh/id_rsa.pub
-
-      - name: Install dependencies
-        run: pip install requests
-
-      - name: Run tests on RunPod
-        env:
-          JOB_ID: "transformer-test"
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-        timeout-minutes: 30
-        run: >-
-          python .github/scripts/runpod_api.py 
-          --gpu-type "NVIDIA L40S"
-          --gpu-count 1
-          --volume-size 100
-          --image "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
-          --test-command "pip install -e .[test] && pytest ./fastvideo/v1/tests/transformers -s"
-
-      - name: Terminate RunPod Instances
-        if: ${{ always() }}
-        env:
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-          JOB_ID: "transformer-test"
-        run: python .github/scripts/runpod_cleanup.py
+    uses: ./.github/workflows/runpod-test.yml
+    with:
+      job_id: "transformer-test"
+      gpu_type: "NVIDIA L40S"
+      gpu_count: 1
+      volume_size: 100
+      image: "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
+      test_command: "pip install -e .[test] && pytest ./fastvideo/v1/tests/transformers -s"
+      timeout_minutes: 30
+    secrets:
+      RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+      RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
       
   ssim-test:
     needs: change-filter
     if: >-
       (github.event_name != 'workflow_dispatch' && github.event.pull_request.draft == false) || 
       (github.event_name == 'workflow_dispatch' && github.event.inputs.run_ssim_test == 'true')
-    runs-on: ubuntu-latest
-    environment: runpod-runners
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.10"
-
-      - name: Set up SSH key
-        run: |
-          mkdir -p ~/.ssh
-          echo "${{ secrets.RUNPOD_PRIVATE_KEY }}" > ~/.ssh/id_rsa
-          chmod 600 ~/.ssh/id_rsa
-          ssh-keygen -y -f ~/.ssh/id_rsa > ~/.ssh/id_rsa.pub
-
-      - name: Install dependencies
-        run: pip install requests
-
-      - name: Run tests on RunPod
-        env:
-          JOB_ID: "ssim-test"
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-        timeout-minutes: 45
-        run: >-
-          python .github/scripts/runpod_api.py 
-          --gpu-type "NVIDIA A40"
-          --gpu-count 2
-          --disk-size 200
-          --volume-size 200
-          --image "ghcr.io/${{ github.repository }}/${{ github.event.inputs.custom_image || 'fastvideo-dev:latest' }}"
-          --test-command "pip install -e .[test] && pytest ./fastvideo/v1/tests/ssim -vs"
-
-      - name: Terminate RunPod Instances
-        if: ${{ always() }}
-        env:
-          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-          JOB_ID: "ssim-test"
-        run: python .github/scripts/runpod_cleanup.py
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: [
+          # {version: "3.10", tag: "latest"},
+          # {version: "3.11", tag: "py3.11-latest"},
+          {version: "3.12", tag: "py3.12-latest"}
+        ]
+    uses: ./.github/workflows/runpod-test.yml
+    with:
+      job_id: "ssim-test-py${{ matrix.python-version.version }}"
+      gpu_type: "NVIDIA A40"
+      gpu_count: 2
+      volume_size: 200
+      disk_size: 200
+      image: "ghcr.io/${{ github.repository }}/fastvideo-dev:${{ matrix.python-version.tag }}"
+      test_command: "pip install -e .[test] && pytest ./fastvideo/v1/tests/ssim -vs"
+      timeout_minutes: 60
+    secrets:
+      RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+      RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
 
   runpod-cleanup:
     needs: [encoder-test, vae-test, transformer-test, ssim-test] # Add other jobs to this list as you create them
@@ -289,7 +179,7 @@ jobs:
 
       - name: Cleanup all RunPod instances
         env:
-          JOB_IDS: '["encoder-test", "vae-test", "transformer-test", "ssim-test"]' # JSON array of job IDs
+          JOB_IDS: '["encoder-test", "vae-test", "transformer-test", "ssim-test-py3.10", "ssim-test-py3.11", "ssim-test-py3.12"]'
           RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
           GITHUB_RUN_ID: ${{ github.run_id }}
         run: python .github/scripts/runpod_cleanup.py

.github/workflows/runpod-test.yml

@@ -0,0 +1,91 @@
+name: RunPod Test
+
+on:
+  workflow_call:
+    inputs:
+      job_id:
+        required: true
+        type: string
+        description: "Unique identifier for this test job"
+      gpu_type:
+        required: true
+        type: string
+        description: "GPU type to use (e.g. NVIDIA A40, NVIDIA L40S)"
+      gpu_count:
+        required: true
+        type: number
+        description: "Number of GPUs to use"
+      volume_size:
+        required: false
+        type: number
+        default: 20
+        description: "Volume size in GB"
+      disk_size:
+        required: false
+        type: number
+        default: 20
+        description: "Disk size in GB"
+      image:
+        required: true
+        type: string
+        description: "Docker image to use"
+      test_command:
+        required: true
+        type: string
+        description: "Command to run tests"
+      timeout_minutes:
+        required: false
+        type: number
+        default: 30
+        description: "Timeout in minutes"
+    secrets:
+      RUNPOD_API_KEY:
+        required: true
+      RUNPOD_PRIVATE_KEY:
+        required: true
+
+jobs:
+  run-test:
+    runs-on: ubuntu-latest
+    environment: runpod-runners
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Set up SSH key
+        run: |
+          mkdir -p ~/.ssh
+          echo "${{ secrets.RUNPOD_PRIVATE_KEY }}" > ~/.ssh/id_rsa
+          chmod 600 ~/.ssh/id_rsa
+          ssh-keygen -y -f ~/.ssh/id_rsa > ~/.ssh/id_rsa.pub
+
+      - name: Install dependencies
+        run: pip install requests
+
+      - name: Run tests on RunPod
+        env:
+          JOB_ID: ${{ inputs.job_id }}
+          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+          GITHUB_RUN_ID: ${{ github.run_id }}
+        timeout-minutes: ${{ inputs.timeout_minutes }}
+        run: >-
+          python .github/scripts/runpod_api.py 
+          --gpu-type "${{ inputs.gpu_type }}"
+          --gpu-count ${{ inputs.gpu_count }}
+          --volume-size ${{ inputs.volume_size }}
+          --disk-size ${{ inputs.disk_size }}
+          --image "${{ inputs.image }}"
+          --test-command "${{ inputs.test_command }}"
+
+      - name: Terminate RunPod Instances
+        if: ${{ always() }}
+        env:
+          RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
+          GITHUB_RUN_ID: ${{ github.run_id }}
+          JOB_ID: ${{ inputs.job_id }}
+        run: python .github/scripts/runpod_cleanup.py 

.pre-commit-config.yaml

@@ -19,9 +19,11 @@ exclude: |
     fastvideo/sample/.*|
     fastvideo/train\.py|
     fastvideo/utils/.*|
-    fastvideo/v1/examples/.*|
+    examples/.*|
     .github/workflows/fastvideo-publish.yml|
-    .github/workflows/sta-publish.yml
+    .github/workflows/sta-publish.yml|
+    .github/workflows/build-image-template.yml|
+    docs/source/inference/support_matrix.md
   )
 repos:
 - repo: https://github.com/google/yapf

README.md

@@ -2,62 +2,109 @@
 <img src=assets/logo.jpg width="30%"/>
 </div>
 
-FastVideo is a lightweight framework for accelerating large video diffusion models.
+**FastVideo is a unified framework for accelerated video generation.**
+
+It features a clean, consistent API that works across popular video models, making it easier for developers to author new models and incorporate system- or kernel-level optimizations.
+With FastVideo's optimizations, you can achieve more than 3x inference improvement compared to other systems.
 
 <p align="center">
-    | <a href="https://hao-ai-lab.github.io/FastVideo"><b>Documentation</b></a> | 🤗 <a href="https://huggingface.co/FastVideo/FastHunyuan"  target="_blank"><b>FastHunyuan</b></a>  | 🤗 <a href="https://huggingface.co/FastVideo/FastMochi-diffusers" target="_blank"><b>FastMochi</b></a> | 🟣💬 <a href="https://join.slack.com/t/fastvideo/shared_invite/zt-2zf6ru791-sRwI9lPIUJQq1mIeB_yjJg" target="_blank"> <b>Slack</b> </a> |
+    | <a href="https://hao-ai-lab.github.io/FastVideo"><b>Documentation</b></a> | <a href="https://hao-ai-lab.github.io/FastVideo/inference/inference_quick_start.html"><b> Quick Start</b></a> | 🤗 <a href="https://huggingface.co/FastVideo/FastHunyuan"  target="_blank"><b>FastHunyuan</b></a>  | 🤗 <a href="https://huggingface.co/FastVideo/FastMochi-diffusers" target="_blank"><b>FastMochi</b></a> | 🟣💬 <a href="https://join.slack.com/t/fastvideo/shared_invite/zt-2zf6ru791-sRwI9lPIUJQq1mIeB_yjJg" target="_blank"> <b>Slack</b> </a> |
 </p>
 
-https://github.com/user-attachments/assets/79af5fb8-707c-4263-b153-9ab2a01d3ac1
+<div align="center">
+<img src=assets/perf.png width="90%"/>
+</div>
 
-FastVideo currently offers: (with more to come)
+## Key Features
 
-- [NEW!] V1 inference API available. Full announcement coming soon!
-- [Sliding Tile Attention](https://hao-ai-lab.github.io/blogs/sta/).
-- FastHunyuan and FastMochi: consistency distilled video diffusion models for 8x inference speedup.
-- First open distillation recipes for video DiT, based on [PCM](https://github.com/G-U-N/Phased-Consistency-Model).
-- Support distilling/finetuning/inferencing state-of-the-art open video DiTs: 1. Mochi 2. Hunyuan.
+FastVideo has the following features:
+- State-of-the-art performance optimizations for inference
+  - [Sliding Tile Attention](https://arxiv.org/pdf/2502.04507)
+  - [TeaCache](https://arxiv.org/pdf/2411.19108)
+  - [Sage Attention](https://arxiv.org/abs/2410.02367)
+- Cutting edge models
+  - Wan2.1 T2V, I2V
+  - HunyuanVideo
+  - FastHunyuan: consistency distilled video diffusion models for 8x inference speedup.
+  - StepVideo T2V
+- Distillation support
+  - Recipes for video DiT, based on [PCM](https://github.com/G-U-N/Phased-Consistency-Model).
+  - Support distilling/finetuning/inferencing state-of-the-art open video DiTs: 1. Mochi 2. Hunyuan.
 - Scalable training with FSDP, sequence parallelism, and selective activation checkpointing, with near linear scaling to 64 GPUs.
 - Memory efficient finetuning with LoRA, precomputed latent, and precomputed text embeddings.
 
-Dev in progress and highly experimental.
-
-## Change Log
-- ```2025/02/20```: FastVideo now supports STA on [StepVideo](https://github.com/stepfun-ai/Step-Video-T2V) with 3.4X speedup!
-- ```2025/02/18```: Release the inference code and kernel for [Sliding Tile Attention](https://hao-ai-lab.github.io/blogs/sta/).
-- ```2025/01/13```: Support Lora finetuning for HunyuanVideo.
-- ```2024/12/25```: Enable single 4090 inference for `FastHunyuan`, please rerun the installation steps to update the environment.
-- ```2024/12/17```: `FastVideo` v0.0.1 is released.
-
 ## Getting Started
+We recommend using an environment manager such as `Conda` to create a clean environment:
+
+```bash
+# Create and activate a new conda environment
+conda create -n fastvideo python=3.12
+conda activate fastvideo
+
+# Install FastVideo
+pip install fastvideo
+```
+
+Please see our [docs](https://hao-ai-lab.github.io/FastVideo/getting_started/installation.html) for more detailed installation instructions.
+
+## Inference
+### Generating Your First Video
+Here's a minimal example to generate a video using the default settings. Create a file called `example.py` with the following code:
+
+```python
+from fastvideo import VideoGenerator
+
+def main():
+    # Create a video generator with a pre-trained model
+    generator = VideoGenerator.from_pretrained(
+        "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        num_gpus=1,  # Adjust based on your hardware
+    )
+
+    # Define a prompt for your video
+    prompt = "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes wide with interest."
+
+    # Generate the video
+    video = generator.generate_video(
+        prompt,
+        return_frames=True,  # Also return frames from this call (defaults to False)
+        output_path="my_videos/",  # Controls where videos are saved
+        save_video=True
+    )
+
+if __name__ == '__main__':
+    main()
+```
+
+Run the script with:
+
+```bash
+python example.py
+```
+
+For a more detailed guide, please see our [inference quick start](https://hao-ai-lab.github.io/FastVideo/inference/inference_quick_start.html).
+
+### Other docs:
 
-- [Install FastVideo](https://hao-ai-lab.github.io/FastVideo/getting_started/installation.html)
 - [Design Overview](https://hao-ai-lab.github.io/FastVideo/design/overview.html)
 - [Contribution Guide](https://hao-ai-lab.github.io/FastVideo/getting_started/installation.html)
 
-### Inference
-- [Quick Start](https://hao-ai-lab.github.io/FastVideo/inference/examples/basic.html)
-- V1 Inference API Guide (Coming soon!)
-
-### Distillation and Finetuning
+## Distillation and Finetuning
 - [Distillation Guide](https://hao-ai-lab.github.io/FastVideo/training/distillation.html)
 - [Finetuning Guide](https://hao-ai-lab.github.io/FastVideo/training/finetuning.html)
 
-### Deprecated APIs
-- [V0 Inference (Deprecated)](https://hao-ai-lab.github.io/FastVideo/inference/v0_inference.html)
-
 ## 📑 Development Plan
 
 <!-- - More distillation methods -->
   <!-- - [ ] Add Distribution Matching Distillation -->
 - More models support
   <!-- - [ ] Add CogvideoX model -->
-  - [ ] Add StepVideo to V1
+  - [x] Add StepVideo to V1
 - Optimization features
-  - [ ] Teacache in V1
-  - [ ] SageAttention in V1
+  - [x] Teacache in V1
+  - [x] SageAttention in V1
 - Code updates
-  - [ ] V1 Configuration API
+  - [x] V1 Configuration API
   - [ ] Support Training in V1
   <!-- - [ ] fp8 support -->
   <!-- - [ ] faster load model and save model support -->

csrc/sliding_tile_attention/test/test_sta.py

@@ -2,7 +2,6 @@ import torch
 from flex_sta_ref import get_sliding_tile_attention_mask
 from st_attn import sliding_tile_attention
 from torch.nn.attention.flex_attention import flex_attention
-# from flash_attn_interface import flash_attn_func
 from tqdm import tqdm
 
 flex_attention = torch.compile(flex_attention, dynamic=False)
@@ -23,7 +22,7 @@ def h100_fwd_kernel_test(Q, K, V, kernel_size):
 def generate_tensor(shape, mean, std, dtype, device):
     tensor = torch.randn(shape, dtype=dtype, device=device)
 
-    magnitude = torch.norm(tensor, dim=-1, keepdim=True)
+    magnitude = torch.linalg.norm(tensor, dim=-1, keepdim=True)
     scaled_tensor = tensor * (torch.randn(magnitude.shape, dtype=dtype, device=device) * std + mean) / magnitude
 
     return scaled_tensor.contiguous()

docker/Dockerfile.python3.10

@@ -29,7 +29,7 @@ RUN echo "# Placeholder" > README.md
 
 RUN conda run -n fastvideo-dev pip install --no-cache-dir --upgrade pip && \
     conda run -n fastvideo-dev pip install --no-cache-dir .[dev] && \
-    conda run -n fastvideo-dev pip install --no-cache-dir flash-attn==2.7.0.post2 --no-build-isolation && \
+    conda run -n fastvideo-dev pip install --no-cache-dir flash-attn==2.7.4.post1 --no-build-isolation && \
     conda clean -afy
 
 COPY . .

docker/Dockerfile.python3.11

@@ -0,0 +1,48 @@
+FROM nvidia/cuda:12.4.1-devel-ubuntu20.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+WORKDIR /FastVideo
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    wget \
+    git \
+    ca-certificates \
+    openssh-server \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
+    bash Miniconda3-latest-Linux-x86_64.sh -b -p /opt/conda && \
+    rm Miniconda3-latest-Linux-x86_64.sh
+
+ENV PATH=/opt/conda/bin:$PATH
+
+RUN conda create --name fastvideo-dev python=3.11.11 -y
+
+SHELL ["/bin/bash", "-c"]
+
+# Copy just the pyproject.toml first to leverage Docker cache
+COPY pyproject.toml ./
+
+# Create a dummy README to satisfy the installation
+RUN echo "# Placeholder" > README.md
+
+RUN conda run -n fastvideo-dev pip install --no-cache-dir --upgrade pip && \
+    conda run -n fastvideo-dev pip install --no-cache-dir .[dev] && \
+    conda run -n fastvideo-dev pip install --no-cache-dir flash-attn==2.7.4.post1 --no-build-isolation && \
+    conda clean -afy
+
+COPY . .
+
+RUN conda run -n fastvideo-dev pip install --no-cache-dir -e .[dev]
+
+# Remove authentication headers
+RUN git config --unset-all http.https://github.com/.extraheader || true
+
+# Set up automatic conda environment activation for all shells
+RUN echo 'source /opt/conda/etc/profile.d/conda.sh' >> /root/.bashrc && \
+    echo 'conda activate fastvideo-dev' >> /root/.bashrc && \
+    # Ensure .bashrc is sourced for SSH login shells
+    echo 'if [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile
+
+EXPOSE 22

docker/Dockerfile.python3.12

@@ -0,0 +1,48 @@
+FROM nvidia/cuda:12.4.1-devel-ubuntu20.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+WORKDIR /FastVideo
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    wget \
+    git \
+    ca-certificates \
+    openssh-server \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
+    bash Miniconda3-latest-Linux-x86_64.sh -b -p /opt/conda && \
+    rm Miniconda3-latest-Linux-x86_64.sh
+
+ENV PATH=/opt/conda/bin:$PATH
+
+RUN conda create --name fastvideo-dev python=3.12.9 -y
+
+SHELL ["/bin/bash", "-c"]
+
+# Copy just the pyproject.toml first to leverage Docker cache
+COPY pyproject.toml ./
+
+# Create a dummy README to satisfy the installation
+RUN echo "# Placeholder" > README.md
+
+RUN conda run -n fastvideo-dev pip install --no-cache-dir --upgrade pip && \
+    conda run -n fastvideo-dev pip install --no-cache-dir .[dev] && \
+    conda run -n fastvideo-dev pip install --no-cache-dir flash-attn==2.7.4.post1 --no-build-isolation && \
+    conda clean -afy
+
+COPY . .
+
+RUN conda run -n fastvideo-dev pip install --no-cache-dir -e .[dev]
+
+# Remove authentication headers
+RUN git config --unset-all http.https://github.com/.extraheader || true
+
+# Set up automatic conda environment activation for all shells
+RUN echo 'source /opt/conda/etc/profile.d/conda.sh' >> /root/.bashrc && \
+    echo 'conda activate fastvideo-dev' >> /root/.bashrc && \
+    # Ensure .bashrc is sourced for SSH login shells
+    echo 'if [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile
+
+EXPOSE 22

docs/requirements-docs.txt

@@ -1,25 +1,15 @@
-sphinx==6.2.1
-sphinx-argparse==0.4.0
-sphinx-book-theme==1.0.1
+sphinx==7.4.7
+sphinx-argparse==0.5.2
+sphinx-autodoc2==0.5.0
+sphinx-book-theme==1.1.4
 sphinx-copybutton==0.5.2
 sphinx-design==0.6.1
 sphinx-togglebutton==0.3.2
 myst-parser==3.0.1
 msgspec
-cloudpickle
+commonmark # Required by sphinx-argparse when using :markdownhelp:
 
 # packages to install to build the documentation
 cachetools
-pydantic >= 2.8
 -f https://download.pytorch.org/whl/cpu
-torch
-py-cpuinfo
-transformers
-mistral_common >= 1.5.4
-aiohttp
-starlette
-openai # Required by docs/source/serving/openai_compatible_server.md's vllm.entrypoints.openai.cli_args
-fastapi # Required by docs/source/serving/openai_compatible_server.md's vllm.entrypoints.openai.cli_args
-partial-json-parser # Required by docs/source/serving/openai_compatible_server.md's vllm.entrypoints.openai.cli_args
-requests
-zmq
+torch

docs/source/api/summary.md

@@ -0,0 +1,19 @@
+# Summary
+
+## Video Generator
+
+```{autodoc2-summary}
+    fastvideo.VideoGenerator
+```
+
+## Initialization Configuration
+
+```{autodoc2-summary}
+    fastvideo.v1.configs.pipelines.PipelineConfig
+```
+
+## Sampling Configuration
+
+```{autodoc2-summary}
+    fastvideo.v1.configs.sample.SamplingParam
+```

docs/source/autodoc2_docstring_parser.py

@@ -0,0 +1,22 @@
+# type: ignore
+# SPDX-License-Identifier: Apache-2.0
+from docutils import nodes
+from myst_parser.parsers.sphinx_ import MystParser
+from sphinx.ext.napoleon import docstring
+
+
+class NapoleonParser(MystParser):
+
+    def parse(self, input_string: str, document: nodes.document) -> None:
+        # Get the Sphinx configuration
+        config = document.settings.env.config
+
+        parsed_content = str(
+            docstring.GoogleDocstring(
+                str(docstring.NumpyDocstring(input_string, config)),
+                config,
+            ))
+        return super().parse(parsed_content, document)
+
+
+Parser = NapoleonParser

docs/source/conf.py

@@ -13,17 +13,19 @@
 # documentation root, use os.path.abspath to make it absolute, like shown here.
 
 import datetime
-import inspect
 import logging
 import os
+import re
 import sys
+from pathlib import Path
 from typing import Optional
 
 import requests
-from sphinx.ext import autodoc
 
 logger = logging.getLogger(__name__)
-sys.path.append(os.path.abspath("../.."))
+REPO_ROOT = Path(__file__).resolve().parent.parent.parent
+print(os.path.abspath(REPO_ROOT))
+sys.path.append(os.path.abspath(REPO_ROOT))
 
 # -- Project information -----------------------------------------------------
 
@@ -41,8 +43,7 @@ extensions = [
     "sphinx.ext.linkcode",
     "sphinx.ext.intersphinx",
     "sphinx_copybutton",
-    "sphinx.ext.autodoc",
-    "sphinx.ext.autosummary",
+    "autodoc2",
     "myst_parser",
     "sphinxarg.ext",
     "sphinx_design",
@@ -50,6 +51,31 @@ extensions = [
 ]
 myst_enable_extensions = [
     "colon_fence",
+    "fieldlist",
+]
+autodoc2_packages = [
+    {
+        "path": "../../fastvideo",
+        "exclude_dirs": ["__pycache__", "third_party"],
+    },
+]
+autodoc2_output_dir = "api"
+autodoc2_render_plugin = "myst"
+autodoc2_hidden_objects = ["dunder", "private", "inherited"]
+autodoc2_docstring_parser_regexes = [
+    (".*", "docs.source.autodoc2_docstring_parser"),
+]
+autodoc2_sort_names = True
+autodoc2_index_template = None
+autodoc2_skip_module_regexes = [
+    "fastvideo.dataset",
+    "fastvideo.distill",
+    "fastvideo.data_preprocess",
+    "fastvideo.models",
+    "fastvideo.sample",
+    "fastvideo.utils",
+    "fastvideo.distill_adv",
+    "fastvideo.train",
 ]
 
 # Add any paths that contain templates here, relative to this directory.
@@ -78,6 +104,11 @@ html_theme_options = {
     'repository_url': 'https://github.com/hao-ai-lab/FastVideo/',
     'use_repository_button': True,
     'use_edit_page_button': True,
+    # Prevents the full API being added to the left sidebar of every page.
+    # Reduces build time by 2.5x and reduces build size from ~225MB to ~95MB.
+    'collapse_navbar': True,
+    # Makes API visible in the right sidebar on API reference pages.
+    'show_toc_level': 3,
 }
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
@@ -160,38 +191,38 @@ def linkcode_resolve(domain, info):
         return None
     if not info['module']:
         return None
-    module = info['module']
 
-    # try to determine the correct file and line number to link to
-    obj = sys.modules[module]
+    # Get path from module name
+    file = Path(f"{info['module'].replace('.', '/')}.py")
+    path = REPO_ROOT / file
+    if not path.exists():
+        path = REPO_ROOT / file.with_suffix("") / "__init__.py"
+    if not path.exists():
+        return None
 
-    # get as specific as we can
-    lineno: int = 0
-    filename: str = ""
-    try:
-        for part in info['fullname'].split('.'):
-            obj = getattr(obj, part)
+    # Get the line number of the object
+    with open(path) as f:
+        lines = f.readlines()
+    name = info['fullname'].split(".")[-1]
+    pattern = fr"^( {{4}})*((def|class) )?{name}\b.*"
+    for lineno, line in enumerate(lines, 1):
+        if not line or line.startswith("#"):
+            continue
+        if re.match(pattern, line):
+            break
 
-            if not (inspect.isclass(obj) or inspect.isfunction(obj)
-                    or inspect.ismethod(obj)):
-                obj = obj.__class__  # type: ignore[assignment]
+    # If the line number is not found, return None
+    if lineno == len(lines):
+        return None
 
-            lineno = inspect.getsourcelines(obj)[1]
-            filename = (inspect.getsourcefile(obj)
-                        or f"{filename}.py").split("FastVideo/", 1)[1]
-    except Exception:
-        # For some things, like a class member, won't work, so
-        # we'll use the line number of the parent (the class)
-        pass
-
-    if filename.startswith("checkouts/"):
+    # If the line number is found, create the URL
+    filename = path.relative_to(REPO_ROOT)
+    if "checkouts" in path.parts:
         # a PR build on readthedocs
-        pr_number = filename.split("/")[1]
-        filename = filename.split("/", 2)[2]
+        pr_number = REPO_ROOT.name
         base, branch = get_repo_base_and_branch(pr_number)
         if base and branch:
             return f"https://github.com/{base}/blob/{branch}/{filename}#L{lineno}"
-
     # Otherwise, link to the source file on the main branch
     return f"https://github.com/hao-ai-lab/FastVideo/blob/main/{filename}#L{lineno}"
 
@@ -203,6 +234,8 @@ autodoc_mock_imports = [
     "cpuinfo",
     "cv2",
     "torch",
+    "huggingface_hub",
+    "torchvision",
     "transformers",
     "psutil",
     "prometheus_client",
@@ -231,18 +264,6 @@ for mock_target in autodoc_mock_imports:
             "been loaded into sys.modules when the sphinx build starts.",
             mock_target)
 
-
-class MockedClassDocumenter(autodoc.ClassDocumenter):
-    """Remove note about base class when a class is derived from object."""
-
-    def add_line(self, line: str, source: str, *lineno: int) -> None:
-        if line == "   Bases: :py:class:`object`":
-            return
-        super().add_line(line, source, *lineno)
-
-
-autodoc.ClassDocumenter = MockedClassDocumenter
-
 intersphinx_mapping = {
     "python": ("https://docs.python.org/3", None),
     "typing_extensions":
@@ -254,7 +275,4 @@ intersphinx_mapping = {
     "psutil": ("https://psutil.readthedocs.io/en/stable", None),
 }
 
-autodoc_preserve_defaults = True
-autodoc_warningiserror = True
-
 navigation_with_keys = False

docs/source/contributing/developer_env/docker.md

@@ -1,3 +1,4 @@
+(docker)=
 # 🐳 Using the FastVideo Docker Image
 
 If you prefer a containerized development environment or want to avoid managing dependencies manually, you can use our prebuilt Docker image:

docs/source/contributing/overview.md

@@ -39,7 +39,7 @@ Now you can install FastVideo and setup git hooks for running linting. By using
 pip install -e .[dev]
 
 # Can also install flash-attn (optional)
-pip install flash-attn==2.7.0.post2 --no-build-isolation 
+pip install flash-attn==2.7.4.post1 --no-build-isolation 
 
 # Linting, formatting and static type checking
 pre-commit install --hook-type pre-commit --hook-type commit-msg

docs/source/generate_examples.py

@@ -9,7 +9,7 @@ from typing import Optional
 
 ROOT_DIR = Path(__file__).parent.parent.parent.resolve()
 ROOT_DIR_RELATIVE = '../../../..'
-EXAMPLE_DIR = ROOT_DIR / "fastvideo/v1/examples"
+EXAMPLE_DIR = ROOT_DIR / "examples"
 EXAMPLE_DOC_DIR = ROOT_DIR / "docs/source/getting_started/examples"
 
 

docs/source/getting_started/installation.md

@@ -4,32 +4,19 @@
 
 FastVideo currently only supports Linux and NVIDIA CUDA GPUs.
 
-FastVideo has been tested on the following GPUs, but it should work on any GPUs that supports CUDA 12.4+, please create an issue if you discover any issues:
-- RTX 4090
-- A40
-- L40S
-- A100
-- H100
-
 ## Requirements
 
-- OS: Linux
-- Python: 3.10-3.12
-- CUDA 12.4+
+- **OS: Linux**
+- **Python: 3.10-3.12**
+- **CUDA 12.4**
+- **At least 1 NVIDIA GPU**
 
-## Installation Options
+## Set up using Python
+### Create a new Python environment
 
-### Option 1: Quick Install
-
-```bash
-pip install fastvideo
-```
-
-### Option 2: Installation from Source
-
-We recommend using a Python environment such as Conda.
-
-#### 1. [Optional] Install Miniconda (if not already installed)
+#### Conda
+You can create a new python environment using [Conda](https://docs.conda.io/projects/conda/en/stable/user-guide/getting-started.html)
+##### 1. Install Miniconda (if not already installed)
 
 ```bash
 wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
@@ -37,38 +24,77 @@ bash Miniconda3-latest-Linux-x86_64.sh
 source ~/.bashrc
 ```
 
-#### 2. [Optional] Create and activate a Conda environment for FastVideo
+##### 2. Create and activate a Conda environment for FastVideo
 
 ```bash
-conda create -n fastvideo python=3.10 -y
+# (Recommended) Create a new conda environment.
+conda create -n fastvideo python=3.12 -y
 conda activate fastvideo
 ```
 
-#### 3. Clone the FastVideo repository
+:::{note}
+[PyTorch has deprecated the conda release channel](https://github.com/pytorch/pytorch/issues/138506). If you use `conda`, please only use it to create Python environment rather than installing packages.
+:::
+
+#### uv
+
+:::{tip}
+We highly recommend using `uv` to install FastVideo. In our experience, `uv` speeds up installation by at least 3x.
+:::
+
+Or you can create a new Python environment using [uv](https://docs.astral.sh/uv/), a very fast Python environment manager. Please follow the [documentation](https://docs.astral.sh/uv/#getting-started) to install `uv`. After installing `uv`, you can create a new Python environment using the following command:
+
+```console
+# (Recommended) Create a new uv environment. Use `--seed` to install `pip` and `setuptools` in the environment.
+uv venv --python 3.12 --seed
+source .venv/bin/activate
+```
+
+### Installation
+
+```bash
+pip install fastvideo
+
+# or if you are using uv
+uv pip install fastvideo
+```
+
+Also optionally install flash-attn:
+
+```bash
+pip install flash-attn==2.7.4.post1 --no-build-isolation
+```
+
+### Installation from Source
+
+#### 1. Clone the FastVideo repository
 
 ```bash
 git clone https://github.com/hao-ai-lab/FastVideo.git && cd FastVideo
 ```
 
-#### 4. Install FastVideo
+#### 2. Install FastVideo
 
 Basic installation:
 
 ```bash
 pip install -e .
+
+# or if you are using uv
+uv pip install -e .
 ```
 
-## Optional Dependencies
+### Optional Dependencies
 
-### Flash Attention
+#### Flash Attention
 
 ```bash
-pip install flash-attn==2.7.0.post2 --no-build-isolation
+pip install flash-attn==2.7.4.post1 --no-build-isolation
 ```
 
-### Sliding Tile Attention (STA) (Requires CUDA 12.4+ and H100)
-
-To try Sliding Tile Attention (optional), please follow the instructions in [csrc/sliding_tile_attention/README.md](#sta-installation) to install STA.
+## Set up using Docker
+We also have prebuilt docker images with FastVideo dependencies pre-installed:
+[Docker Images](#docker)
 
 ## Development Environment Setup
 
@@ -78,8 +104,7 @@ If you're planning to contribute to FastVideo please see the following page:
 ## Hardware Requirements
 
 ### For Basic Inference
-- NVIDIA GPU with CUDA support
-- Minimum 20GB VRAM for quantized models (e.g., single RTX 4090)
+- NVIDIA GPU with CUDA 12.4 support
 
 ### For Lora Finetuning
 - 40GB GPU memory each for 2 GPUs with lora

docs/source/getting_started/v1_api.md

@@ -0,0 +1,83 @@
+# V1 API
+
+FastVideo's V1 API provides a streamlined interface for video generation tasks with powerful customization options. This page documents the primary components of the API.
+
+## Video Generator
+
+This class will be the primary Python API for generating videos and images.
+
+```{autodoc2-summary}
+    fastvideo.VideoGenerator
+```
+
+`````{py:class} VideoGenerator(fastvideo_args: fastvideo.v1.fastvideo_args.FastVideoArgs, executor_class: type[fastvideo.v1.worker.executor.Executor], log_stats: bool)
+:canonical: fastvideo.v1.entrypoints.video_generator.VideoGenerator
+
+```{autodoc2-docstring} fastvideo.v1.entrypoints.video_generator.VideoGenerator
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+`VideoGenerator.from_pretrained()` should be the primary way of creating a new video generator.
+
+````{py:method} from_pretrained(model_path: str, device: typing.Optional[str] = None, torch_dtype: typing.Optional[torch.dtype] = None, pipeline_config: typing.Optional[typing.Union[str | fastvideo.v1.configs.pipelines.PipelineConfig]] = None, **kwargs) -> fastvideo.v1.entrypoints.video_generator.VideoGenerator
+:canonical: fastvideo.v1.entrypoints.video_generator.VideoGenerator.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.v1.entrypoints.video_generator.VideoGenerator.from_pretrained
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+
+## Configuring FastVideo
+
+The follow two classes `PipelineConfig` and `SamplingParam` are used to configure initialization and sampling parameters, respectively.
+
+### PipelineConfig
+```{autodoc2-summary}
+    fastvideo.PipelineConfig
+```
+
+`````{py:class} PipelineConfig
+:canonical: fastvideo.v1.configs.pipelines.base.PipelineConfig
+
+```{autodoc2-docstring} fastvideo.v1.configs.pipelines.base.PipelineConfig
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+````{py:method} from_pretrained(model_path: str) -> fastvideo.v1.configs.pipelines.base.PipelineConfig
+:canonical: fastvideo.v1.configs.pipelines.base.PipelineConfig.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.v1.configs.pipelines.base.PipelineConfig.from_pretrained
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+
+````{py:method} dump_to_json(file_path: str)
+:canonical: fastvideo.v1.configs.pipelines.base.PipelineConfig.dump_to_json
+
+```{autodoc2-docstring} fastvideo.v1.configs.pipelines.base.PipelineConfig.dump_to_json
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+
+### SamplingParam
+
+```{autodoc2-summary}
+    fastvideo.SamplingParam
+```
+
+`````{py:class} SamplingParam
+:canonical: fastvideo.v1.configs.sample.base.SamplingParam
+
+```{autodoc2-docstring} fastvideo.v1.configs.sample.base.SamplingParam
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+````{py:method} from_pretrained(model_path: str) -> fastvideo.v1.configs.sample.base.SamplingParam
+:canonical: fastvideo.v1.configs.sample.base.SamplingParam.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.v1.configs.sample.base.SamplingParam.from_pretrained
+:parser: docs.source.autodoc2_docstring_parser
+```

docs/source/index.md

@@ -9,7 +9,7 @@
 
 :::{raw} html
 <p style="text-align:center">
-<strong>FastVideo is a lightweight framework for accelerating large video diffusion models.
+<strong>FastVideo is a unified framework for accelerated video generation.
 </strong>
 </p>
 
@@ -21,27 +21,31 @@
 </p>
 :::
 
-FastVideo is a lightweight framework for accelerating large video diffusion models developed by the [Hao AI Lab](https://hao-ai-lab.github.io/).
+It features a clean, consistent API that works across popular video models, making it easier for developers to author new models and incorporate system- or kernel-level optimizations.
+With FastVideo's optimizations, you can achieve more than 3x inference improvement compared to other systems.
 
 <div style="text-align: center;">
-  <video controls width="800">
-    <source src="https://github.com/user-attachments/assets/79af5fb8-707c-4263-b153-9ab2a01d3ac1" type="video/mp4">
-    Your browser does not support the video tag.
-  </video>
+  <img src=_static/images/perf.png width="100%"/>
 </div>
 
-FastVideo currently offers: (with more to come)
+## Key Features
 
-- [NEW!] V1 inference API available. Full announcement coming soon!
-- [Sliding Tile Attention](https://hao-ai-lab.github.io/blogs/sta/).
-- FastHunyuan and FastMochi: consistency distilled video diffusion models for 8x inference speedup.
-- First open distillation recipes for video DiT, based on [PCM](https://github.com/G-U-N/Phased-Consistency-Model).
-- Support distilling/finetuning/inferencing state-of-the-art open video DiTs: 1. Mochi 2. Hunyuan.
+FastVideo has the following features:
+- State-of-the-art performance optimizations for inference
+  - [Sliding Tile Attention](https://arxiv.org/pdf/2502.04507)
+  - [TeaCache](https://arxiv.org/pdf/2411.19108)
+  - [Sage Attention](https://arxiv.org/abs/2410.02367)
+- Cutting edge models
+  - Wan2.1 T2V, I2V
+  - HunyuanVideo
+  - FastHunyuan: consistency distilled video diffusion models for 8x inference speedup.
+  - StepVideo T2V
+- Distillation support
+  - Recipes for video DiT, based on [PCM](https://github.com/G-U-N/Phased-Consistency-Model).
+  - Support distilling/finetuning/inferencing state-of-the-art open video DiTs: 1. Mochi 2. Hunyuan.
 - Scalable training with FSDP, sequence parallelism, and selective activation checkpointing, with near linear scaling to 64 GPUs.
 - Memory efficient finetuning with LoRA, precomputed latent, and precomputed text embeddings.
 
-Dev in progress and highly experimental.
-
 ## Documentation
 
 % How to start using FastVideo?
@@ -51,14 +55,20 @@ Dev in progress and highly experimental.
 :maxdepth: 1
 
 getting_started/installation
-<!-- getting_started/examples/examples_index -->
+<!-- getting_started/v1_api -->
 :::
 
 :::{toctree}
 :caption: Inference
 :maxdepth: 1
 
+inference/inference_quick_start
+inference/configuration
+inference/optimizations
+inference/support_matrix
 inference/examples/examples_inference_index
+inference/cli
+inference/add_pipeline
 inference/v0_inference
 :::
 
@@ -93,7 +103,14 @@ design/overview
 
 contributing/overview
 contributing/developer_env/index
-contributing/add_pipeline
+:::
+
+:::{toctree}
+:caption: API Reference
+:maxdepth: 2
+
+<!-- api/summary -->
+api/fastvideo/fastvideo
 :::
 
 ## Indices and tables

docs/source/inference/add_pipeline.md

@@ -1,6 +1,6 @@
 (add-pipeline)=
 
-# 🏗️ Adding a New Diffusion Pipeline
+# 🏗️ Adding a New Pipeline
 
 This guide explains how to implement a custom diffusion pipeline in FastVideo, leveraging the framework's modular architecture for high-performance video generation.
 

docs/source/inference/cli.md

@@ -0,0 +1,151 @@
+# FastVideo CLI Inference
+
+The FastVideo CLI provides a quick way to access the FastVideo inference pipeline for video generation. For more advanced usage,
+see the Python interface [here](https://hao-ai-lab.github.io/FastVideo/inference/examples/basic.html).
+
+## Basic Usage
+
+The basic command to generate a video is:
+
+```bash
+fastvideo generate --model-path {MODEL_PATH} --prompt {PROMPT}
+```
+
+### Required Parameters
+
+- `--model-path {MODEL_PATH}`: Path to the model or model ID
+- `--prompt {PROMPT}`: Text description for the video you want to generate
+
+## Common Arguments
+
+To see all the options, you can use the `--help` flag:
+
+```bash
+fastvideo generate --help
+```
+
+### Hardware Configuration
+
+- `--num-gpus {NUM_GPUS}`: Number of GPUs to use
+- `--tp-size {TP_SIZE}`: Tensor parallelism size (Typically should match the number of GPUs)
+- `--sp-size {SP_SIZE}`: Sequence parallelism size (Typically should match the number of GPUs)
+
+#### Video Configuration
+
+- `--height {HEIGHT}`: Height of the generated video
+- `--width {WIDTH}`: Width of the generated video
+- `--num-frames {NUM_FRAMES}`: Number of frames to generate
+- `--fps {FPS}`: Frames per second for the saved video
+
+#### Generation Parameters
+
+- `--num-inference-steps {STEPS}`: Number of denoising steps
+- `--negative-prompt {PROMPT}`: Negative prompt to guide generation away from certain concepts
+- `--seed {SEED}`: Random seed for reproducible generation
+
+#### Output Options
+
+- `--output-path {PATH}`: Directory to save the generated video
+- `--save-video`: Whether to save the video to disk
+- `--return-frames`: Whether to return the raw frames
+
+## Using Configuration Files
+
+Instead of specifying all parameters on the command line, you can use a configuration file:
+
+```bash
+fastvideo generate --config {CONFIG_FILE_PATH}
+```
+
+The config file should be in JSON or YAML format with the same parameter names as the CLI options. Command-line arguments will take precedence over settings in the configuration file, allowing you to override specific values while keeping the rest from the config file.
+
+Example configuration file (config.json):
+
+```json
+{
+    "model_path": "FastVideo/FastHunyuan-diffusers",
+    "prompt": "A beautiful woman in a red dress walking down a street",
+    "output_path": "outputs/",
+    "num_gpus": 2,
+    "sp_size": 2,
+    "tp_size": 2,
+    "num_frames": 45,
+    "height": 720,
+    "width": 1280,
+    "num_inference_steps": 6,
+    "seed": 1024,
+    "fps": 24,
+    "precision": "bf16",
+    "vae_precision": "fp16",
+    "vae_tiling": true,
+    "vae_sp": true,
+    "vae_config": {
+        "load_encoder": false,
+        "load_decoder": true,
+        "tile_sample_min_height": 256,
+        "tile_sample_min_width": 256
+    },
+    "text_encoder_precisions": [
+        "fp16",
+        "fp16"
+    ],
+    "mask_strategy_file_path": null,
+    "enable_torch_compile": false
+}
+```
+
+Or using YAML format (config.yaml):
+
+```yaml
+model_path: "FastVideo/FastHunyuan-diffusers"
+prompt: "A beautiful woman in a red dress walking down a street"
+output_path: "outputs/"
+num_gpus: 2
+sp_size: 2
+tp_size: 2
+num_frames: 45
+height: 720
+width: 1280
+num_inference_steps: 6
+seed: 1024
+fps: 24
+precision: "bf16"
+vae_precision: "fp16"
+vae_tiling: true
+vae_sp: true
+vae_config:
+  load_encoder: false
+  load_decoder: true
+  tile_sample_min_height: 256
+  tile_sample_min_width: 256
+text_encoder_precisions:
+  - "fp16"
+  - "fp16"
+mask_strategy_file_path: null
+enable_torch_compile: false
+```
+
+## Examples
+
+Generating a simple video:
+
+```bash
+fastvideo generate --model-path FastVideo/FastHunyuan-diffusers --prompt "A cat playing with a ball of yarn" --num-frames 45 --height 720 --width 1280 --num-inference-steps 6 --seed 1024 --output-path outputs/
+```
+
+Using a negative prompt to avoid certain elements:
+
+```bash
+fastvideo generate --model-path FastVideo/FastHunyuan-diffusers --prompt "A beautiful forest landscape" --negative-prompt "people, buildings, roads"
+```
+
+Combining command line arguments and a configuration file:
+
+```bash
+fastvideo generate --config config.json --prompt "A capybara lounging in a hammock"
+```
+
+## Troubleshooting
+
+- If you encounter CUDA out-of-memory errors, try reducing the video dimensions or number of frames, or the number of inference steps.
+- For reproducible results, set the same seed value between runs.

docs/source/inference/configuration.md

@@ -0,0 +1,77 @@
+(inference-configuration)=
+# Configuration
+
+## Multi-GPU Setup
+
+FastVideo automatically distributes the generation process when multiple GPUs are specified:
+
+```python
+# Will use 4 GPUs in parallel for faster generation
+generator = VideoGenerator.from_pretrained(
+    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+    num_gpus=4,
+)
+```
+
+## Customizing Generation
+
+- `PipelineConfig`: Initialization time parameters
+- `SamplingParam`: Generation time parameters
+
+You can customize various parameters when generating videos using the `PipelineConfig` and `SamplingParam` class:
+
+```python
+from fastvideo import VideoGenerator, SamplingParam, PipelineConfig
+
+def main():
+    model_name = "Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
+    config = PipelineConfig.from_pretrained(model_name)
+    config.vae_precision = "fp16"
+    config.use_cpu_offload = True
+
+    # Create the generator
+    generator = VideoGenerator.from_pretrained(
+        model_name,
+        num_gpus=1,
+        pipeline_config=config
+    )
+
+    # Create and customize sampling parameters
+    sampling_param = SamplingParam.from_pretrained("Wan-AI/Wan2.1-T2V-1.3B-Diffusers")
+    
+    # How many frames to generate
+    sampling_param.num_frames = 45
+    
+    # Video resolution (width, height)
+    sampling_param.width = 1024
+    sampling_param.height = 576
+    
+    # How many steps we denoise the video (higher = better quality, slower generation)
+    sampling_param.num_inference_steps = 30
+    
+    # How strongly the video conforms to the prompt (higher = more faithful to prompt)
+    sampling_param.guidance_scale = 7.5
+    
+    # Random seed for reproducibility
+    sampling_param.seed = 42  # Optional, leave unset for random results
+
+    # Generate video with custom parameters
+    prompt = "A beautiful sunset over a calm ocean, with gentle waves."
+    video = generator.generate_video(
+        prompt, 
+        sampling_param=sampling_param, 
+        output_path="my_videos/",  # Controls where videos are saved
+        return_frames=True,  # Also return frames from this call (defaults to False)
+        save_video=True
+    )
+
+    # If return_frames=True, video contains the generated frames as a NumPy array
+    print(f"Generated {len(video)} frames")
+
+if __name__ == '__main__':
+    main()
+```
+
+## Performance Optimization
+
+For configuring optimizations, please see our [optimizations guide](#inference-optimizations)

docs/source/inference/fasthunyuan.md

@@ -1,33 +0,0 @@
-(fasthunyuan)=
-
-# FastHunyuan
-## Inference FastHunyuan on single RTX4090
-We now support NF4 and LLM-INT8 quantized inference using BitsAndBytes for FastHunyuan. With NF4 quantization, inference can be performed on a single RTX 4090 GPU, requiring just 20GB of VRAM.
-
-```bash
-# Download the model weight
-python scripts/huggingface/download_hf.py --repo_id=FastVideo/FastHunyuan-diffusers --local_dir=data/FastHunyuan-diffusers --repo_type=model
-# CLI inference
-bash scripts/inference/inference_hunyuan_hf_quantization.sh
-```
-
-For more information about the VRAM requirements for BitsAndBytes quantization, please refer to the table below (timing measured on an H100 GPU):
-
-| Configuration                  | Memory to Init Transformer | Peak Memory After Init Pipeline (Denoise) | Diffusion Time | End-to-End Time |
-|--------------------------------|----------------------------|--------------------------------------------|----------------|-----------------|
-| BF16 + Pipeline CPU Offload    | 23.883G                   | 33.744G                                    | 81s            | 121.5s          |
-| INT8 + Pipeline CPU Offload    | 13.911G                   | 27.979G                                    | 88s            | 116.7s          |
-| NF4 + Pipeline CPU Offload     | 9.453G                    | 19.26G                                     | 78s            | 114.5s          |
-  
-For improved quality in generated videos, we recommend using a GPU with 80GB of memory to run the BF16 model with the original Hunyuan pipeline. To execute the inference, use the following section:
-
-## FastHunyuan
-
-```bash
-# Download the model weight
-python scripts/huggingface/download_hf.py --repo_id=FastVideo/FastHunyuan --local_dir=data/FastHunyuan --repo_type=model
-# CLI inference
-bash scripts/inference/inference_hunyuan.sh
-```
-
-You can also inference FastHunyuan in the [official Hunyuan github](https://github.com/Tencent/HunyuanVideo).

docs/source/inference/fastmochi.md

@@ -1,9 +0,0 @@
-(fastmochi)=
-
-# FastMochi
-
-```bash
-# Download the model weight
-python scripts/huggingface/download_hf.py --repo_id=FastVideo/FastMochi-diffusers --local_dir=data/FastMochi-diffusers --repo_type=model
-# CLI inference
-bash scripts/inference/inference_mochi_sp.sh

docs/source/inference/hunyuanvideo.md

@@ -1,18 +0,0 @@
-(hunyuanvideo)=
-
-# HunyuanVideo
-## Inference HunyuanVideo with Sliding Tile Attention
-First, download the model:
-
-```bash
-python scripts/huggingface/download_hf.py --repo_id=FastVideo/hunyuan --local_dir=data/hunyuan --repo_type=model 
-```
-
-We provide two examples in the following script to run inference with STA + [TeaCache](https://github.com/ali-vilab/TeaCache) and STA only.
-
-```bash
-sh scripts/inference/inference_hunyuan_STA.sh
-```
-
-## Video Demos using STA + Teacache
-Visit our [demo website](https://fast-video.github.io/) to explore our complete collection of examples. We shorten a single video generation process from 945s to 317s on H100.

docs/source/inference/inference_quick_start.md

@@ -0,0 +1,124 @@
+# Inference Quick Start
+
+This page contains step-by-step instructions to get you quickly started with video generation using FastVideo.
+
+## Requirements
+- **OS**: Linux (Tested on Ubuntu 22.04+)
+- **Python**: 3.10-3.12
+- **CUDA**: 12.4
+- **GPU**: At least one NVIDIA GPU
+
+## Installation
+
+We recommend using an environment manager such as `Conda` to create a clean environment:
+
+```bash
+# Create and activate a new conda environment
+conda create -n fastvideo python=3.12
+conda activate fastvideo
+
+# Install FastVideo
+pip install fastvideo
+```
+
+For advanced installation options, see the [Installation Guide](installation.md).
+
+## Generating Your First Video
+Here's a minimal example to generate a video using the default settings. Create a file called `example.py` with the following code:
+
+```python
+from fastvideo import VideoGenerator
+
+def main():
+    # Create a video generator with a pre-trained model
+    generator = VideoGenerator.from_pretrained(
+        "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        num_gpus=1,  # Adjust based on your hardware
+    )
+
+    # Define a prompt for your video
+    prompt = "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes wide with interest."
+
+    # Generate the video
+    video = generator.generate_video(
+        prompt,
+        return_frames=True,  # Also return frames from this call (defaults to False)
+        output_path="my_videos/",  # Controls where videos are saved
+        save_video=True
+    )
+
+if __name__ == '__main__':
+    main()
+```
+
+Run the script with:
+
+```bash
+python example.py
+```
+
+The generated video will be saved in the current directory under `my_videos/`.
+
+## Available Models
+
+Please see the [support matrix](#support-matrix) for the list of supported models and their available optimizations.
+
+## Image-to-Video Generation
+
+You can generate a video starting from an initial image:
+
+```python
+from fastvideo import VideoGenerator, SamplingParam
+
+def main():
+    # Create the generator
+    model_name = "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers"
+    generator = VideoGenerator.from_pretrained(model_name, num_gpus=1)
+
+    # Set up parameters with an initial image
+    sampling_param = SamplingParam.from_pretrained(model_name)
+    sampling_param.image_path = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
+    sampling_param.num_frames = 107
+    sampling_param.image_strength = 0.8  # How much to preserve the original image (0-1)
+
+    # Generate video based on the image
+    prompt = "A photograph coming to life with gentle movement"
+    generator.generate_video(prompt, sampling_param=sampling_param,
+                             output_path="my_videos/",
+                             save_video=True)
+
+if __name__ == '__main__':
+    main()
+```
+
+## Troubleshooting
+
+Common issues and their solutions:
+
+### Out of Memory Errors
+If you encounter CUDA out of memory errors:
+- Reduce `num_frames` or video resolution
+- Enable memory optimization with `enable_model_cpu_offload`
+- Try a smaller model or use distilled versions
+- Use `num_gpus` > 1 if multiple GPUs are available
+
+### Slow Generation
+To speed up generation:
+- Reduce `num_inference_steps` (20-30 is usually sufficient)
+- Use half precision (`fp16`) for the VAE
+- Use multiple GPUs if available
+
+### Unexpected Results
+If the generated video doesn't match your prompt:
+- Try increasing `guidance_scale` (7.0-9.0 works well)
+- Make your prompt more detailed and specific
+- Experiment with different random seeds
+- Try a different model
+
+## Next Steps
+
+- Learn about [Advanced Inference Configurations](#inference-configuration)
+- Learn about using [Optimizations](#inference-optimizations)
+- See [Examples](../examples/examples_inference_index.md) for more usage scenarios
+- Join our [Community Discord](https://discord.gg/JA7cksDz86).
+- Join our [Community Slack](https://join.slack.com/t/fastvideo/shared_invite/zt-2zf6ru791-sRwI9lPIUJQq1mIeB_yjJg).

docs/source/inference/optimizations.md

@@ -0,0 +1,148 @@
+(inference-optimizations)=
+# Optimizations
+
+This page describes the various options for speeding up generation times in FastVideo.
+
+## Table of Contents
+- Optimized Attention Backends
+  - [Flash Attention](#optimizations-flash)
+  - [Sliding Tile Attention](#optimizations-sta)
+  - [Sage Attention](#optimizations-sage)
+
+- Caching Techniques
+  - [TeaCache](#optimizations-teacache)
+
+(optimizations-backends)=
+## Attention Backends
+
+### Available Backends
+- Torch SDPA: `FASTVIDEO_ATTENTION_BACKEND=TORCH_SDPA`
+- Flash Attention 2 and 3: `FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN`
+- Sliding Tile Attention: `FASTVIDEO_ATTENTION_BACKEND=SLIDING_TILE_ATTN`
+- Sage Attention: `FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN`
+
+### Configuring Backends
+
+There are two ways to configure the attention backend in FastVideo.
+
+#### 1. In Python
+In python, set the `FASTVIDEO_ATTENTION_BACKEND` environment variable before instantiating `VideoGenerator` like this:
+
+```python
+os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "SLIDING_TILE_ATTN"
+```
+
+#### 2. In CLI
+You can also set the environment variable on the command line:
+
+```bash
+FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN python example.py
+```
+
+(optimizations-flash)=
+### Flash Attention
+
+**`FLASH_ATTN`**
+
+We recommend always installing [Flash Attention 2](https://github.com/Dao-AILab/flash-attention):
+
+```bash
+pip install flash-attn==2.7.4.post1 --no-build-isolation
+```
+
+And if using a Hopper+ GPU (ie H100), installing [Flash Attention 3](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#flashattention-3-beta-release) by compiling it from source (takes about 10 minutes for me):
+
+```bash
+git clone https://github.com/Dao-AILab/flash-attention.git && cd flash-attention
+
+cd hopper
+pip install ninja 
+python setup.py install
+```
+
+:::{note}
+FastVideo will automatically detect and use `FA3` if it is installed when using `FLASH_ATTN` backend.
+:::
+
+(optimizations-sta)=
+### Sliding Tile Attention
+**`SLIDING_TILE_ATTN`**
+
+```bash
+pip install st_attn==0.0.4
+```
+
+Please see [this page](#sta-installation) for more installation instructions.
+
+(optimizations-sage)=
+### Sage Attention
+**`SAGE_ATTN`**
+
+To use [SageAttention](https://github.com/thu-ml/SageAttention) 2.1.1, please compile from source:
+
+```bash
+git clone https://github.com/thu-ml/SageAttention.git
+cd sageattention 
+python setup.py install  # or pip install -e .
+```
+
+(optimizations-teacache)=
+## Teacache
+TeaCache is an optimization technique supported in FastVideo that can significantly speed up video generation by skipping redundant calculations across diffusion steps. This guide explains how to enable and configure TeaCache for optimal performance in FastVideo.
+
+### What is TeaCache?
+
+See the official [TeaCache](https://github.com/ali-vilab/TeaCache) repo and their paper for more details.
+
+### How to Enable TeaCache
+
+Enabling TeaCache is straightforward - simply add the `enable_teacache=True` parameter to your `generate_video()` call:
+
+```python
+# ... previous code
+generator.generate_video(
+    prompt="Your prompt here",
+    sampling_param=params,
+    enable_teacache=True
+)
+# more code ...
+```
+
+### Complete Example
+
+At the bottom is a complete example of using TeaCache for faster video generation. You can run it using the following command:
+
+```bash
+python examples/inference/optimizations/teacache_example.py
+```
+
+### Advanced Configuration
+
+While TeaCache works well with default settings, you can fine-tune its behavior by adjusting the threshold value:
+
+1. Lower threshold values (e.g., 0.1) will result in more skipped calculations and faster generation with slightly more potential for quality degradation
+2. Higher threshold values (e.g., 0.15-0.23) will skip fewer calculations but maintain quality closer to the original
+
+Note that the optimal threshold depends on your specific model and content.
+
+## Benchmarking different optimizations
+
+To benchmark the performance improvement, try generating the same video with and without TeaCache enabled and compare the generation times:
+
+```python
+# Without TeaCache
+start_time = time.perf_counter()
+generator.generate_video(prompt="Your prompt", enable_teacache=False)
+standard_time = time.perf_counter() - start_time
+
+# With TeaCache
+start_time = time.perf_counter()
+generator.generate_video(prompt="Your prompt", enable_teacache=True)
+teacache_time = time.perf_counter() - start_time
+
+print(f"Standard generation: {standard_time:.2f} seconds")
+print(f"TeaCache generation: {teacache_time:.2f} seconds")
+print(f"Speedup: {standard_time/teacache_time:.2f}x")
+```
+
+Note: If you want to benchmark different attention backends, you'll need to reinstantiate `VideoGenerator`.

docs/source/inference/stepvideo.md

@@ -1,16 +0,0 @@
-(stepvideo)=
-
-# StepVideo
-## Inference StepVideo with Sliding Tile Attention
-First, download the model:
-
-```
-python scripts/huggingface/download_hf.py --repo_id=stepfun-ai/stepvideo-t2v --local_dir=data/stepvideo-t2v --repo_type=model 
-```
-
-Use the following scripts to run inference for StepVideo. When using STA for inference, the generated videos will have dimensions of 204×768×768 (currently, this is the only supported shape).
-
-```bash
-sh scripts/inference/inference_stepvideo_STA.sh # Inference stepvideo with STA
-sh scripts/inference/inference_stepvideo.sh # Inference original stepvideo
-```

docs/source/inference/support_matrix.md

@@ -0,0 +1,92 @@
+(support-matrix)=
+# Compatibility Matrix
+The table below shows every supported model and optimizations supported for them.
+
+The symbols used have the following meanings:
+
+- ✅ = Full compatibility
+- ❌ = No compatibility
+
+## Models x Optimization
+The `HuggingFace Model ID` can be directly pass to `from_pretrained()` methods and FastVideo will use the optimal default parameters when initializing and generating videos.
+
+:::{raw} html
+<style>
+  /* Make smaller to try to improve readability  */
+  td {
+    font-size: 0.9rem;
+    text-align: center;
+  }
+
+  th {
+    text-align: center;
+    font-size: 0.9rem;
+  }
+</style>
+:::
+
+:::{list-table}
+:header-rows: 1
+:stub-columns: 3
+:widths: auto
+:class: vertical-table-header
+
+- * Model Name
+  * HuggingFace Model ID
+  * Resolutions
+  * TeaCache
+  * Sliding Tile Attn
+  * Sage Attn
+- * HunyuanVideo
+  * `hunyuanvideo-community/HunyuanVideo`
+  * 720px1280p<br>544px960p
+  * ❌
+  * ✅
+  * ✅
+- * FastHunyuan
+  * `FastVideo/FastHunyuan-diffusers`
+  * 720px1280p<br>544px960p
+  * ❌
+  * ✅
+  * ✅
+- * Wan T2V 1.3B
+  * `Wan-AI/Wan2.1-T2V-1.3B-Diffusers`
+  * 480P
+  * ✅
+  * ✅*
+  * ✅
+- * Wan T2V 14B
+  * `Wan-AI/Wan2.1-T2V-14B-Diffusers`
+  * 480P, 720P
+  * ✅
+  * ✅*
+  * ✅
+- * Wan I2V 480P
+  * `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers`
+  * 480P
+  * ✅
+  * ✅*
+  * ✅
+- * Wan I2V 720P
+  * `Wan-AI/Wan2.1-I2V-14B-720P-Diffusers`
+  * 720P
+  * ✅
+  * ✅*
+  * ✅
+- * StepVideo T2V
+  * `FastVideo/stepvideo-t2v-diffusers`
+  * 768px768px204f<br>544px992px204f<br>544px992px136f
+  * ❌
+  * ❌
+  * ✅
+:::
+
+**Note**: there are some known quality issues with Wan2.1 + Sliding Tile Attn. We are working on fixing this issue.
+
+## Special requirements
+
+### StepVideo T2V
+- The self-attention in text-encoder (step_llm) only supports CUDA capabilities sm_80 sm_86 and sm_90
+
+### Sliding Tile Attention
+- Currently only Hopper GPUs (H100s) are supported.

docs/source/inference/wanvideo.md

@@ -1,44 +0,0 @@
-(wanvideo)=
-
-# WanVideo
-## Inference T2V with WanVideo
-First, download the model:
-
-```bash
-python scripts/huggingface/download_hf.py --repo_id=Wan-AI/Wan2.1-T2V-1.3B-Diffusers --local_dir=YOUR_LOCAL_DIR --repo_type=model 
-```
-
-or
-
-```bash
-python scripts/huggingface/download_hf.py --repo_id=Wan-AI/Wan2.1-T2V-14B-Diffusers --local_dir=YOUR_LOCAL_DIR --repo_type=model 
-```
-
-Then run the inference using:
-
-```bash
-sh scripts/inference/v1_inference_wan.sh
-```
-
-Remember to set `MODEL_BASE` and `num_gpus` accordingly.
-
-## Inference I2V with WanVideo
-First, download the model:
-
-```bash
-python scripts/huggingface/download_hf.py --repo_id=Wan-AI/Wan2.1-I2V-14B-480P-Diffusers --local_dir=YOUR_LOCAL_DIR --repo_type=model
-```
-
-or
-
-```bash
-python scripts/huggingface/download_hf.py --repo_id=Wan-AI/Wan2.1-I2V-14B-720P-Diffusers --local_dir=YOUR_LOCAL_DIR --repo_type=model
-```
-
-Then run the inference using:
-
-```bash
-sh scripts/inference/v1_inference_wan_i2v.sh
-```
-
-Remember to set `MODEL_BASE` and `num_gpus` accordingly.

examples/inference/basic/README.md

@@ -1,3 +1,41 @@
-# Basic
+# Basic Video Generation Tutorial
+The `VideoGenerator` class provides the primary Python interface for doing offline video generation, which is interacting with a diffusion pipeline without using a separate inference api server.
 
-The class provides the main python interface for using FastVideo's inference pipeline.
+## Requirements
+- At least a single NVIDIA GPU with CUDA 12.4.
+- Python 3.10-3.12
+
+## Installation
+If you have not installed FastVideo, please following these [instructions](https://hao-ai-lab.github.io/FastVideo/getting_started/installation.html) first.
+
+## Usage
+The first script in this example shows the most basic usage of FastVideo. If you are new to Python and FastVideo, you should start here.
+
+```bash
+# if you have not cloned the directory:
+git clone https://github.com/hao-ai-lab/FastVideo.git && cd FastVideo
+
+python examples/inference/basic/basic.py
+```
+
+## Basic Walkthrough
+
+All you need to generate videos using multi-gpus from state-of-the-art diffusion pipelines is the following few lines!
+
+```python
+from fastvideo import VideoGenerator
+
+def main():
+    generator = VideoGenerator.from_pretrained(
+        "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        num_gpus=1,
+    )
+
+    prompt = ("A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
+             "wide with interest. The playful yet serene atmosphere is complemented by soft "
+             "natural light filtering through the petals. Mid-shot, warm and cheerful tones.")
+    video = generator.generate_video(prompt)
+
+if __name__ == "__main__":
+    main()
+```

examples/inference/basic/basic.py

@@ -1 +1,41 @@
-print('Hello, world!')
+from fastvideo import VideoGenerator
+
+# from fastvideo.v1.configs.sample import SamplingParam
+
+
+def main():
+    # FastVideo will automatically use the optimal default arguments for the
+    # model.
+    # If a local path is provided, FastVideo will make a best effort
+    # attempt to identify the optimal arguments.
+    generator = VideoGenerator.from_pretrained(
+        "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        # if num_gpus > 1, FastVideo will automatically handle distributed setup
+        num_gpus=1,
+    )
+
+    # sampling_param = SamplingParam.from_pretrained("Wan-AI/Wan2.1-T2V-1.3B-Diffusers")
+    # sampling_param.num_frames = 45
+    # sampling_param.image_path = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
+    # Generate videos with the same simple API, regardless of GPU count
+    prompt = (
+        "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
+        "wide with interest. The playful yet serene atmosphere is complemented by soft "
+        "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
+    )
+    video = generator.generate_video(prompt)
+    # video = generator.generate_video(prompt, sampling_param=sampling_param, output_path="wan_t2v_videos/")
+
+    # Generate another video with a different prompt, without reloading the
+    # model!
+    prompt2 = (
+        "A majestic lion strides across the golden savanna, its powerful frame "
+        "glistening under the warm afternoon sun. The tall grass ripples gently in "
+        "the breeze, enhancing the lion's commanding presence. The tone is vibrant, "
+        "embodying the raw energy of the wild. Low angle, steady tracking shot, "
+        "cinematic.")
+    video2 = generator.generate_video(prompt2)
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/default_args.py

@@ -0,0 +1,62 @@
+from fastvideo import VideoGenerator
+
+
+def main():
+
+    # This is the config class for the model initialization
+    config = PipelineConfig.from_pretrained("FastVideo/FastHunyuan-Diffusers")
+    # can be used to dump the config to a yaml file
+    config.dump_to_yaml("config.yaml")
+    print(config)
+    # {
+    #    'vae_config': {
+    #        'scale_factor': 8,
+    #        'sp': True,
+    #        'tiling': True,
+    #        'precision': 'fp16'
+    #    },
+    #    'text_encoder_config': {
+    #        'precision': 'fp16'
+    #    },
+    #    'dit_config': {
+    #        'precision': 'fp16'
+    #    },
+    #    'inference_args': {
+    #        'guidance_scale': 7.5,
+    #        'num_inference_steps': 5,
+    #        'seed': 1024,
+    #        'guidance_rescale': 0.0,
+    #        'flow_shift': 17,
+    #        'num_inference_steps': 5,
+    #    }
+    # }
+
+    config.vae_config.scale_factor = 16
+
+    # FastVideo will automatically used the optimal default arguments for the model
+    # If a local path is provided, FastVideo will make a best effort attempt to
+    # identify the optimal arguments.
+    generator = VideoGenerator.from_pretrained(
+        "FastVideo/FastHunyuan-Diffusers",
+        num_gpus=4,
+        config=config,
+        # or
+        config_path="config.yaml",
+    )
+
+    sampling_param = SamplingParam.from_pretrained(
+        "FastVideo/FastHunyuan-Diffusers")
+    sampling_param.num_inference_steps = 5
+
+    # Generate videos with the same simple API, regardless of GPU count
+    prompt = "A beautiful woman in a red dress walking down a street"
+    video = generator.generate_video(prompt,
+                                     sampling_param=sampling_param,
+                                     num_inference_steps=6)
+
+    video2 = generator.generate_video(prompt2)
+    prompt2 = "A beautiful woman in a blue dress walking down a street"
+
+
+if __name__ == "__main__":
+    main()

examples/inference/gradio/README.md

@@ -1,4 +1,4 @@
-# FastVideo VideoGenerator Gradio Demo
+# FastVideo Gradio Demo
 
 This is a Gradio-based web interface for generating videos using the FastVideo framework. The demo allows users to create videos from text prompts with various customization options.
 
@@ -13,25 +13,12 @@ The demo uses the FastVideo framework to generate videos based on text prompts.
 
 ---
 
-## Requirements
-
-- Linux-based OS
-- Python 3.10
-- Cuda 12.4
-- FastVideo
-
-## Installation
-
-```bash
-pip install fastvideo
-```
-
 ## Usage
 
 Run the demo with:
 
 ```bash
-python fastvideo/v1/examples/inference/gradio/gradio_demo.py
+python examples/inference/gradio/gradio_demo.py
 ```
 
 This will start a web server at `http://0.0.0.0:7860` where you can access the interface.
@@ -40,15 +27,6 @@ This will start a web server at `http://0.0.0.0:7860` where you can access the i
 
 ## Model Initialization
 
-```python
-args = FastVideoArgs(model_path="FastVideo/FastHunyuan-Diffusers", num_gpus=2)
-
-generator = VideoGenerator.from_pretrained(
-    model_path=args.model_path,
-    num_gpus=args.num_gpus
-)
-```
-
 This demo initializes a `VideoGenerator` with the minimum required arguments for inference. Users can seamlessly adjust inference options between generations, including prompts, resolution, video length, or even the number of inference steps, *without ever needing to reload the model*.
 
 ## Video Generation
@@ -78,4 +56,4 @@ The interface is built with several components:
 - **Guidance Scale**: Control how closely the generation follows the prompt
 - **Inference Steps**: More steps can improve quality but take longer
 - **Negative Prompt**: Specify what you don't want to see in the video
-- **Seed**: Control randomness for reproducible results
+- **Seed**: Control randomness for reproducible results

examples/inference/gradio/gradio_demo.py

@@ -0,0 +1,169 @@
+import argparse
+import os
+from copy import deepcopy
+
+import gradio as gr
+import torch
+
+from fastvideo import VideoGenerator
+from fastvideo.v1.configs.sample.base import SamplingParam
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="FastVideo Gradio Demo")
+    parser.add_argument("--model_path",
+                        type=str,
+                        default="FastVideo/FastHunyuan-diffusers",
+                        help="Path to the model")
+    parser.add_argument("--num_gpus",
+                        type=int,
+                        default=1,
+                        help="Number of GPUs to use")
+    parser.add_argument("--output_path",
+                        type=str,
+                        default="outputs",
+                        help="Path to save generated videos")
+    parsed_args = parser.parse_args()
+
+    # args = FastVideoArgs(model_path="FastVideo/FastHunyuan-Diffusers", num_gpus=2)
+
+    generator = VideoGenerator.from_pretrained(
+        model_path=parsed_args.model_path, num_gpus=parsed_args.num_gpus)
+
+    default_params = SamplingParam.from_pretrained(parsed_args.model_path)
+
+    def generate_video(
+        prompt,
+        negative_prompt,
+        use_negative_prompt,
+        seed,
+        guidance_scale,
+        num_frames,
+        height,
+        width,
+        num_inference_steps,
+        randomize_seed=False,
+    ):
+        params = deepcopy(default_params)
+        params.prompt = prompt
+        params.negative_prompt = negative_prompt
+        params.seed = seed
+        params.guidance_scale = guidance_scale
+        params.num_frames = num_frames
+        params.height = height
+        params.width = width
+        params.num_inference_steps = num_inference_steps
+
+        if randomize_seed:
+            params.seed = torch.randint(0, 1000000, (1, )).item()
+
+        if not use_negative_prompt:
+            params.negative_prompt = None
+
+        generator.generate_video(prompt=prompt, sampling_param=params)
+
+        output_path = os.path.join(parsed_args.output_path,
+                                   f"{params.prompt[:100]}.mp4")
+
+        return output_path, params.seed
+
+    examples = [
+        "A hand enters the frame, pulling a sheet of plastic wrap over three balls of dough placed on a wooden surface. The plastic wrap is stretched to cover the dough more securely. The hand adjusts the wrap, ensuring that it is tight and smooth over the dough. The scene focuses on the hand’s movements as it secures the edges of the plastic wrap. No new objects appear, and the camera remains stationary, focusing on the action of covering the dough.",
+        "A vintage train snakes through the mountains, its plume of white steam rising dramatically against the jagged peaks. The cars glint in the late afternoon sun, their deep crimson and gold accents lending a touch of elegance. The tracks carve a precarious path along the cliffside, revealing glimpses of a roaring river far below. Inside, passengers peer out the large windows, their faces lit with awe as the landscape unfolds.",
+        "A crowded rooftop bar buzzes with energy, the city skyline twinkling like a field of stars in the background. Strings of fairy lights hang above, casting a warm, golden glow over the scene. Groups of people gather around high tables, their laughter blending with the soft rhythm of live jazz. The aroma of freshly mixed cocktails and charred appetizers wafts through the air, mingling with the cool night breeze.",
+    ]
+
+    with gr.Blocks() as demo:
+        gr.Markdown("# FastVideo Inference Demo")
+
+        with gr.Group():
+            with gr.Row():
+                prompt = gr.Text(
+                    label="Prompt",
+                    show_label=False,
+                    max_lines=1,
+                    placeholder="Enter your prompt",
+                    container=False,
+                )
+                run_button = gr.Button("Run", scale=0)
+            result = gr.Video(label="Result", show_label=False)
+
+        with gr.Accordion("Advanced options", open=False):
+            with gr.Group():
+                with gr.Row():
+                    height = gr.Slider(
+                        label="Height",
+                        minimum=256,
+                        maximum=1024,
+                        step=32,
+                        value=default_params.height,
+                    )
+                    width = gr.Slider(label="Width",
+                                      minimum=256,
+                                      maximum=1024,
+                                      step=32,
+                                      value=default_params.width)
+
+                with gr.Row():
+                    num_frames = gr.Slider(
+                        label="Number of Frames",
+                        minimum=21,
+                        maximum=163,
+                        value=default_params.num_frames,
+                    )
+                    guidance_scale = gr.Slider(
+                        label="Guidance Scale",
+                        minimum=1,
+                        maximum=12,
+                        value=default_params.guidance_scale,
+                    )
+                    num_inference_steps = gr.Slider(
+                        label="Inference Steps",
+                        minimum=4,
+                        maximum=100,
+                        value=default_params.num_inference_steps,
+                    )
+
+                with gr.Row():
+                    use_negative_prompt = gr.Checkbox(
+                        label="Use negative prompt", value=False)
+                    negative_prompt = gr.Text(
+                        label="Negative prompt",
+                        max_lines=1,
+                        placeholder="Enter a negative prompt",
+                        visible=False,
+                    )
+
+                seed = gr.Slider(label="Seed",
+                                 minimum=0,
+                                 maximum=1000000,
+                                 step=1,
+                                 value=default_params.seed)
+                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                seed_output = gr.Number(label="Used Seed")
+
+        gr.Examples(examples=examples, inputs=prompt)
+
+        use_negative_prompt.change(
+            fn=lambda x: gr.update(visible=x),
+            inputs=use_negative_prompt,
+            outputs=default_params.negative_prompt,
+        )
+
+        run_button.click(
+            fn=generate_video,
+            inputs=[
+                prompt,
+                negative_prompt,
+                use_negative_prompt,
+                seed,
+                guidance_scale,
+                num_frames,
+                height,
+                width,
+                num_inference_steps,
+                randomize_seed,
+            ],
+            outputs=[result, seed_output],
+        )
+
+    demo.queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)

examples/inference/optimizations/README.md

@@ -0,0 +1,9 @@
+# Optimization Examples
+
+```bash
+python examples/inference/optimizations/attention_example.py
+```
+
+```bash
+python examples/inference/optimizations/teacache_example.py
+```

examples/inference/optimizations/attention_example.py

@@ -0,0 +1,33 @@
+import os
+import time
+
+from fastvideo import VideoGenerator
+
+def main():
+    # set the attention backend 
+    os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
+
+    start_time = time.perf_counter()
+    gen = VideoGenerator.from_pretrained(
+        model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        num_gpus=1,
+    )
+    load_time = time.perf_counter() - start_time
+    print(f"Model loading time: {load_time:.2f} seconds")
+
+    gen_start_time = time.perf_counter()
+
+    gen.generate_video(
+        prompt=
+        "Will Smith casually eats noodles, his relaxed demeanor contrasting with the energetic background of a bustling street food market. The scene captures a mix of humor and authenticity. Mid-shot framing, vibrant lighting.",
+        seed=1024,
+        output_path="example_outputs/")
+    
+    generation_time = time.perf_counter() - gen_start_time
+    print(f"Video generation time: {generation_time:.2f} seconds")
+
+    total_time = time.perf_counter() - start_time
+    print(f"Total execution time: {total_time:.2f} seconds")
+
+if __name__ == "__main__":
+    main()

examples/inference/optimizations/teacache_example.py

@@ -0,0 +1,44 @@
+import time
+
+from fastvideo import VideoGenerator, SamplingParam
+
+
+def main():
+    start_time = time.perf_counter()
+
+    gen = VideoGenerator.from_pretrained(
+        model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        num_gpus=1,
+        use_cpu_offload=False,
+    )
+    load_time = time.perf_counter() - start_time
+    print(f"Model loading time: {load_time:.2f} seconds")
+
+    gen_start_time = time.perf_counter()
+
+    params = SamplingParam.from_pretrained(
+        model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+    )
+    # this controls the threshold for the tea cache
+    params.teacache_params.teacache_thresh = 0.08
+    gen.generate_video(
+        prompt=
+        "Will Smith casually eats noodles, his relaxed demeanor contrasting with the energetic background of a bustling street food market. The scene captures a mix of humor and authenticity. Mid-shot framing, vibrant lighting.",
+        sampling_param=params,
+        height=480,
+        width=832,
+        num_frames=61,  # 85 ,77 
+        num_inference_steps=50,
+        enable_teacache=True,
+        seed=1024,
+        output_path="example_outputs/")
+    
+    generation_time = time.perf_counter() - gen_start_time
+    print(f"Video generation time: {generation_time:.2f} seconds")
+
+    total_time = time.perf_counter() - start_time
+    print(f"Total execution time: {total_time:.2f} seconds")
+
+
+if __name__ == "__main__":
+    main()

fastvideo/__init__.py

@@ -1,5 +1,6 @@
 from fastvideo.v1.configs.pipelines import PipelineConfig
 from fastvideo.v1.configs.sample import SamplingParam
 from fastvideo.v1.entrypoints.video_generator import VideoGenerator
+from fastvideo.version import __version__
 
-__all__ = ["VideoGenerator", "PipelineConfig", "SamplingParam"]
+__all__ = ["VideoGenerator", "PipelineConfig", "SamplingParam", "__version__"]

fastvideo/distill.py

@@ -452,7 +452,7 @@ def main(args):
         return phase
 
     for step in range(init_steps + 1, args.max_train_steps + 1):
-        start_time = time.time()
+        start_time = time.perf_counter()
         assert args.multi_phased_distill_schedule is not None
         num_phases = get_num_phases(args.multi_phased_distill_schedule, step)
 
@@ -482,7 +482,7 @@ def main(args):
             args.hunyuan_teacher_disable_cfg,
         )
 
-        step_time = time.time() - start_time
+        step_time = time.perf_counter() - start_time
         step_times.append(step_time)
         avg_step_time = sum(step_times) / len(step_times)
 

fastvideo/distill_adv.py

@@ -517,7 +517,7 @@ def main(args):
     for step in range(init_steps + 1, args.max_train_steps + 1):
         assert args.multi_phased_distill_schedule is not None
         num_phases = get_num_phases(args.multi_phased_distill_schedule, step)
-        start_time = time.time()
+        start_time = time.perf_counter()
         (
             generator_loss,
             generator_grad_norm,
@@ -547,7 +547,7 @@ def main(args):
             args.discriminator_head_stride,
         )
 
-        step_time = time.time() - start_time
+        step_time = time.perf_counter() - start_time
         step_times.append(step_time)
         avg_step_time = sum(step_times) / len(step_times)
 

fastvideo/models/hunyuan/idle_config.py

@@ -237,7 +237,7 @@ def add_inference_args(parser: argparse.ArgumentParser):
         type=str,
         default="540p",
         choices=["540p", "720p"],
-        help="Root path of all the models, including t2v models and extra models.",
+        help="The resolution of the model.",
     )
     group.add_argument(
         "--load-key",
@@ -361,7 +361,7 @@ def add_parallel_args(parser: argparse.ArgumentParser):
         "--ring-degree",
         type=int,
         default=1,
-        help="Ulysses degree.",
+        help="Ring degree.",
     )
 
     return parser

fastvideo/models/hunyuan/inference.py

@@ -17,7 +17,7 @@ from fastvideo.models.hunyuan.vae import load_vae
 from fastvideo.utils.parallel_states import nccl_info
 
 
-class Inference(object):
+class Inference:
 
     def __init__(
         self,
@@ -452,7 +452,7 @@ class HunyuanVideoSampler(Inference):
         # ========================================================================
         # Pipeline inference
         # ========================================================================
-        start_time = time.time()
+        start_time = time.perf_counter()
         samples = self.pipeline(
             prompt=prompt,
             height=target_height,
@@ -476,7 +476,7 @@ class HunyuanVideoSampler(Inference):
         out_dict["samples"] = samples
         out_dict["prompts"] = prompt
 
-        gen_time = time.time() - start_time
+        gen_time = time.perf_counter() - start_time
         logger.info(f"Success, time: {gen_time}")
 
         return out_dict

fastvideo/models/hunyuan/prompt_rewrite.py

@@ -41,7 +41,7 @@ def get_rewrite_prompt(ori_prompt, mode="Normal"):
     elif mode == "Master":
         prompt = master_mode_prompt.format(input=ori_prompt)
     else:
-        raise Exception("Only supports Normal and Normal", mode)
+        raise Exception("Only supports Normal and Master mode, but got {}".format(mode))
     return prompt
 
 

fastvideo/models/stepvideo/text_encoder/stepllm.py

@@ -267,25 +267,25 @@ class Step1Model(PreTrainedModel):
 class STEP1TextEncoder(torch.nn.Module):
 
     def __init__(self, model_dir, max_length=320):
-        super(STEP1TextEncoder, self).__init__()
+        super()
         self.max_length = max_length
         self.text_tokenizer = Wrapped_StepChatTokenizer(os.path.join(model_dir, 'step1_chat_tokenizer.model'))
         text_encoder = Step1Model.from_pretrained(model_dir)
         self.text_encoder = text_encoder.eval().to(torch.bfloat16)
 
     @torch.no_grad
+    @torch.autocast(device_type='cuda', dtype=torch.bfloat16)
     def forward(self, prompts, with_mask=True, max_length=None):
         self.device = next(self.text_encoder.parameters()).device
-        with torch.no_grad(), torch.cuda.amp.autocast(dtype=torch.bfloat16):
-            if type(prompts) is str:
-                prompts = [prompts]
+        if type(prompts) is str:
+            prompts = [prompts]
 
-            txt_tokens = self.text_tokenizer(prompts,
-                                             max_length=max_length or self.max_length,
-                                             padding="max_length",
-                                             truncation=True,
-                                             return_tensors="pt")
-            y = self.text_encoder(txt_tokens.input_ids.to(self.device),
+        txt_tokens = self.text_tokenizer(prompts,
+                                            max_length=max_length or self.max_length,
+                                            padding="max_length",
+                                            truncation=True,
+                                            return_tensors="pt")
+        y = self.text_encoder(txt_tokens.input_ids.to(self.device),
                                   attention_mask=txt_tokens.attention_mask.to(self.device) if with_mask else None)
-            y_mask = txt_tokens.attention_mask
+        y_mask = txt_tokens.attention_mask
         return y.transpose(0, 1), y_mask

fastvideo/train.py

@@ -364,7 +364,7 @@ def main(args):
     for i in range(init_steps):
         next(loader)
     for step in range(init_steps + 1, args.max_train_steps + 1):
-        start_time = time.time()
+        start_time = time.perf_counter()
         loss, grad_norm = train_one_step(
             transformer,
             args.model_type,
@@ -383,7 +383,7 @@ def main(args):
             args.mode_scale,
         )
 
-        step_time = time.time() - start_time
+        step_time = time.perf_counter() - start_time
         step_times.append(step_time)
         avg_step_time = sum(step_times) / len(step_times)
 

fastvideo/utils/collect_env.py

@@ -0,0 +1,776 @@
+# SPDX-License-Identifier: Apache-2.0
+
+# type: ignore
+# ruff: noqa
+# code borrowed from https://github.com/pytorch/pytorch/blob/main/torch/utils/collect_env.py
+# and vllm: https://github.com/vllm-project/vllm/blob/main/vllm/collect_env.py
+
+import datetime
+import locale
+import os
+import re
+import subprocess
+import sys
+# Unlike the rest of the PyTorch this file must be python2 compliant.
+# This script outputs relevant system environment info
+# Run it with `python collect_env.py` or `python -m torch.utils.collect_env`
+from collections import namedtuple
+
+from fastvideo.v1.envs import environment_variables
+
+try:
+    import torch
+    TORCH_AVAILABLE = True
+except (ImportError, NameError, AttributeError, OSError):
+    TORCH_AVAILABLE = False
+
+# System Environment Information
+SystemEnv = namedtuple(
+    'SystemEnv',
+    [
+        'torch_version',
+        'is_debug_build',
+        'cuda_compiled_version',
+        'gcc_version',
+        'clang_version',
+        'cmake_version',
+        'os',
+        'libc_version',
+        'python_version',
+        'python_platform',
+        'is_cuda_available',
+        'cuda_runtime_version',
+        'cuda_module_loading',
+        'nvidia_driver_version',
+        'nvidia_gpu_models',
+        'cudnn_version',
+        'pip_version',  # 'pip' or 'pip3'
+        'pip_packages',
+        'conda_packages',
+        'hip_compiled_version',
+        'hip_runtime_version',
+        'miopen_runtime_version',
+        'caching_allocator_config',
+        'is_xnnpack_available',
+        'cpu_info',
+        'fastvideo_version',
+        'fastvideo_build_flags',
+        'gpu_topo',
+        'env_vars',
+    ])
+
+DEFAULT_CONDA_PATTERNS = {
+    "torch",
+    "numpy",
+    "cudatoolkit",
+    "soumith",
+    "mkl",
+    "magma",
+    "triton",
+    "optree",
+    "nccl",
+    "transformers",
+    "accelerate",
+    "peft",
+    "zmq",
+    "nvidia",
+    "pynvml",
+}
+
+DEFAULT_PIP_PATTERNS = {
+    "torch",
+    "numpy",
+    "mypy",
+    "flake8",
+    "triton",
+    "optree",
+    "onnx",
+    "nccl",
+    "transformers",
+    "accelerate",
+    "peft",
+    "zmq",
+    "nvidia",
+    "pynvml",
+}
+
+
+def run(command):
+    """Return (return-code, stdout, stderr)."""
+    shell = True if type(command) is str else False
+    p = subprocess.Popen(command,
+                         stdout=subprocess.PIPE,
+                         stderr=subprocess.PIPE,
+                         shell=shell)
+    raw_output, raw_err = p.communicate()
+    rc = p.returncode
+    if get_platform() == 'win32':
+        enc = 'oem'
+    else:
+        enc = locale.getpreferredencoding()
+    output = raw_output.decode(enc)
+    if command == 'nvidia-smi topo -m':
+        # don't remove the leading whitespace of `nvidia-smi topo -m`
+        #   because they are meaningful
+        output = output.rstrip()
+    else:
+        output = output.strip()
+    err = raw_err.decode(enc)
+    return rc, output, err.strip()
+
+
+def run_and_read_all(run_lambda, command):
+    """Run command using run_lambda; reads and returns entire output if rc is 0."""
+    rc, out, _ = run_lambda(command)
+    if rc != 0:
+        return None
+    return out
+
+
+def run_and_parse_first_match(run_lambda, command, regex):
+    """Run command using run_lambda, returns the first regex match if it exists."""
+    rc, out, _ = run_lambda(command)
+    if rc != 0:
+        return None
+    match = re.search(regex, out)
+    if match is None:
+        return None
+    return match.group(1)
+
+
+def run_and_return_first_line(run_lambda, command):
+    """Run command using run_lambda and returns first line if output is not empty."""
+    rc, out, _ = run_lambda(command)
+    if rc != 0:
+        return None
+    return out.split('\n')[0]
+
+
+def get_conda_packages(run_lambda, patterns=None):
+    if patterns is None:
+        patterns = DEFAULT_CONDA_PATTERNS
+    conda = os.environ.get('CONDA_EXE', 'conda')
+    out = run_and_read_all(run_lambda, "{} list".format(conda))
+    if out is None:
+        return out
+
+    return "\n".join(line for line in out.splitlines()
+                     if not line.startswith("#") and any(name in line
+                                                         for name in patterns))
+
+
+def get_gcc_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'gcc --version', r'gcc (.*)')
+
+
+def get_clang_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'clang --version',
+                                     r'clang version (.*)')
+
+
+def get_cmake_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'cmake --version',
+                                     r'cmake (.*)')
+
+
+def get_nvidia_driver_version(run_lambda):
+    if get_platform() == 'darwin':
+        cmd = 'kextstat | grep -i cuda'
+        return run_and_parse_first_match(run_lambda, cmd,
+                                         r'com[.]nvidia[.]CUDA [(](.*?)[)]')
+    smi = get_nvidia_smi()
+    return run_and_parse_first_match(run_lambda, smi, r'Driver Version: (.*?) ')
+
+
+def get_gpu_info(run_lambda):
+    if get_platform() == 'darwin' or (TORCH_AVAILABLE and hasattr(
+            torch.version, 'hip') and torch.version.hip is not None):
+        if TORCH_AVAILABLE and torch.cuda.is_available():
+            if torch.version.hip is not None:
+                prop = torch.cuda.get_device_properties(0)
+                if hasattr(prop, "gcnArchName"):
+                    gcnArch = " ({})".format(prop.gcnArchName)
+                else:
+                    gcnArch = "NoGCNArchNameOnOldPyTorch"
+            else:
+                gcnArch = ""
+            return torch.cuda.get_device_name(None) + gcnArch
+        return None
+    smi = get_nvidia_smi()
+    uuid_regex = re.compile(r' \(UUID: .+?\)')
+    rc, out, _ = run_lambda(smi + ' -L')
+    if rc != 0:
+        return None
+    # Anonymize GPUs by removing their UUID
+    return re.sub(uuid_regex, '', out)
+
+
+def get_running_cuda_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'nvcc --version',
+                                     r'release .+ V(.*)')
+
+
+def get_cudnn_version(run_lambda):
+    """Return a list of libcudnn.so; it's hard to tell which one is being used."""
+    if get_platform() == 'win32':
+        system_root = os.environ.get('SYSTEMROOT', 'C:\\Windows')
+        cuda_path = os.environ.get('CUDA_PATH', "%CUDA_PATH%")
+        where_cmd = os.path.join(system_root, 'System32', 'where')
+        cudnn_cmd = '{} /R "{}\\bin" cudnn*.dll'.format(where_cmd, cuda_path)
+    elif get_platform() == 'darwin':
+        # CUDA libraries and drivers can be found in /usr/local/cuda/. See
+        # https://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#install
+        # https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#installmac
+        # Use CUDNN_LIBRARY when cudnn library is installed elsewhere.
+        cudnn_cmd = 'ls /usr/local/cuda/lib/libcudnn*'
+    else:
+        cudnn_cmd = 'ldconfig -p | grep libcudnn | rev | cut -d" " -f1 | rev'
+    rc, out, _ = run_lambda(cudnn_cmd)
+    # find will return 1 if there are permission errors or if not found
+    if len(out) == 0 or (rc != 1 and rc != 0):
+        l = os.environ.get('CUDNN_LIBRARY')
+        if l is not None and os.path.isfile(l):
+            return os.path.realpath(l)
+        return None
+    files_set = set()
+    for fn in out.split('\n'):
+        fn = os.path.realpath(fn)  # eliminate symbolic links
+        if os.path.isfile(fn):
+            files_set.add(fn)
+    if not files_set:
+        return None
+    # Alphabetize the result because the order is non-deterministic otherwise
+    files = sorted(files_set)
+    if len(files) == 1:
+        return files[0]
+    result = '\n'.join(files)
+    return 'Probably one of the following:\n{}'.format(result)
+
+
+def get_nvidia_smi():
+    # Note: nvidia-smi is currently available only on Windows and Linux
+    smi = 'nvidia-smi'
+    if get_platform() == 'win32':
+        system_root = os.environ.get('SYSTEMROOT', 'C:\\Windows')
+        program_files_root = os.environ.get('PROGRAMFILES', 'C:\\Program Files')
+        legacy_path = os.path.join(program_files_root, 'NVIDIA Corporation',
+                                   'NVSMI', smi)
+        new_path = os.path.join(system_root, 'System32', smi)
+        smis = [new_path, legacy_path]
+        for candidate_smi in smis:
+            if os.path.exists(candidate_smi):
+                smi = '"{}"'.format(candidate_smi)
+                break
+    return smi
+
+
+def get_fastvideo_version():
+    return ""
+    from fastvideo import __version__, __version_tuple__
+
+    if __version__ == "dev":
+        return "N/A (dev)"
+    version_str = __version_tuple__[-1]
+    if isinstance(version_str, str) and version_str.startswith('g'):
+        # it's a dev build
+        if '.' in version_str:
+            # it's a dev build containing local changes
+            git_sha = version_str.split('.')[0][1:]
+            date = version_str.split('.')[-1][1:]
+            return f"{__version__} (git sha: {git_sha}, date: {date})"
+        else:
+            # it's a dev build without local changes
+            git_sha = version_str[1:]  # type: ignore
+            return f"{__version__} (git sha: {git_sha})"
+    return __version__
+
+
+def summarize_fastvideo_build_flags():
+    # This could be a static method if the flags are constant, or dynamic if you need to check environment variables, etc.
+    return 'CUDA Archs: {}; ROCm: {}; Neuron: {}'.format(
+        os.environ.get('TORCH_CUDA_ARCH_LIST', 'Not Set'),
+        'Enabled' if os.environ.get('ROCM_HOME') else 'Disabled',
+        'Enabled' if os.environ.get('NEURON_CORES') else 'Disabled',
+    )
+
+
+def get_gpu_topo(run_lambda):
+    output = None
+
+    if get_platform() == 'linux':
+        output = run_and_read_all(run_lambda, 'nvidia-smi topo -m')
+        if output is None:
+            output = run_and_read_all(run_lambda, 'rocm-smi --showtopo')
+
+    return output
+
+
+# example outputs of CPU infos
+#  * linux
+#    Architecture:            x86_64
+#      CPU op-mode(s):        32-bit, 64-bit
+#      Address sizes:         46 bits physical, 48 bits virtual
+#      Byte Order:            Little Endian
+#    CPU(s):                  128
+#      On-line CPU(s) list:   0-127
+#    Vendor ID:               GenuineIntel
+#      Model name:            Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
+#        CPU family:          6
+#        Model:               106
+#        Thread(s) per core:  2
+#        Core(s) per socket:  32
+#        Socket(s):           2
+#        Stepping:            6
+#        BogoMIPS:            5799.78
+#        Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr
+#                             sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl
+#                             xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16
+#                             pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand
+#                             hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced
+#                             fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap
+#                             avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1
+#                             xsaves wbnoinvd ida arat avx512vbmi pku ospke avx512_vbmi2 gfni vaes vpclmulqdq
+#                             avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid md_clear flush_l1d arch_capabilities
+#    Virtualization features:
+#      Hypervisor vendor:     KVM
+#      Virtualization type:   full
+#    Caches (sum of all):
+#      L1d:                   3 MiB (64 instances)
+#      L1i:                   2 MiB (64 instances)
+#      L2:                    80 MiB (64 instances)
+#      L3:                    108 MiB (2 instances)
+#    NUMA:
+#      NUMA node(s):          2
+#      NUMA node0 CPU(s):     0-31,64-95
+#      NUMA node1 CPU(s):     32-63,96-127
+#    Vulnerabilities:
+#      Itlb multihit:         Not affected
+#      L1tf:                  Not affected
+#      Mds:                   Not affected
+#      Meltdown:              Not affected
+#      Mmio stale data:       Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
+#      Retbleed:              Not affected
+#      Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl and seccomp
+#      Spectre v1:            Mitigation; usercopy/swapgs barriers and __user pointer sanitization
+#      Spectre v2:            Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
+#      Srbds:                 Not affected
+#      Tsx async abort:       Not affected
+#  * win32
+#    Architecture=9
+#    CurrentClockSpeed=2900
+#    DeviceID=CPU0
+#    Family=179
+#    L2CacheSize=40960
+#    L2CacheSpeed=
+#    Manufacturer=GenuineIntel
+#    MaxClockSpeed=2900
+#    Name=Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
+#    ProcessorType=3
+#    Revision=27142
+#
+#    Architecture=9
+#    CurrentClockSpeed=2900
+#    DeviceID=CPU1
+#    Family=179
+#    L2CacheSize=40960
+#    L2CacheSpeed=
+#    Manufacturer=GenuineIntel
+#    MaxClockSpeed=2900
+#    Name=Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
+#    ProcessorType=3
+#    Revision=27142
+
+
+def get_cpu_info(run_lambda):
+    rc, out, err = 0, '', ''
+    if get_platform() == 'linux':
+        rc, out, err = run_lambda('lscpu')
+    elif get_platform() == 'win32':
+        rc, out, err = run_lambda(
+            'wmic cpu get Name,Manufacturer,Family,Architecture,ProcessorType,DeviceID, \
+        CurrentClockSpeed,MaxClockSpeed,L2CacheSize,L2CacheSpeed,Revision /VALUE'
+        )
+    elif get_platform() == 'darwin':
+        rc, out, err = run_lambda("sysctl -n machdep.cpu.brand_string")
+    cpu_info = 'None'
+    if rc == 0:
+        cpu_info = out
+    else:
+        cpu_info = err
+    return cpu_info
+
+
+def get_platform():
+    if sys.platform.startswith('linux'):
+        return 'linux'
+    elif sys.platform.startswith('win32'):
+        return 'win32'
+    elif sys.platform.startswith('cygwin'):
+        return 'cygwin'
+    elif sys.platform.startswith('darwin'):
+        return 'darwin'
+    else:
+        return sys.platform
+
+
+def get_mac_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'sw_vers -productVersion',
+                                     r'(.*)')
+
+
+def get_windows_version(run_lambda):
+    system_root = os.environ.get('SYSTEMROOT', 'C:\\Windows')
+    wmic_cmd = os.path.join(system_root, 'System32', 'Wbem', 'wmic')
+    findstr_cmd = os.path.join(system_root, 'System32', 'findstr')
+    return run_and_read_all(
+        run_lambda,
+        '{} os get Caption | {} /v Caption'.format(wmic_cmd, findstr_cmd))
+
+
+def get_lsb_version(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'lsb_release -a',
+                                     r'Description:\t(.*)')
+
+
+def check_release_file(run_lambda):
+    return run_and_parse_first_match(run_lambda, 'cat /etc/*-release',
+                                     r'PRETTY_NAME="(.*)"')
+
+
+def get_os(run_lambda):
+    from platform import machine
+    platform = get_platform()
+
+    if platform == 'win32' or platform == 'cygwin':
+        return get_windows_version(run_lambda)
+
+    if platform == 'darwin':
+        version = get_mac_version(run_lambda)
+        if version is None:
+            return None
+        return 'macOS {} ({})'.format(version, machine())
+
+    if platform == 'linux':
+        # Ubuntu/Debian based
+        desc = get_lsb_version(run_lambda)
+        if desc is not None:
+            return '{} ({})'.format(desc, machine())
+
+        # Try reading /etc/*-release
+        desc = check_release_file(run_lambda)
+        if desc is not None:
+            return '{} ({})'.format(desc, machine())
+
+        return '{} ({})'.format(platform, machine())
+
+    # Unknown platform
+    return platform
+
+
+def get_python_platform():
+    import platform
+    return platform.platform()
+
+
+def get_libc_version():
+    import platform
+    if get_platform() != 'linux':
+        return 'N/A'
+    return '-'.join(platform.libc_ver())
+
+
+def get_pip_packages(run_lambda, patterns=None):
+    """Return `pip list` output. Note: will also find conda-installed pytorch and numpy packages."""
+    if patterns is None:
+        patterns = DEFAULT_PIP_PATTERNS
+
+    def run_with_pip():
+        try:
+            import importlib.util
+            pip_spec = importlib.util.find_spec('pip')
+            pip_available = pip_spec is not None
+        except ImportError:
+            pip_available = False
+
+        if pip_available:
+            cmd = [sys.executable, '-mpip', 'list', '--format=freeze']
+        elif os.environ.get("UV") is not None:
+            print("uv is set")
+            cmd = ["uv", "pip", "list", "--format=freeze"]
+        else:
+            raise RuntimeError(
+                "Could not collect pip list output (pip or uv module not available)"
+            )
+
+        out = run_and_read_all(run_lambda, cmd)
+        return "\n".join(line for line in out.splitlines()
+                         if any(name in line for name in patterns))
+
+    pip_version = 'pip3' if sys.version[0] == '3' else 'pip'
+    out = run_with_pip()
+    return pip_version, out
+
+
+def get_cachingallocator_config():
+    ca_config = os.environ.get('PYTORCH_CUDA_ALLOC_CONF', '')
+    return ca_config
+
+
+def get_cuda_module_loading_config():
+    if TORCH_AVAILABLE and torch.cuda.is_available():
+        torch.cuda.init()
+        config = os.environ.get('CUDA_MODULE_LOADING', '')
+        return config
+    else:
+        return "N/A"
+
+
+def is_xnnpack_available():
+    if TORCH_AVAILABLE:
+        import torch.backends.xnnpack
+        return str(torch.backends.xnnpack.enabled)  # type: ignore[attr-defined]
+    else:
+        return "N/A"
+
+
+def get_env_vars():
+    env_vars = ''
+    secret_terms = ('secret', 'token', 'api', 'access', 'password')
+    report_prefix = ("TORCH", "NCCL", "PYTORCH", "CUDA", "CUBLAS", "CUDNN",
+                     "OMP_", "MKL_", "NVIDIA")
+    for k, v in os.environ.items():
+        if any(term in k.lower() for term in secret_terms):
+            continue
+        if k in environment_variables:
+            env_vars = env_vars + "{}={}".format(k, v) + "\n"
+        if k.startswith(report_prefix):
+            env_vars = env_vars + "{}={}".format(k, v) + "\n"
+
+    return env_vars
+
+
+def get_env_info():
+    run_lambda = run
+    pip_version, pip_list_output = get_pip_packages(run_lambda)
+
+    if TORCH_AVAILABLE:
+        version_str = torch.__version__
+        debug_mode_str = str(torch.version.debug)
+        cuda_available_str = str(torch.cuda.is_available())
+        cuda_version_str = torch.version.cuda
+        if not hasattr(torch.version,
+                       'hip') or torch.version.hip is None:  # cuda version
+            hip_compiled_version = hip_runtime_version = miopen_runtime_version = 'N/A'
+        else:  # HIP version
+
+            def get_version_or_na(cfg, prefix):
+                _lst = [s.rsplit(None, 1)[-1] for s in cfg if prefix in s]
+                return _lst[0] if _lst else 'N/A'
+
+            cfg = torch._C._show_config().split('\n')
+            hip_runtime_version = get_version_or_na(cfg, 'HIP Runtime')
+            miopen_runtime_version = get_version_or_na(cfg, 'MIOpen')
+            cuda_version_str = 'N/A'
+            hip_compiled_version = torch.version.hip
+    else:
+        version_str = debug_mode_str = cuda_available_str = cuda_version_str = 'N/A'
+        hip_compiled_version = hip_runtime_version = miopen_runtime_version = 'N/A'
+
+    sys_version = sys.version.replace("\n", " ")
+
+    conda_packages = get_conda_packages(run_lambda)
+
+    fastvideo_version = get_fastvideo_version()
+    fastvideo_build_flags = summarize_fastvideo_build_flags()
+    gpu_topo = get_gpu_topo(run_lambda)
+
+    return SystemEnv(
+        torch_version=version_str,
+        is_debug_build=debug_mode_str,
+        python_version='{} ({}-bit runtime)'.format(
+            sys_version,
+            sys.maxsize.bit_length() + 1),
+        python_platform=get_python_platform(),
+        is_cuda_available=cuda_available_str,
+        cuda_compiled_version=cuda_version_str,
+        cuda_runtime_version=get_running_cuda_version(run_lambda),
+        cuda_module_loading=get_cuda_module_loading_config(),
+        nvidia_gpu_models=get_gpu_info(run_lambda),
+        nvidia_driver_version=get_nvidia_driver_version(run_lambda),
+        cudnn_version=get_cudnn_version(run_lambda),
+        hip_compiled_version=hip_compiled_version,
+        hip_runtime_version=hip_runtime_version,
+        miopen_runtime_version=miopen_runtime_version,
+        pip_version=pip_version,
+        pip_packages=pip_list_output,
+        conda_packages=conda_packages,
+        os=get_os(run_lambda),
+        libc_version=get_libc_version(),
+        gcc_version=get_gcc_version(run_lambda),
+        clang_version=get_clang_version(run_lambda),
+        cmake_version=get_cmake_version(run_lambda),
+        caching_allocator_config=get_cachingallocator_config(),
+        is_xnnpack_available=is_xnnpack_available(),
+        cpu_info=get_cpu_info(run_lambda),
+        fastvideo_version=fastvideo_version,
+        fastvideo_build_flags=fastvideo_build_flags,
+        gpu_topo=gpu_topo,
+        env_vars=get_env_vars(),
+    )
+
+
+env_info_fmt = """
+PyTorch version: {torch_version}
+Is debug build: {is_debug_build}
+CUDA used to build PyTorch: {cuda_compiled_version}
+ROCM used to build PyTorch: {hip_compiled_version}
+
+OS: {os}
+GCC version: {gcc_version}
+Clang version: {clang_version}
+CMake version: {cmake_version}
+Libc version: {libc_version}
+
+Python version: {python_version}
+Python platform: {python_platform}
+Is CUDA available: {is_cuda_available}
+CUDA runtime version: {cuda_runtime_version}
+CUDA_MODULE_LOADING set to: {cuda_module_loading}
+GPU models and configuration: {nvidia_gpu_models}
+Nvidia driver version: {nvidia_driver_version}
+cuDNN version: {cudnn_version}
+HIP runtime version: {hip_runtime_version}
+MIOpen runtime version: {miopen_runtime_version}
+Is XNNPACK available: {is_xnnpack_available}
+
+CPU:
+{cpu_info}
+
+Versions of relevant libraries:
+{pip_packages}
+{conda_packages}
+""".strip()
+
+# both the above code and the following code use `strip()` to
+# remove leading/trailing whitespaces, so we need to add a newline
+# in between to separate the two sections
+env_info_fmt += "\n"
+
+env_info_fmt += """
+FastVideo Version: {fastvideo_version}
+FastVideo Build Flags:
+{fastvideo_build_flags}
+GPU Topology:
+{gpu_topo}
+
+{env_vars}
+""".strip()
+
+
+def pretty_str(envinfo):
+
+    def replace_nones(dct, replacement='Could not collect'):
+        for key in dct.keys():
+            if dct[key] is not None:
+                continue
+            dct[key] = replacement
+        return dct
+
+    def replace_bools(dct, true='Yes', false='No'):
+        for key in dct.keys():
+            if dct[key] is True:
+                dct[key] = true
+            elif dct[key] is False:
+                dct[key] = false
+        return dct
+
+    def prepend(text, tag='[prepend]'):
+        lines = text.split('\n')
+        updated_lines = [tag + line for line in lines]
+        return '\n'.join(updated_lines)
+
+    def replace_if_empty(text, replacement='No relevant packages'):
+        if text is not None and len(text) == 0:
+            return replacement
+        return text
+
+    def maybe_start_on_next_line(string):
+        # If `string` is multiline, prepend a \n to it.
+        if string is not None and len(string.split('\n')) > 1:
+            return '\n{}\n'.format(string)
+        return string
+
+    mutable_dict = envinfo._asdict()
+
+    # If nvidia_gpu_models is multiline, start on the next line
+    mutable_dict['nvidia_gpu_models'] = \
+        maybe_start_on_next_line(envinfo.nvidia_gpu_models)
+
+    # If the machine doesn't have CUDA, report some fields as 'No CUDA'
+    dynamic_cuda_fields = [
+        'cuda_runtime_version',
+        'nvidia_gpu_models',
+        'nvidia_driver_version',
+    ]
+    all_cuda_fields = dynamic_cuda_fields + ['cudnn_version']
+    all_dynamic_cuda_fields_missing = all(mutable_dict[field] is None
+                                          for field in dynamic_cuda_fields)
+    if TORCH_AVAILABLE and not torch.cuda.is_available(
+    ) and all_dynamic_cuda_fields_missing:
+        for field in all_cuda_fields:
+            mutable_dict[field] = 'No CUDA'
+        if envinfo.cuda_compiled_version is None:
+            mutable_dict['cuda_compiled_version'] = 'None'
+
+    # Replace True with Yes, False with No
+    mutable_dict = replace_bools(mutable_dict)
+
+    # Replace all None objects with 'Could not collect'
+    mutable_dict = replace_nones(mutable_dict)
+
+    # If either of these are '', replace with 'No relevant packages'
+    mutable_dict['pip_packages'] = replace_if_empty(
+        mutable_dict['pip_packages'])
+    mutable_dict['conda_packages'] = replace_if_empty(
+        mutable_dict['conda_packages'])
+
+    # Tag conda and pip packages with a prefix
+    # If they were previously None, they'll show up as ie '[conda] Could not collect'
+    if mutable_dict['pip_packages']:
+        mutable_dict['pip_packages'] = prepend(
+            mutable_dict['pip_packages'], '[{}] '.format(envinfo.pip_version))
+    if mutable_dict['conda_packages']:
+        mutable_dict['conda_packages'] = prepend(mutable_dict['conda_packages'],
+                                                 '[conda] ')
+    mutable_dict['cpu_info'] = envinfo.cpu_info
+    return env_info_fmt.format(**mutable_dict)
+
+
+def get_pretty_env_info():
+    return pretty_str(get_env_info())
+
+
+def main():
+    print("Collecting environment information...")
+    output = get_pretty_env_info()
+    print(output)
+
+    if TORCH_AVAILABLE and hasattr(torch, 'utils') and hasattr(
+            torch.utils, '_crash_handler'):
+        minidump_dir = torch.utils._crash_handler.DEFAULT_MINIDUMP_DIR
+        if sys.platform == "linux" and os.path.exists(minidump_dir):
+            dumps = [
+                os.path.join(minidump_dir, dump)
+                for dump in os.listdir(minidump_dir)
+            ]
+            latest = max(dumps, key=os.path.getctime)
+            ctime = os.path.getctime(latest)
+            creation_time = datetime.datetime.fromtimestamp(ctime).strftime(
+                '%Y-%m-%d %H:%M:%S')
+            msg = "\n*** Detected a minidump at {} created on {}, ".format(latest, creation_time) + \
+                  "if this is related to your bug please include it when you file a report ***"
+            print(msg, file=sys.stderr)
+
+
+if __name__ == '__main__':
+    main()

fastvideo/utils/env_utils.py

@@ -1,38 +0,0 @@
-import platform
-
-import accelerate
-import peft
-import torch
-import transformers
-from transformers.utils import is_torch_cuda_available, is_torch_npu_available
-
-VERSION = "1.2.0"
-
-if __name__ == "__main__":
-    info = {
-        "FastVideo version": VERSION,
-        "Platform": platform.platform(),
-        "Python version": platform.python_version(),
-        "PyTorch version": torch.__version__,
-        "Transformers version": transformers.__version__,
-        "Accelerate version": accelerate.__version__,
-        "PEFT version": peft.__version__,
-    }
-
-    if is_torch_cuda_available():
-        info["PyTorch version"] += " (GPU)"
-        info["GPU type"] = torch.cuda.get_device_name()
-
-    if is_torch_npu_available():
-        info["PyTorch version"] += " (NPU)"
-        info["NPU type"] = torch.npu.get_device_name()
-        info["CANN version"] = torch.version.cann  # codespell:ignore
-
-    try:
-        import bitsandbytes
-
-        info["Bitsandbytes version"] = bitsandbytes.__version__
-    except Exception:
-        pass
-
-    print("\n" + "\n".join([f"- {key}: {value}" for key, value in info.items()]) + "\n")

fastvideo/v1/attention/backends/abstract.py

@@ -56,20 +56,6 @@ class AttentionMetadata:
     # Current step of diffusion process
     current_timestep: int
 
-    # @property
-    # @abstractmethod
-    # def inference_metadata(self) -> Optional["AttentionMetadata"]:
-    #     """Return the attention metadata that's required to run prefill
-    #     attention."""
-    #     pass
-
-    # @property
-    # @abstractmethod
-    # def training_metadata(self) -> Optional["AttentionMetadata"]:
-    #     """Return the attention metadata that's required to run decode
-    #     attention."""
-    #     pass
-
     def asdict_zerocopy(self,
                         skip_fields: Optional[Set[str]] = None
                         ) -> Dict[str, Any]:
@@ -86,55 +72,6 @@ class AttentionMetadata:
 
 T = TypeVar("T", bound=AttentionMetadata)
 
-# class AttentionState(ABC, Generic[T]):
-#     """Holds attention backend-specific objects reused during the
-#     lifetime of the model runner."""
-
-#     @abstractmethod
-#     def __init__(self, runner: "ModelRunnerBase"):
-#         ...
-
-#     @abstractmethod
-#     @contextmanager
-#     def graph_capture(self, max_batch_size: int):
-#         """Context manager used when capturing CUDA graphs."""
-#         yield
-
-#     @abstractmethod
-#     def graph_clone(self, batch_size: int) -> "AttentionState[T]":
-#         """Clone attention state to save in CUDA graph metadata."""
-#         ...
-
-#     @abstractmethod
-#     def graph_capture_get_metadata_for_batch(
-#             self,
-#             batch_size: int,
-#             is_encoder_decoder_model: bool = False) -> T:
-#         """Get attention metadata for CUDA graph capture of batch_size."""
-#         ...
-
-#     @abstractmethod
-#     def get_graph_input_buffers(
-#             self,
-#             attn_metadata: T,
-#             is_encoder_decoder_model: bool = False) -> Dict[str, Any]:
-#         """Get attention-specific input buffers for CUDA graph capture."""
-#         ...
-
-#     @abstractmethod
-#     def prepare_graph_input_buffers(
-#             self,
-#             input_buffers: Dict[str, Any],
-#             attn_metadata: T,
-#             is_encoder_decoder_model: bool = False) -> None:
-#         """In-place modify input buffers dict for CUDA graph replay."""
-#         ...
-
-#     @abstractmethod
-#     def begin_forward(self, model_input: "ModelRunnerInputBase") -> None:
-#         """Prepare state for forward pass."""
-#         ...
-
 
 class AttentionMetadataBuilder(ABC, Generic[T]):
     """Abstract class for attention metadata builders."""

fastvideo/v1/attention/layer.py

@@ -84,9 +84,6 @@ class DistributedAttention(nn.Module):
         # Check input shapes
         assert q.dim() == 4 and k.dim() == 4 and v.dim(
         ) == 4, "Expected 4D tensors"
-        # assert bs = 1
-        assert q.shape[
-            0] == 1, "Batch size must be 1, and there should be no padding tokens"
         batch_size, seq_len, num_heads, head_dim = q.shape
         local_rank = get_sequence_model_parallel_rank()
         world_size = get_sequence_model_parallel_world_size()

fastvideo/v1/configs/fasthunyuan_t2v.json

@@ -0,0 +1,48 @@
+{
+  "embedded_cfg_scale": 6,
+  "flow_shift": 17,
+  "use_cpu_offload": false,
+  "disable_autocast": false,
+  "precision": "bf16",
+  "vae_precision": "fp16",
+  "vae_tiling": true,
+  "vae_sp": true,
+  "vae_config": {
+    "load_encoder": false,
+    "load_decoder": true,
+    "tile_sample_min_height": 256,
+    "tile_sample_min_width": 256,
+    "tile_sample_min_num_frames": 16,
+    "tile_sample_stride_height": 192,
+    "tile_sample_stride_width": 192,
+    "tile_sample_stride_num_frames": 12,
+    "blend_num_frames": 4,
+    "use_tiling": true,
+    "use_temporal_tiling": true,
+    "use_parallel_tiling": true
+  },
+  "dit_config": {
+    "prefix": "Hunyuan",
+    "quant_config": null
+  },
+  "text_encoder_precisions": [
+    "fp16",
+    "fp16"
+  ],
+  "text_encoder_configs": [
+    {
+      "prefix": "llama",
+      "quant_config": null,
+      "lora_config": null
+    },
+    {
+      "prefix": "clip",
+      "quant_config": null,
+      "lora_config": null,
+      "num_hidden_layers_override": null,
+      "require_post_norm": null
+    }
+  ],
+  "mask_strategy_file_path": null,
+  "enable_torch_compile": false
+}

fastvideo/v1/configs/models/base.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass, fields
+from dataclasses import dataclass, field, fields
 from typing import Any, Dict
 
 from fastvideo.v1.logger import init_logger
@@ -18,7 +18,7 @@ class ArchConfig:
 class ModelConfig:
     # Every model config parameter can be categorized into either ArchConfig or everything else
     # Diffuser/Transformer parameters
-    arch_config: ArchConfig = ArchConfig()
+    arch_config: ArchConfig = field(default_factory=ArchConfig)
 
     # FastVideo-specific parameters here
     # i.e. STA, quantization, teacache
@@ -30,6 +30,16 @@ class ModelConfig:
         raise AttributeError(
             f"'{type(self).__name__}' object has no attribute '{name}'")
 
+    def __getstate__(self):
+        # Return a dictionary of attributes to pickle
+        # Convert to dict and exclude any problematic attributes
+        state = self.__dict__.copy()
+        return state
+
+    def __setstate__(self, state):
+        # Restore instance attributes from the unpickled state
+        self.__dict__.update(state)
+
     # This should be used only when loading from transformers/diffusers
     def update_model_arch(self, source_model_dict: Dict[str, Any]) -> None:
         arch_config = self.arch_config

fastvideo/v1/configs/models/dits/__init__.py

@@ -1,4 +1,5 @@
 from fastvideo.v1.configs.models.dits.hunyuanvideo import HunyuanVideoConfig
+from fastvideo.v1.configs.models.dits.stepvideo import StepVideoConfig
 from fastvideo.v1.configs.models.dits.wanvideo import WanVideoConfig
 
-__all__ = ["HunyuanVideoConfig", "WanVideoConfig"]
+__all__ = ["HunyuanVideoConfig", "WanVideoConfig", "StepVideoConfig"]

fastvideo/v1/configs/models/dits/base.py

@@ -1,14 +1,15 @@
 from dataclasses import dataclass, field
-from typing import Optional, Tuple
+from typing import Any, Optional, Tuple
 
 from fastvideo.v1.configs.models.base import ArchConfig, ModelConfig
-from fastvideo.v1.configs.quantization import QuantizationConfig
+from fastvideo.v1.layers.quantization import QuantizationConfig
 from fastvideo.v1.platforms import _Backend
 
 
 @dataclass
 class DiTArchConfig(ArchConfig):
     _fsdp_shard_conditions: list = field(default_factory=list)
+    _compile_conditions: list = field(default_factory=list)
     _param_names_mapping: dict = field(default_factory=dict)
     _supported_attention_backends: Tuple[_Backend,
                                          ...] = (_Backend.SLIDING_TILE_ATTN,
@@ -20,11 +21,36 @@ class DiTArchConfig(ArchConfig):
     num_attention_heads: int = 0
     num_channels_latents: int = 0
 
+    def __post_init__(self) -> None:
+        if not self._compile_conditions:
+            self._compile_conditions = self._fsdp_shard_conditions.copy()
+
 
 @dataclass
 class DiTConfig(ModelConfig):
-    arch_config: DiTArchConfig = DiTArchConfig()
+    arch_config: DiTArchConfig = field(default_factory=DiTArchConfig)
 
     # FastVideoDiT-specific parameters
     prefix: str = ""
     quant_config: Optional[QuantizationConfig] = None
+
+    @staticmethod
+    def add_cli_args(parser: Any, prefix: str = "dit-config") -> Any:
+        """Add CLI arguments for DiTConfig fields"""
+        parser.add_argument(
+            f"--{prefix}.prefix",
+            type=str,
+            dest=f"{prefix.replace('-', '_')}.prefix",
+            default=DiTConfig.prefix,
+            help="Prefix for the DiT model",
+        )
+
+        parser.add_argument(
+            f"--{prefix}.quant-config",
+            type=str,
+            dest=f"{prefix.replace('-', '_')}.quant_config",
+            default=None,
+            help="Quantization configuration for the DiT model",
+        )
+
+        return parser

fastvideo/v1/configs/models/dits/hunyuanvideo.py

@@ -18,12 +18,19 @@ def is_refiner_block(n: str, m) -> bool:
     return "refiner" in n and str.isdigit(n.split(".")[-1])
 
 
+def is_txt_in(n: str, m) -> bool:
+    return n.split(".")[-1] == "txt_in"
+
+
 @dataclass
 class HunyuanVideoArchConfig(DiTArchConfig):
     _fsdp_shard_conditions: list = field(
         default_factory=lambda:
         [is_double_block, is_single_block, is_refiner_block])
 
+    _compile_conditions: list = field(
+        default_factory=lambda: [is_double_block, is_single_block, is_txt_in])
+
     _param_names_mapping: dict = field(
         default_factory=lambda: {
             # 1. context_embedder.time_text_embed submodules (specific rules, applied first):
@@ -158,12 +165,13 @@ class HunyuanVideoArchConfig(DiTArchConfig):
     qk_norm: str = "rms_norm"
 
     def __post_init__(self):
+        super().__post_init__()
         self.hidden_size: int = self.attention_head_dim * self.num_attention_heads
         self.num_channels_latents: int = self.in_channels
 
 
 @dataclass
 class HunyuanVideoConfig(DiTConfig):
-    arch_config: DiTArchConfig = HunyuanVideoArchConfig()
+    arch_config: DiTArchConfig = field(default_factory=HunyuanVideoArchConfig)
 
     prefix: str = "Hunyuan"

fastvideo/v1/configs/models/dits/stepvideo.py

@@ -0,0 +1,65 @@
+from dataclasses import dataclass, field
+from typing import List, Optional, Tuple, Union
+
+from fastvideo.v1.configs.models.dits.base import DiTArchConfig, DiTConfig
+
+
+def is_blocks(n: str, m) -> bool:
+    return "blocks" in n and str.isdigit(n.split(".")[-1])
+
+
+@dataclass
+class StepVideoArchConfig(DiTArchConfig):
+    _fsdp_shard_conditions: list = field(default_factory=lambda: [is_blocks])
+
+    _param_names_mapping: dict = field(
+        default_factory=lambda: {
+            # transformer block
+            r"^transformer_blocks\.(\d+)\.norm1\.(weight|bias)$":
+            r"transformer_blocks.\1.norm1.norm.\2",
+            r"^transformer_blocks\.(\d+)\.norm2\.(weight|bias)$":
+            r"transformer_blocks.\1.norm2.norm.\2",
+            r"^transformer_blocks\.(\d+)\.ff\.net\.0\.proj\.weight$":
+            r"transformer_blocks.\1.ff.fc_in.weight",
+            r"^transformer_blocks\.(\d+)\.ff\.net\.2\.weight$":
+            r"transformer_blocks.\1.ff.fc_out.weight",
+
+            # adanorm block
+            r"^adaln_single\.emb\.timestep_embedder\.linear_1\.(weight|bias)$":
+            r"adaln_single.emb.mlp.fc_in.\1",
+            r"^adaln_single\.emb\.timestep_embedder\.linear_2\.(weight|bias)$":
+            r"adaln_single.emb.mlp.fc_out.\1",
+
+            # caption projection
+            r"^caption_projection\.linear_1\.(weight|bias)$":
+            r"caption_projection.fc_in.\1",
+            r"^caption_projection\.linear_2\.(weight|bias)$":
+            r"caption_projection.fc_out.\1",
+        })
+
+    num_attention_heads: int = 48
+    attention_head_dim: int = 128
+    in_channels: int = 64
+    out_channels: Optional[int] = 64
+    num_layers: int = 48
+    dropout: float = 0.0
+    patch_size: int = 1
+    norm_type: str = "ada_norm_single"
+    norm_elementwise_affine: bool = False
+    norm_eps: float = 1e-6
+    caption_channels: Optional[Union[int, List[int], Tuple[int, ...]]] = field(
+        default_factory=lambda: [6144, 1024])
+    attention_type: Optional[str] = "torch"
+    use_additional_conditions: Optional[bool] = False
+
+    def __post_init__(self):
+        self.hidden_size = self.num_attention_heads * self.attention_head_dim
+        self.out_channels = self.in_channels if self.out_channels is None else self.out_channels
+        self.num_channels_latents = self.out_channels
+
+
+@dataclass
+class StepVideoConfig(DiTConfig):
+    arch_config: DiTArchConfig = field(default_factory=StepVideoArchConfig)
+
+    prefix: str = "StepVideo"

fastvideo/v1/configs/models/dits/wanvideo.py

@@ -70,6 +70,7 @@ class WanVideoArchConfig(DiTArchConfig):
     rope_max_seq_len: int = 1024
 
     def __post_init__(self):
+        super().__post_init__()
         self.out_channels = self.out_channels or self.in_channels
         self.hidden_size = self.num_attention_heads * self.attention_head_dim
         self.num_channels_latents = self.in_channels if self.added_kv_proj_dim is None else self.out_channels
@@ -77,6 +78,6 @@ class WanVideoArchConfig(DiTArchConfig):
 
 @dataclass
 class WanVideoConfig(DiTConfig):
-    arch_config: DiTArchConfig = WanVideoArchConfig()
+    arch_config: DiTArchConfig = field(default_factory=WanVideoArchConfig)
 
     prefix: str = "Wan"

fastvideo/v1/configs/models/encoders/__init__.py

@@ -1,4 +1,5 @@
-from fastvideo.v1.configs.models.encoders.base import (EncoderConfig,
+from fastvideo.v1.configs.models.encoders.base import (BaseEncoderOutput,
+                                                       EncoderConfig,
                                                        ImageEncoderConfig,
                                                        TextEncoderConfig)
 from fastvideo.v1.configs.models.encoders.clip import (CLIPTextConfig,
@@ -8,5 +9,6 @@ from fastvideo.v1.configs.models.encoders.t5 import T5Config
 
 __all__ = [
     "EncoderConfig", "TextEncoderConfig", "ImageEncoderConfig",
-    "CLIPTextConfig", "CLIPVisionConfig", "LlamaConfig", "T5Config"
+    "BaseEncoderOutput", "CLIPTextConfig", "CLIPVisionConfig", "LlamaConfig",
+    "T5Config"
 ]

fastvideo/v1/configs/models/encoders/base.py

@@ -1,8 +1,10 @@
 from dataclasses import dataclass, field
-from typing import Any, List, Optional, Tuple
+from typing import Any, Dict, List, Optional, Tuple
+
+import torch
 
 from fastvideo.v1.configs.models.base import ArchConfig, ModelConfig
-from fastvideo.v1.configs.quantization import QuantizationConfig
+from fastvideo.v1.layers.quantization import QuantizationConfig
 from fastvideo.v1.platforms import _Backend
 
 
@@ -30,15 +32,33 @@ class TextEncoderArchConfig(EncoderArchConfig):
     scalable_attention: bool = True
     tie_word_embeddings: bool = False
 
+    tokenizer_kwargs: Dict[str, Any] = field(default_factory=dict)
+
+    def __post_init__(self) -> None:
+        self.tokenizer_kwargs = {
+            "truncation": True,
+            "max_length": self.text_len,
+            "return_tensors": "pt",
+        }
+
 
 @dataclass
 class ImageEncoderArchConfig(EncoderArchConfig):
     pass
 
 
+@dataclass
+class BaseEncoderOutput:
+    last_hidden_state: Optional[torch.FloatTensor] = None
+    pooler_output: Optional[torch.FloatTensor] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
+    attentions: Optional[Tuple[torch.FloatTensor, ...]] = None
+    attention_mask: Optional[torch.Tensor] = None
+
+
 @dataclass
 class EncoderConfig(ModelConfig):
-    arch_config: ArchConfig = EncoderArchConfig()
+    arch_config: ArchConfig = field(default_factory=EncoderArchConfig)
 
     prefix: str = ""
     quant_config: Optional[QuantizationConfig] = None
@@ -47,9 +67,9 @@ class EncoderConfig(ModelConfig):
 
 @dataclass
 class TextEncoderConfig(EncoderConfig):
-    arch_config: ArchConfig = TextEncoderArchConfig()
+    arch_config: ArchConfig = field(default_factory=TextEncoderArchConfig)
 
 
 @dataclass
 class ImageEncoderConfig(EncoderConfig):
-    arch_config: ArchConfig = ImageEncoderArchConfig()
+    arch_config: ArchConfig = field(default_factory=ImageEncoderArchConfig)

fastvideo/v1/configs/models/encoders/clip.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (ImageEncoderArchConfig,
@@ -48,7 +48,8 @@ class CLIPVisionArchConfig(ImageEncoderArchConfig):
 
 @dataclass
 class CLIPTextConfig(TextEncoderConfig):
-    arch_config: TextEncoderArchConfig = CLIPTextArchConfig()
+    arch_config: TextEncoderArchConfig = field(
+        default_factory=CLIPTextArchConfig)
 
     num_hidden_layers_override: Optional[int] = None
     require_post_norm: Optional[bool] = None
@@ -57,7 +58,8 @@ class CLIPTextConfig(TextEncoderConfig):
 
 @dataclass
 class CLIPVisionConfig(ImageEncoderConfig):
-    arch_config: ImageEncoderArchConfig = CLIPVisionArchConfig()
+    arch_config: ImageEncoderArchConfig = field(
+        default_factory=CLIPVisionArchConfig)
 
     num_hidden_layers_override: Optional[int] = None
     require_post_norm: Optional[bool] = None

fastvideo/v1/configs/models/encoders/llama.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (TextEncoderArchConfig,
@@ -35,6 +35,6 @@ class LlamaArchConfig(TextEncoderArchConfig):
 
 @dataclass
 class LlamaConfig(TextEncoderConfig):
-    arch_config: TextEncoderArchConfig = LlamaArchConfig()
+    arch_config: TextEncoderArchConfig = field(default_factory=LlamaArchConfig)
 
     prefix: str = "llama"

fastvideo/v1/configs/models/encoders/t5.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (TextEncoderArchConfig,
@@ -31,15 +31,25 @@ class T5ArchConfig(TextEncoderArchConfig):
 
     # Referenced from https://github.com/huggingface/transformers/blob/main/src/transformers/models/t5/configuration_t5.py
     def __post_init__(self):
+        super().__post_init__()
         act_info = self.feed_forward_proj.split("-")
         self.dense_act_fn: str = act_info[-1]
         self.is_gated_act: bool = act_info[0] == "gated"
         if self.feed_forward_proj == "gated-gelu":
             self.dense_act_fn = "gelu_new"
 
+        self.tokenizer_kwargs = {
+            "padding": "max_length",
+            "truncation": True,
+            "max_length": self.text_len,
+            "add_special_tokens": True,
+            "return_attention_mask": True,
+            "return_tensors": "pt",
+        }
+
 
 @dataclass
 class T5Config(TextEncoderConfig):
-    arch_config: TextEncoderArchConfig = T5ArchConfig()
+    arch_config: TextEncoderArchConfig = field(default_factory=T5ArchConfig)
 
     prefix: str = "t5"

fastvideo/v1/configs/models/vaes/__init__.py

@@ -1,7 +1,9 @@
 from fastvideo.v1.configs.models.vaes.hunyuanvae import HunyuanVAEConfig
+from fastvideo.v1.configs.models.vaes.stepvideovae import StepVideoVAEConfig
 from fastvideo.v1.configs.models.vaes.wanvae import WanVAEConfig
 
 __all__ = [
     "HunyuanVAEConfig",
     "WanVAEConfig",
+    "StepVideoVAEConfig",
 ]

fastvideo/v1/configs/models/vaes/base.py

@@ -1,9 +1,10 @@
-from dataclasses import dataclass
-from typing import Union
+from dataclasses import dataclass, field
+from typing import Any, Union
 
 import torch
 
 from fastvideo.v1.configs.models.base import ArchConfig, ModelConfig
+from fastvideo.v1.utils import StoreBoolean
 
 
 @dataclass
@@ -16,7 +17,7 @@ class VAEArchConfig(ArchConfig):
 
 @dataclass
 class VAEConfig(ModelConfig):
-    arch_config: VAEArchConfig = VAEArchConfig()
+    arch_config: VAEArchConfig = field(default_factory=VAEArchConfig)
 
     # FastVideoVAE-specific parameters
     load_encoder: bool = True
@@ -33,6 +34,97 @@ class VAEConfig(ModelConfig):
     use_tiling: bool = True
     use_temporal_tiling: bool = True
     use_parallel_tiling: bool = True
+    use_temporal_scaling_frames: bool = True
 
     def __post_init__(self):
         self.blend_num_frames = self.tile_sample_min_num_frames - self.tile_sample_stride_num_frames
+
+    @staticmethod
+    def add_cli_args(parser: Any, prefix: str = "vae-config") -> Any:
+        """Add CLI arguments for VAEConfig fields"""
+        parser.add_argument(
+            f"--{prefix}.load-encoder",
+            action=StoreBoolean,
+            dest=f"{prefix.replace('-', '_')}.load_encoder",
+            default=VAEConfig.load_encoder,
+            help="Whether to load the VAE encoder",
+        )
+        parser.add_argument(
+            f"--{prefix}.load-decoder",
+            action=StoreBoolean,
+            dest=f"{prefix.replace('-', '_')}.load_decoder",
+            default=VAEConfig.load_decoder,
+            help="Whether to load the VAE decoder",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-min-height",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_min_height",
+            default=VAEConfig.tile_sample_min_height,
+            help="Minimum height for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-min-width",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_min_width",
+            default=VAEConfig.tile_sample_min_width,
+            help="Minimum width for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-min-num-frames",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_min_num_frames",
+            default=VAEConfig.tile_sample_min_num_frames,
+            help="Minimum number of frames for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-stride-height",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_stride_height",
+            default=VAEConfig.tile_sample_stride_height,
+            help="Stride height for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-stride-width",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_stride_width",
+            default=VAEConfig.tile_sample_stride_width,
+            help="Stride width for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.tile-sample-stride-num-frames",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.tile_sample_stride_num_frames",
+            default=VAEConfig.tile_sample_stride_num_frames,
+            help="Stride number of frames for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.blend-num-frames",
+            type=int,
+            dest=f"{prefix.replace('-', '_')}.blend_num_frames",
+            default=VAEConfig.blend_num_frames,
+            help="Number of frames to blend for VAE tile sampling",
+        )
+        parser.add_argument(
+            f"--{prefix}.use-tiling",
+            action=StoreBoolean,
+            dest=f"{prefix.replace('-', '_')}.use_tiling",
+            default=VAEConfig.use_tiling,
+            help="Whether to use tiling for VAE",
+        )
+        parser.add_argument(
+            f"--{prefix}.use-temporal-tiling",
+            action=StoreBoolean,
+            dest=f"{prefix.replace('-', '_')}.use_temporal_tiling",
+            default=VAEConfig.use_temporal_tiling,
+            help="Whether to use temporal tiling for VAE",
+        )
+        parser.add_argument(
+            f"--{prefix}.use-parallel-tiling",
+            action=StoreBoolean,
+            dest=f"{prefix.replace('-', '_')}.use_parallel_tiling",
+            default=VAEConfig.use_parallel_tiling,
+            help="Whether to use parallel tiling for VAE",
+        )
+
+        return parser

fastvideo/v1/configs/models/vaes/hunyuanvae.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Tuple
 
 from fastvideo.v1.configs.models.vaes.base import VAEArchConfig, VAEConfig
@@ -37,4 +37,4 @@ class HunyuanVAEArchConfig(VAEArchConfig):
 
 @dataclass
 class HunyuanVAEConfig(VAEConfig):
-    arch_config: VAEArchConfig = HunyuanVAEArchConfig()
+    arch_config: VAEArchConfig = field(default_factory=HunyuanVAEArchConfig)

fastvideo/v1/configs/models/vaes/stepvideovae.py

@@ -0,0 +1,28 @@
+from dataclasses import dataclass, field
+
+from fastvideo.v1.configs.models.vaes.base import VAEArchConfig, VAEConfig
+
+
+@dataclass
+class StepVideoVAEArchConfig(VAEArchConfig):
+    in_channels: int = 3
+    out_channels: int = 3
+    z_channels: int = 64
+    num_res_blocks: int = 2
+    version: int = 2
+    frame_len: int = 17
+    world_size: int = 1
+
+    spatial_compression_ratio: int = 16
+    temporal_compression_ratio: int = 8
+
+    scaling_factor: float = 1.0
+
+
+@dataclass
+class StepVideoVAEConfig(VAEConfig):
+    arch_config: VAEArchConfig = field(default_factory=StepVideoVAEArchConfig)
+    use_tiling: bool = False
+    use_temporal_tiling: bool = False
+    use_parallel_tiling: bool = False
+    use_temporal_scaling_frames: bool = False

fastvideo/v1/configs/models/vaes/wanvae.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Tuple
 
 import torch
@@ -63,7 +63,7 @@ class WanVAEArchConfig(VAEArchConfig):
 
 @dataclass
 class WanVAEConfig(VAEConfig):
-    arch_config: VAEArchConfig = WanVAEArchConfig()
+    arch_config: VAEArchConfig = field(default_factory=WanVAEArchConfig)
     use_feature_cache: bool = True
 
     use_tiling: bool = False

fastvideo/v1/configs/pipelines/__init__.py

@@ -4,11 +4,15 @@ from fastvideo.v1.configs.pipelines.hunyuan import (FastHunyuanConfig,
                                                     HunyuanConfig)
 from fastvideo.v1.configs.pipelines.registry import (
     get_pipeline_config_cls_for_name)
+from fastvideo.v1.configs.pipelines.stepvideo import StepVideoT2VConfig
 from fastvideo.v1.configs.pipelines.wan import (WanI2V480PConfig,
-                                                WanT2V480PConfig)
+                                                WanI2V720PConfig,
+                                                WanT2V480PConfig,
+                                                WanT2V720PConfig)
 
 __all__ = [
     "HunyuanConfig", "FastHunyuanConfig", "PipelineConfig",
     "SlidingTileAttnConfig", "WanT2V480PConfig", "WanI2V480PConfig",
+    "WanT2V720PConfig", "WanI2V720PConfig", "StepVideoT2VConfig",
     "get_pipeline_config_cls_for_name"
 ]

fastvideo/v1/configs/pipelines/base.py

@@ -1,15 +1,26 @@
 import json
-from dataclasses import asdict, dataclass, fields
-from typing import Any, Dict, Optional
+from dataclasses import asdict, dataclass, field, fields
+from typing import Any, Callable, Dict, Optional, Tuple, cast
+
+import torch
 
 from fastvideo.v1.configs.models import (DiTConfig, EncoderConfig, ModelConfig,
                                          VAEConfig)
+from fastvideo.v1.configs.models.encoders import BaseEncoderOutput
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.utils import shallow_asdict
 
 logger = init_logger(__name__)
 
 
+def preprocess_text(prompt: str) -> str:
+    return prompt
+
+
+def postprocess_text(output: BaseEncoderOutput) -> torch.tensor:
+    raise NotImplementedError
+
+
 @dataclass
 class PipelineConfig:
     """Base configuration for all pipeline architectures."""
@@ -26,17 +37,26 @@ class PipelineConfig:
     vae_precision: str = "fp16"
     vae_tiling: bool = True
     vae_sp: bool = True
-    vae_config: VAEConfig = VAEConfig()
+    vae_config: VAEConfig = field(default_factory=VAEConfig)
 
     # DiT configuration
-    dit_config: DiTConfig = DiTConfig()
+    dit_config: DiTConfig = field(default_factory=DiTConfig)
 
     # Text encoder configuration
-    text_encoder_precision: str = "fp16"
-    text_encoder_config: EncoderConfig = EncoderConfig()
+    text_encoder_precisions: Tuple[str, ...] = field(
+        default_factory=lambda: ("fp16", ))
+    text_encoder_configs: Tuple[EncoderConfig, ...] = field(
+        default_factory=lambda: (EncoderConfig(), ))
+    preprocess_text_funcs: Tuple[Callable[[str], str], ...] = field(
+        default_factory=lambda: (preprocess_text, ))
+    postprocess_text_funcs: Tuple[Callable[[BaseEncoderOutput], torch.tensor],
+                                  ...] = field(default_factory=lambda:
+                                               (postprocess_text, ))
 
     # STA (Spatial-Temporal Attention) parameters
     mask_strategy_file_path: Optional[str] = None
+
+    # Compilation
     enable_torch_compile: bool = False
 
     @classmethod
@@ -52,16 +72,32 @@ class PipelineConfig:
                 model_path)
             pipeline_config = cls()
 
-        return pipeline_config
+        return cast(PipelineConfig, pipeline_config)
 
     def dump_to_json(self, file_path: str):
         output_dict = shallow_asdict(self)
+        del_keys = []
         for key, value in output_dict.items():
             if isinstance(value, ModelConfig):
                 model_dict = asdict(value)
                 # Model Arch Config should be hidden away from the users
                 model_dict.pop("arch_config")
                 output_dict[key] = model_dict
+            elif isinstance(value, tuple) and all(
+                    isinstance(v, ModelConfig) for v in value):
+                model_dicts = []
+                for v in value:
+                    model_dict = asdict(v)
+                    # Model Arch Config should be hidden away from the users
+                    model_dict.pop("arch_config")
+                    model_dicts.append(model_dict)
+                output_dict[key] = model_dicts
+            elif isinstance(value, tuple) and all(callable(f) for f in value):
+                # Skip dumping functions
+                del_keys.append(key)
+
+        for key in del_keys:
+            output_dict.pop(key, None)
 
         with open(file_path, "w") as f:
             json.dump(output_dict, f, indent=2)
@@ -82,6 +118,14 @@ class PipelineConfig:
                 # If it's a nested ModelConfig, update it recursively
                 if isinstance(current_value, ModelConfig):
                     current_value.update_model_config(new_value)
+                elif isinstance(current_value, tuple) and all(
+                        isinstance(v, ModelConfig) for v in current_value):
+                    assert len(current_value) == len(
+                        new_value
+                    ), "Users shouldn't delete or add text encoder config objects in your json"
+                    for target_config, source_config in zip(
+                            current_value, new_value):
+                        target_config.update_model_config(source_config)
                 else:
                     setattr(self, key, new_value)
 

fastvideo/v1/configs/pipelines/hunyuan.py

@@ -1,11 +1,59 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
+from typing import Callable, Tuple, TypedDict
+
+import torch
 
 from fastvideo.v1.configs.models import DiTConfig, EncoderConfig, VAEConfig
 from fastvideo.v1.configs.models.dits import HunyuanVideoConfig
-from fastvideo.v1.configs.models.encoders import CLIPTextConfig, LlamaConfig
+from fastvideo.v1.configs.models.encoders import (BaseEncoderOutput,
+                                                  CLIPTextConfig, LlamaConfig)
 from fastvideo.v1.configs.models.vaes import HunyuanVAEConfig
 from fastvideo.v1.configs.pipelines.base import PipelineConfig
 
+PROMPT_TEMPLATE_ENCODE_VIDEO = (
+    "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
+    "1. The main content and theme of the video."
+    "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+    "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+    "4. background environment, light, style and atmosphere."
+    "5. camera angles, movements, and transitions used in the video:<|eot_id|>"
+    "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>")
+
+
+class PromptTemplate(TypedDict):
+    template: str
+    crop_start: int
+
+
+prompt_template_video: PromptTemplate = {
+    "template": PROMPT_TEMPLATE_ENCODE_VIDEO,
+    "crop_start": 95,
+}
+
+
+def llama_preprocess_text(prompt: str) -> str:
+    return prompt_template_video["template"].format(prompt)
+
+
+def llama_postprocess_text(outputs: BaseEncoderOutput) -> torch.tensor:
+    hidden_state_skip_layer = 2
+    assert outputs.hidden_states is not None
+    hidden_states: Tuple[torch.Tensor, ...] = outputs.hidden_states
+    last_hidden_state: torch.tensor = hidden_states[-(hidden_state_skip_layer +
+                                                      1)]
+    crop_start = prompt_template_video.get("crop_start", -1)
+    last_hidden_state = last_hidden_state[:, crop_start:]
+    return last_hidden_state
+
+
+def clip_preprocess_text(prompt: str) -> str:
+    return prompt
+
+
+def clip_postprocess_text(outputs: BaseEncoderOutput) -> torch.tensor:
+    pooler_output: torch.tensor = outputs.pooler_output
+    return pooler_output
+
 
 @dataclass
 class HunyuanConfig(PipelineConfig):
@@ -13,25 +61,31 @@ class HunyuanConfig(PipelineConfig):
 
     # HunyuanConfig-specific parameters with defaults
     # DiT
-    dit_config: DiTConfig = HunyuanVideoConfig()
+    dit_config: DiTConfig = field(default_factory=HunyuanVideoConfig)
     # VAE
-    vae_config: VAEConfig = HunyuanVAEConfig()
+    vae_config: VAEConfig = field(default_factory=HunyuanVAEConfig)
     # Denoising stage
     embedded_cfg_scale: int = 6
     flow_shift: int = 7
 
+    # Video parameters
+    use_cpu_offload: bool = True
+
     # Text encoding stage
-    text_encoder_config: EncoderConfig = LlamaConfig()
+    text_encoder_configs: Tuple[EncoderConfig, ...] = field(
+        default_factory=lambda: (LlamaConfig(), CLIPTextConfig()))
+    preprocess_text_funcs: Tuple[Callable[[str], str], ...] = field(
+        default_factory=lambda: (llama_preprocess_text, clip_preprocess_text))
+    postprocess_text_funcs: Tuple[
+        Callable[[BaseEncoderOutput], torch.tensor],
+        ...] = field(default_factory=lambda:
+                     (llama_postprocess_text, clip_postprocess_text))
 
     # Precision for each component
     precision: str = "bf16"
     vae_precision: str = "fp16"
-    text_encoder_precision: str = "fp16"
-
-    # HunyuanConfig-specific added parameters
-    # Secondary text encoder
-    text_encoder_config_2: EncoderConfig = CLIPTextConfig()
-    text_encoder_precision_2: str = "fp16"
+    text_encoder_precisions: Tuple[str, ...] = field(
+        default_factory=lambda: ("fp16", "fp16"))
 
     def __post_init__(self):
         self.vae_config.load_encoder = False

fastvideo/v1/configs/pipelines/registry.py

@@ -6,8 +6,11 @@ from typing import Callable, Dict, Optional, Type
 from fastvideo.v1.configs.pipelines.base import PipelineConfig
 from fastvideo.v1.configs.pipelines.hunyuan import (FastHunyuanConfig,
                                                     HunyuanConfig)
+from fastvideo.v1.configs.pipelines.stepvideo import StepVideoT2VConfig
 from fastvideo.v1.configs.pipelines.wan import (WanI2V480PConfig,
-                                                WanT2V480PConfig)
+                                                WanI2V720PConfig,
+                                                WanT2V480PConfig,
+                                                WanT2V720PConfig)
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.utils import (maybe_download_model_index,
                                 verify_model_config_and_directory)
@@ -19,7 +22,10 @@ WEIGHT_CONFIG_REGISTRY: Dict[str, Type[PipelineConfig]] = {
     "FastVideo/FastHunyuan-diffusers": FastHunyuanConfig,
     "hunyuanvideo-community/HunyuanVideo": HunyuanConfig,
     "Wan-AI/Wan2.1-T2V-1.3B-Diffusers": WanT2V480PConfig,
-    "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers": WanI2V480PConfig
+    "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers": WanI2V480PConfig,
+    "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers": WanI2V720PConfig,
+    "Wan-AI/Wan2.1-T2V-14B-Diffusers": WanT2V720PConfig,
+    "FastVideo/stepvideo-t2v-diffusers": StepVideoT2VConfig,
     # Add other specific weight variants
 }
 
@@ -28,6 +34,7 @@ PIPELINE_DETECTOR: Dict[str, Callable[[str], bool]] = {
     "hunyuan": lambda id: "hunyuan" in id.lower(),
     "wanpipeline": lambda id: "wanpipeline" in id.lower(),
     "wanimagetovideo": lambda id: "wanimagetovideo" in id.lower(),
+    "stepvideo": lambda id: "stepvideo" in id.lower(),
     # Add other pipeline architecture detectors
 }
 
@@ -38,6 +45,7 @@ PIPELINE_FALLBACK_CONFIG: Dict[str, Type[PipelineConfig]] = {
     "wanpipeline":
     WanT2V480PConfig,  # Base Wan config as fallback for any Wan variant
     "wanimagetovideo": WanI2V480PConfig,
+    "stepvideo": StepVideoT2VConfig
     # Other fallbacks by architecture
 }
 

fastvideo/v1/configs/pipelines/stepvideo.py

@@ -0,0 +1,32 @@
+from dataclasses import dataclass, field
+
+from fastvideo.v1.configs.models import DiTConfig, VAEConfig
+from fastvideo.v1.configs.models.dits import StepVideoConfig
+from fastvideo.v1.configs.models.vaes import StepVideoVAEConfig
+from fastvideo.v1.configs.pipelines.base import PipelineConfig
+
+
+@dataclass
+class StepVideoT2VConfig(PipelineConfig):
+    """Base configuration for StepVideo pipeline architecture."""
+
+    # WanConfig-specific parameters with defaults
+    # DiT
+    dit_config: DiTConfig = field(default_factory=StepVideoConfig)
+    # VAE
+    vae_config: VAEConfig = field(default_factory=StepVideoVAEConfig)
+    vae_tiling: bool = False
+    vae_sp: bool = False
+
+    # Video parameters
+    use_cpu_offload: bool = True
+
+    # Denoising stage
+    flow_shift: int = 13
+    timesteps_scale: bool = False
+    pos_magic: str = "超高清、HDR 视频、环境光、杜比全景声、画面稳定、流畅动作、逼真的细节、专业级构图、超现实主义、自然、生动、超细节、清晰。"
+    neg_magic: str = "画面暗、低分辨率、不良手、文本、缺少手指、多余的手指、裁剪、低质量、颗粒状、签名、水印、用户名、模糊。"
+
+    # Precision for each component
+    precision: str = "bf16"
+    vae_precision: str = "bf16"

fastvideo/v1/configs/pipelines/wan.py

@@ -1,21 +1,39 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
+from typing import Callable, Tuple
+
+import torch
 
 from fastvideo.v1.configs.models import DiTConfig, EncoderConfig, VAEConfig
 from fastvideo.v1.configs.models.dits import WanVideoConfig
-from fastvideo.v1.configs.models.encoders import CLIPVisionConfig, T5Config
+from fastvideo.v1.configs.models.encoders import (BaseEncoderOutput,
+                                                  CLIPVisionConfig, T5Config)
 from fastvideo.v1.configs.models.vaes import WanVAEConfig
 from fastvideo.v1.configs.pipelines.base import PipelineConfig
 
 
+def t5_postprocess_text(outputs: BaseEncoderOutput) -> torch.tensor:
+    mask: torch.tensor = outputs.attention_mask
+    hidden_state: torch.tensor = outputs.last_hidden_state
+    seq_lens = mask.gt(0).sum(dim=1).long()
+    assert torch.isnan(hidden_state).sum() == 0
+    prompt_embeds = [u[:v] for u, v in zip(hidden_state, seq_lens)]
+    prompt_embeds_tensor: torch.tensor = torch.stack([
+        torch.cat([u, u.new_zeros(512 - u.size(0), u.size(1))])
+        for u in prompt_embeds
+    ],
+                                                     dim=0)
+    return prompt_embeds_tensor
+
+
 @dataclass
 class WanT2V480PConfig(PipelineConfig):
     """Base configuration for Wan T2V 1.3B pipeline architecture."""
 
     # WanConfig-specific parameters with defaults
     # DiT
-    dit_config: DiTConfig = WanVideoConfig()
+    dit_config: DiTConfig = field(default_factory=WanVideoConfig)
     # VAE
-    vae_config: VAEConfig = WanVAEConfig()
+    vae_config: VAEConfig = field(default_factory=WanVAEConfig)
     vae_tiling: bool = False
     vae_sp: bool = False
 
@@ -26,12 +44,17 @@ class WanT2V480PConfig(PipelineConfig):
     flow_shift: int = 3
 
     # Text encoding stage
-    text_encoder_config: EncoderConfig = T5Config()
+    text_encoder_configs: Tuple[EncoderConfig, ...] = field(
+        default_factory=lambda: (T5Config(), ))
+    postprocess_text_funcs: Tuple[Callable[[BaseEncoderOutput], torch.tensor],
+                                  ...] = field(default_factory=lambda:
+                                               (t5_postprocess_text, ))
 
     # Precision for each component
     precision: str = "bf16"
     vae_precision: str = "fp16"
-    text_encoder_precision: str = "fp32"
+    text_encoder_precisions: Tuple[str, ...] = field(
+        default_factory=lambda: ("fp32", ))
 
     # WanConfig-specific added parameters
 
@@ -40,6 +63,16 @@ class WanT2V480PConfig(PipelineConfig):
         self.vae_config.load_decoder = True
 
 
+@dataclass
+class WanT2V720PConfig(WanT2V480PConfig):
+    """Base configuration for Wan T2V 14B 720P pipeline architecture."""
+
+    # WanConfig-specific parameters with defaults
+
+    # Denoising stage
+    flow_shift: int = 5
+
+
 @dataclass
 class WanI2V480PConfig(WanT2V480PConfig):
     """Base configuration for Wan I2V 14B 480P pipeline architecture."""
@@ -47,9 +80,20 @@ class WanI2V480PConfig(WanT2V480PConfig):
     # WanConfig-specific parameters with defaults
 
     # Precision for each component
-    image_encoder_config: EncoderConfig = CLIPVisionConfig()
+    image_encoder_config: EncoderConfig = field(
+        default_factory=CLIPVisionConfig)
     image_encoder_precision: str = "fp32"
 
     def __post_init__(self):
         self.vae_config.load_encoder = True
         self.vae_config.load_decoder = True
+
+
+@dataclass
+class WanI2V720PConfig(WanI2V480PConfig):
+    """Base configuration for Wan I2V 14B 720P pipeline architecture."""
+
+    # WanConfig-specific parameters with defaults
+
+    # Denoising stage
+    flow_shift: int = 5

fastvideo/v1/configs/quantization/__init__.py

@@ -1,3 +0,0 @@
-from fastvideo.v1.configs.quantization.base import QuantizationConfig
-
-__all__ = ["QuantizationConfig"]

fastvideo/v1/configs/quantization/base.py

@@ -1,6 +0,0 @@
-from dataclasses import dataclass
-
-
-@dataclass
-class QuantizationConfig:
-    pass

fastvideo/v1/configs/sample/base.py

@@ -8,6 +8,9 @@ logger = init_logger(__name__)
 
 @dataclass
 class SamplingParam:
+    """
+    Sampling parameters for video generation.
+    """
     # All fields below are copied from ForwardBatch
     data_type: str = "video"
 
@@ -26,6 +29,7 @@ class SamplingParam:
 
     # Original dimensions (before VAE scaling)
     num_frames: int = 125
+    num_frames_round_down: bool = False  # Whether to round down num_frames if it's not divisible by num_gpus
     height: int = 720
     width: int = 1280
     fps: int = 24
@@ -35,6 +39,9 @@ class SamplingParam:
     guidance_scale: float = 1.0
     guidance_rescale: float = 0.0
 
+    # TeaCache parameters
+    enable_teacache: bool = False
+
     # Misc
     save_video: bool = True
     return_frames: bool = False
@@ -70,3 +77,115 @@ class SamplingParam:
             sampling_param = cls()
 
         return sampling_param
+
+    @staticmethod
+    def add_cli_args(parser: Any) -> Any:
+        """Add CLI arguments for SamplingParam fields"""
+        parser.add_argument(
+            "--prompt",
+            type=str,
+            default=SamplingParam.prompt,
+            help="Text prompt for video generation",
+        )
+        parser.add_argument(
+            "--negative-prompt",
+            type=str,
+            default=SamplingParam.negative_prompt,
+            help="Negative text prompt for video generation",
+        )
+        parser.add_argument(
+            "--prompt-path",
+            type=str,
+            default=SamplingParam.prompt_path,
+            help="Path to a text file containing the prompt",
+        )
+        parser.add_argument(
+            "--output-path",
+            type=str,
+            default=SamplingParam.output_path,
+            help="Path to save the generated video",
+        )
+        parser.add_argument(
+            "--num-videos-per-prompt",
+            type=int,
+            default=SamplingParam.num_videos_per_prompt,
+            help="Number of videos to generate per prompt",
+        )
+        parser.add_argument(
+            "--seed",
+            type=int,
+            default=SamplingParam.seed,
+            help="Random seed for generation",
+        )
+        parser.add_argument(
+            "--num-frames",
+            type=int,
+            default=SamplingParam.num_frames,
+            help="Number of frames to generate",
+        )
+        parser.add_argument(
+            "--height",
+            type=int,
+            default=SamplingParam.height,
+            help="Height of generated video",
+        )
+        parser.add_argument(
+            "--width",
+            type=int,
+            default=SamplingParam.width,
+            help="Width of generated video",
+        )
+        parser.add_argument(
+            "--fps",
+            type=int,
+            default=SamplingParam.fps,
+            help="Frames per second for saved video",
+        )
+        parser.add_argument(
+            "--num-inference-steps",
+            type=int,
+            default=SamplingParam.num_inference_steps,
+            help="Number of denoising steps",
+        )
+        parser.add_argument(
+            "--guidance-scale",
+            type=float,
+            default=SamplingParam.guidance_scale,
+            help="Classifier-free guidance scale",
+        )
+        parser.add_argument(
+            "--guidance-rescale",
+            type=float,
+            default=SamplingParam.guidance_rescale,
+            help="Guidance rescale factor",
+        )
+        parser.add_argument(
+            "--save-video",
+            action="store_true",
+            default=SamplingParam.save_video,
+            help="Whether to save the video to disk",
+        )
+        parser.add_argument(
+            "--no-save-video",
+            action="store_false",
+            dest="save_video",
+            help="Don't save the video to disk",
+        )
+        parser.add_argument(
+            "--return-frames",
+            action="store_true",
+            default=SamplingParam.return_frames,
+            help="Whether to return the raw frames",
+        )
+        parser.add_argument(
+            "--image-path",
+            type=str,
+            default=SamplingParam.image_path,
+            help="Path to input image for image-to-video generation",
+        )
+        return parser
+
+
+@dataclass
+class CacheParams:
+    cache_type: str = "none"

fastvideo/v1/configs/sample/hunyuan.py

@@ -1,6 +1,7 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 from fastvideo.v1.configs.sample.base import SamplingParam
+from fastvideo.v1.configs.sample.teacache import TeaCacheParams
 
 
 @dataclass
@@ -14,6 +15,14 @@ class HunyuanSamplingParam(SamplingParam):
 
     guidance_scale: float = 1.0
 
+    teacache_params: TeaCacheParams = field(
+        default_factory=lambda: TeaCacheParams(
+            teacache_thresh=0.15,
+            coefficients=[
+                7.33226126e+02, -4.01131952e+02, 6.75869174e+01,
+                -3.14987800e+00, 9.61237896e-02
+            ]))
+
 
 @dataclass
 class FastHunyuanSamplingParam(HunyuanSamplingParam):

fastvideo/v1/configs/sample/registry.py

@@ -3,8 +3,11 @@ from typing import Any, Callable, Dict, Optional
 
 from fastvideo.v1.configs.sample.hunyuan import (FastHunyuanSamplingParam,
                                                  HunyuanSamplingParam)
-from fastvideo.v1.configs.sample.wan import (WanI2V480PSamplingParam,
-                                             WanT2V480PSamplingParam)
+from fastvideo.v1.configs.sample.stepvideo import StepVideoT2VSamplingParam
+from fastvideo.v1.configs.sample.wan import (WanI2V_14B_480P_SamplingParam,
+                                             WanI2V_14B_720P_SamplingParam,
+                                             WanT2V_1_3B_SamplingParam,
+                                             WanT2V_14B_SamplingParam)
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.utils import (maybe_download_model_index,
                                 verify_model_config_and_directory)
@@ -14,8 +17,11 @@ logger = init_logger(__name__)
 SAMPLING_PARAM_REGISTRY: Dict[str, Any] = {
     "FastVideo/FastHunyuan-diffusers": FastHunyuanSamplingParam,
     "hunyuanvideo-community/HunyuanVideo": HunyuanSamplingParam,
-    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers": WanT2V480PSamplingParam,
-    "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers": WanI2V480PSamplingParam
+    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers": WanT2V_1_3B_SamplingParam,
+    "Wan-AI/Wan2.1-T2V-14B-Diffusers": WanT2V_14B_SamplingParam,
+    "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers": WanI2V_14B_480P_SamplingParam,
+    "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers": WanI2V_14B_720P_SamplingParam,
+    "FastVideo/stepvideo-t2v-diffusers": StepVideoT2VSamplingParam,
     # Add other specific weight variants
 }
 
@@ -24,6 +30,7 @@ SAMPLING_PARAM_DETECTOR: Dict[str, Callable[[str], bool]] = {
     "hunyuan": lambda id: "hunyuan" in id.lower(),
     "wanpipeline": lambda id: "wanpipeline" in id.lower(),
     "wanimagetovideo": lambda id: "wanimagetovideo" in id.lower(),
+    "stepvideo": lambda id: "stepvideo" in id.lower(),
     # Add other pipeline architecture detectors
 }
 
@@ -32,8 +39,9 @@ SAMPLING_FALLBACK_PARAM: Dict[str, Any] = {
     "hunyuan":
     HunyuanSamplingParam,  # Base Hunyuan config as fallback for any Hunyuan variant
     "wanpipeline":
-    WanT2V480PSamplingParam,  # Base Wan config as fallback for any Wan variant
-    "wanimagetovideo": WanI2V480PSamplingParam,
+    WanT2V_1_3B_SamplingParam,  # Base Wan config as fallback for any Wan variant
+    "wanimagetovideo": WanI2V_14B_480P_SamplingParam,
+    "stepvideo": StepVideoT2VSamplingParam
     # Other fallbacks by architecture
 }
 

fastvideo/v1/configs/sample/stepvideo.py

@@ -0,0 +1,19 @@
+from dataclasses import dataclass
+
+from fastvideo.v1.configs.sample.base import SamplingParam
+
+
+@dataclass
+class StepVideoT2VSamplingParam(SamplingParam):
+    # Video parameters
+    height: int = 720
+    width: int = 1280
+    num_frames: int = 81
+
+    # Denoising stage
+    guidance_scale: float = 9.0
+    num_inference_steps: int = 50
+
+    # neg magic and pos magic
+    # pos_magic: str = "超高清、HDR 视频、环境光、杜比全景声、画面稳定、流畅动作、逼真的细节、专业级构图、超现实主义、自然、生动、超细节、清晰。"
+    # neg_magic: str = "画面暗、低分辨率、不良手、文本、缺少手指、多余的手指、裁剪、低质量、颗粒状、签名、水印、用户名、模糊。"

fastvideo/v1/configs/sample/teacache.py

@@ -0,0 +1,40 @@
+from dataclasses import dataclass, field
+
+from fastvideo.v1.configs.sample.base import CacheParams
+
+
+@dataclass
+class TeaCacheParams(CacheParams):
+    cache_type: str = "teacache"
+    teacache_thresh: float = 0.0
+    coefficients: list[float] = field(default_factory=list)
+
+
+@dataclass
+class WanTeaCacheParams(CacheParams):
+    # Unfortunately, TeaCache is very different for Wan than other models
+    cache_type: str = "teacache"
+    teacache_thresh: float = 0.0
+    use_ret_steps: bool = True
+    ret_steps_coeffs: list[float] = field(default_factory=list)
+    non_ret_steps_coeffs: list[float] = field(default_factory=list)
+
+    @property
+    def coefficients(self) -> list[float]:
+        if self.use_ret_steps:
+            return self.ret_steps_coeffs
+        else:
+            return self.non_ret_steps_coeffs
+
+    @property
+    def ret_steps(self) -> int:
+        if self.use_ret_steps:
+            return 5 * 2
+        else:
+            return 1 * 2
+
+    def get_cutoff_steps(self, num_inference_steps: int) -> int:
+        if self.use_ret_steps:
+            return num_inference_steps * 2
+        else:
+            return num_inference_steps * 2 - 2

fastvideo/v1/configs/sample/wan.py

@@ -1,10 +1,11 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 from fastvideo.v1.configs.sample.base import SamplingParam
+from fastvideo.v1.configs.sample.teacache import WanTeaCacheParams
 
 
 @dataclass
-class WanT2V480PSamplingParam(SamplingParam):
+class WanT2V_1_3B_SamplingParam(SamplingParam):
     # Video parameters
     height: int = 480
     width: int = 832
@@ -16,9 +17,79 @@ class WanT2V480PSamplingParam(SamplingParam):
     negative_prompt: str = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"
     num_inference_steps: int = 50
 
+    teacache_params: WanTeaCacheParams = field(
+        default_factory=lambda: WanTeaCacheParams(
+            teacache_thresh=0.08,
+            ret_steps_coeffs=[
+                -5.21862437e+04, 9.23041404e+03, -5.28275948e+02,
+                1.36987616e+01, -4.99875664e-02
+            ],
+            non_ret_steps_coeffs=[
+                2.39676752e+03, -1.31110545e+03, 2.01331979e+02,
+                -8.29855975e+00, 1.37887774e-01
+            ]))
+
 
 @dataclass
-class WanI2V480PSamplingParam(WanT2V480PSamplingParam):
+class WanT2V_14B_SamplingParam(SamplingParam):
+    # Video parameters
+    height: int = 720
+    width: int = 1280
+    num_frames: int = 81
+    fps: int = 16
+
+    # Denoising stage
+    guidance_scale: float = 5.0
+    negative_prompt: str = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"
+    num_inference_steps: int = 50
+
+    teacache_params: WanTeaCacheParams = field(
+        default_factory=lambda: WanTeaCacheParams(
+            teacache_thresh=0.20,
+            use_ret_steps=False,
+            ret_steps_coeffs=[
+                -3.03318725e+05, 4.90537029e+04, -2.65530556e+03,
+                5.87365115e+01, -3.15583525e-01
+            ],
+            non_ret_steps_coeffs=[
+                -5784.54975374, 5449.50911966, -1811.16591783, 256.27178429,
+                -13.02252404
+            ]))
+
+
+@dataclass
+class WanI2V_14B_480P_SamplingParam(WanT2V_1_3B_SamplingParam):
     # Denoising stage
     guidance_scale: float = 5.0
     num_inference_steps: int = 40
+
+    teacache_params: WanTeaCacheParams = field(
+        default_factory=lambda: WanTeaCacheParams(
+            teacache_thresh=0.26,
+            ret_steps_coeffs=[
+                -3.03318725e+05, 4.90537029e+04, -2.65530556e+03,
+                5.87365115e+01, -3.15583525e-01
+            ],
+            non_ret_steps_coeffs=[
+                -5784.54975374, 5449.50911966, -1811.16591783, 256.27178429,
+                -13.02252404
+            ]))
+
+
+@dataclass
+class WanI2V_14B_720P_SamplingParam(WanT2V_14B_SamplingParam):
+    # Denoising stage
+    guidance_scale: float = 5.0
+    num_inference_steps: int = 40
+
+    teacache_params: WanTeaCacheParams = field(
+        default_factory=lambda: WanTeaCacheParams(
+            teacache_thresh=0.3,
+            ret_steps_coeffs=[
+                -3.03318725e+05, 4.90537029e+04, -2.65530556e+03,
+                5.87365115e+01, -3.15583525e-01
+            ],
+            non_ret_steps_coeffs=[
+                -5784.54975374, 5449.50911966, -1811.16591783, 256.27178429,
+                -13.02252404
+            ]))

fastvideo/v1/configs/wan_1.3B_t2v_pipeline.json

@@ -26,12 +26,16 @@
     "prefix": "Wan",
     "quant_config": null
   },
-  "text_encoder_precision": "fp32",
-  "text_encoder_config": {
-    "prefix": "t5",
-    "quant_config": null,
-    "lora_config": null
-  },
+  "text_encoder_precisions": [
+    "fp32"
+  ],
+  "text_encoder_configs": [
+    {
+      "prefix": "t5",
+      "quant_config": null,
+      "lora_config": null
+    }
+  ],
   "mask_strategy_file_path": null,
   "enable_torch_compile": false
 }

fastvideo/v1/configs/wan_14B_i2v_480p_pipeline.json

@@ -26,12 +26,16 @@
     "prefix": "Wan",
     "quant_config": null
   },
-  "text_encoder_precision": "fp32",
-  "text_encoder_config": {
-    "prefix": "t5",
-    "quant_config": null,
-    "lora_config": null
-  },
+  "text_encoder_precisions": [
+    "fp32"
+  ],
+  "text_encoder_configs": [
+    {
+      "prefix": "t5",
+      "quant_config": null,
+      "lora_config": null
+    }
+  ],
   "mask_strategy_file_path": null,
   "enable_torch_compile": false,
   "image_encoder_config": {

fastvideo/v1/default_configs/v1_inference_hunyuan_config.yaml

@@ -1,16 +0,0 @@
-num_gpus: 4
-model_path: FastVideo/FastHunyuan-diffusers
-master_port: 29503
-sp_size: 4 
-tp_size: 4 
-height: 720 
-width: 1280 
-num_frames: 125 
-num_inference_steps: 6 
-guidance_scale: 1 
-embedded_cfg_scale: 6 
-flow_shift: 17 
-prompt_path: ./assets/prompt.txt 
-seed: 1024 
-output_path: outputs_video/ 
-vae-sp: True

fastvideo/v1/distributed/utils.py

@@ -94,14 +94,14 @@ class StatelessProcessGroup:
         key = f"send_to/{dst}/{self.send_dst_counter[dst]}"
         self.store.set(key, pickle.dumps(obj))
         self.send_dst_counter[dst] += 1
-        self.entries.append((key, time.time()))
+        self.entries.append((key, time.perf_counter()))
 
     def expire_data(self) -> None:
         """Expire data that is older than `data_expiration_seconds` seconds."""
         while self.entries:
             # check the oldest entry
             key, timestamp = self.entries[0]
-            if time.time() - timestamp > self.data_expiration_seconds:
+            if time.perf_counter() - timestamp > self.data_expiration_seconds:
                 self.store.delete_key(key)
                 self.entries.popleft()
             else:
@@ -125,7 +125,7 @@ class StatelessProcessGroup:
                    f"{self.broadcast_send_counter}")
             self.store.set(key, pickle.dumps(obj))
             self.broadcast_send_counter += 1
-            self.entries.append((key, time.time()))
+            self.entries.append((key, time.perf_counter()))
             return obj
         else:
             key = (f"broadcast_from/{src}/"

fastvideo/v1/entrypoints/cli/generate.py

@@ -2,13 +2,20 @@
 # adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/entrypoints/cli/serve.py
 
 import argparse
-from typing import List, cast
+import dataclasses
+import os
+from typing import Any, Dict, List, Optional, cast
 
-from fastvideo.v1.entrypoints.cli import utils
+from fastvideo import PipelineConfig, VideoGenerator
+from fastvideo.v1.configs.sample.base import SamplingParam
 from fastvideo.v1.entrypoints.cli.cli_types import CLISubcommand
+from fastvideo.v1.entrypoints.cli.utils import RaiseNotImplementedAction
 from fastvideo.v1.fastvideo_args import FastVideoArgs
+from fastvideo.v1.logger import init_logger
 from fastvideo.v1.utils import FlexibleArgumentParser
 
+logger = init_logger(__name__)
+
 
 class GenerateSubcommand(CLISubcommand):
     """The `generate` subcommand for the FastVideo CLI"""
@@ -16,45 +23,80 @@ class GenerateSubcommand(CLISubcommand):
     def __init__(self) -> None:
         self.name = "generate"
         super().__init__()
+        self.init_arg_names = self._get_init_arg_names()
+        self.generation_arg_names = self._get_generation_arg_names()
+
+    def _get_init_arg_names(self) -> List[str]:
+        """Get names of arguments for VideoGenerator initialization"""
+        return ["num_gpus", "tp_size", "sp_size", "model_path"]
+
+    def _get_generation_arg_names(self) -> List[str]:
+        """Get names of arguments for generate_video method"""
+        return [field.name for field in dataclasses.fields(SamplingParam)]
 
     def cmd(self, args: argparse.Namespace) -> None:
-        excluded_args = [
-            'subparser', 'config', 'num_gpus', 'master_port',
-            'dispatch_function'
-        ]
+        excluded_args = ['subparser', 'config', 'dispatch_function']
 
-        # Create a filtered dictionary of arguments
-        filtered_args = {
+        FastVideoArgs.from_cli_args(args)
+
+        provided_args = {}
+        for k, v in vars(args).items():
+            if (k not in excluded_args and v is not None
+                    and hasattr(args, '_provided') and k in args._provided):
+                provided_args[k] = v
+
+        if 'model_path' in vars(args) and args.model_path is not None:
+            provided_args['model_path'] = args.model_path
+
+        if 'prompt' in vars(args) and args.prompt is not None:
+            provided_args['prompt'] = args.prompt
+
+        merged_args = {**provided_args}
+
+        logger.info('CLI Args: %s', merged_args)
+
+        if 'model_path' not in merged_args or not merged_args['model_path']:
+            raise ValueError(
+                "model_path must be provided either in config file or via --model-path"
+            )
+
+        if 'prompt' not in merged_args or not merged_args['prompt']:
+            raise ValueError(
+                "prompt must be provided either in config file or via --prompt")
+
+        init_args = {
             k: v
-            for k, v in vars(args).items()
-            if k not in excluded_args and v is not None
+            for k, v in merged_args.items() if k in self.init_arg_names
+        }
+        generation_args = {
+            k: v
+            for k, v in merged_args.items() if k in self.generation_arg_names
         }
 
-        main_args = []
+        pipeline_config = PipelineConfig.from_pretrained(
+            merged_args['model_path'])
 
-        for key, value in filtered_args.items():
-            # Convert underscores to dashes in argument names
-            arg_name = f"--{key.replace('_', '-')}"
+        update_config_from_args(pipeline_config.dit_config, merged_args,
+                                "dit_config")
+        update_config_from_args(pipeline_config.vae_config, merged_args,
+                                "vae_config")
+        update_config_from_args(pipeline_config, merged_args)
 
-            # Handle boolean flags
-            if isinstance(value, bool):
-                if value:
-                    main_args.append(arg_name)
-            else:
-                main_args.append(arg_name)
-                main_args.append(str(value))
+        model_path = init_args.pop('model_path')
+        prompt = generation_args.pop('prompt')
 
-        utils.launch_distributed(args.num_gpus,
-                                 main_args,
-                                 master_port=args.master_port)
+        generator = VideoGenerator.from_pretrained(
+            model_path=model_path, **init_args, pipeline_config=pipeline_config)
+
+        generator.generate_video(prompt=prompt, **generation_args)
 
     def validate(self, args: argparse.Namespace) -> None:
+        """Validate the arguments for this command"""
         if args.num_gpus is not None and args.num_gpus <= 0:
             raise ValueError("Number of gpus must be positive")
 
-        if args.master_port is not None and (args.master_port < 1024
-                                             or args.master_port > 65535):
-            raise ValueError("Master port must be between 1024 and 65535")
+        if args.config and not os.path.exists(args.config):
+            raise ValueError(f"Config file not found: {args.config}")
 
     def subparser_init(
             self,
@@ -63,7 +105,7 @@ class GenerateSubcommand(CLISubcommand):
             "generate",
             help="Run inference on a model",
             usage=
-            "fastvideo generate --model-path MODEL_PATH_OR_ID --prompt PROMPT [OPTIONS]"
+            "fastvideo generate (--model-path MODEL_PATH_OR_ID --prompt PROMPT) | --config CONFIG_FILE [OPTIONS]"
         )
 
         generate_parser.add_argument(
@@ -71,17 +113,53 @@ class GenerateSubcommand(CLISubcommand):
             type=str,
             default='',
             required=False,
-            help="Read CLI options from a config YAML file.")
-
-        generate_parser.add_argument("--master-port",
-                                     type=int,
-                                     default=None,
-                                     help="Port for the master process")
+            help=
+            "Read CLI options from a config JSON or YAML file. If provided, --model-path and --prompt are optional."
+        )
 
         generate_parser = FastVideoArgs.add_cli_args(generate_parser)
+        generate_parser = SamplingParam.add_cli_args(generate_parser)
+
+        generate_parser.add_argument(
+            "--text-encoder-configs",
+            action=RaiseNotImplementedAction,
+            help=
+            "JSON array of text encoder configurations (NOT YET IMPLEMENTED)",
+        )
 
         return cast(FlexibleArgumentParser, generate_parser)
 
 
 def cmd_init() -> List[CLISubcommand]:
     return [GenerateSubcommand()]
+
+
+def update_config_from_args(config: Any,
+                            args_dict: Dict[str, Any],
+                            prefix: Optional[str] = None) -> None:
+    """
+    Update configuration object from arguments dictionary.
+    
+    Args:
+        config: The configuration object to update
+        args_dict: Dictionary containing arguments
+        prefix: Prefix for the configuration parameters in the args_dict.
+               If None, assumes direct attribute mapping without prefix.
+    """
+    # Handle top-level attributes (no prefix)
+    if prefix is None:
+        for key, value in args_dict.items():
+            if hasattr(config, key) and value is not None:
+                if key == "text_encoder_precisions" and isinstance(value, list):
+                    setattr(config, key, tuple(value))
+                else:
+                    setattr(config, key, value)
+        return
+
+    # Handle nested attributes with prefix
+    prefix_with_dot = f"{prefix}."
+    for key, value in args_dict.items():
+        if key.startswith(prefix_with_dot) and value is not None:
+            attr_name = key[len(prefix_with_dot):]
+            if hasattr(config, attr_name):
+                setattr(config, attr_name, value)

fastvideo/v1/entrypoints/cli/utils.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 
+import argparse
 import os
 import subprocess
 import sys
@@ -10,6 +11,13 @@ from fastvideo.v1.logger import init_logger
 logger = init_logger(__name__)
 
 
+class RaiseNotImplementedAction(argparse.Action):
+
+    def __call__(self, parser, namespace, values, option_string=None):
+        raise NotImplementedError(
+            f"The {option_string} option is not yet implemented")
+
+
 def launch_distributed(num_gpus: int,
                        args: List[str],
                        master_port: Optional[int] = None) -> int:

fastvideo/v1/entrypoints/video_generator.py

@@ -6,9 +6,10 @@ This module provides a consolidated interface for generating videos using
 diffusion models.
 """
 
+import gc
+import math
 import os
 import time
-from dataclasses import asdict
 from typing import Any, Dict, List, Optional, Union
 
 import imageio
@@ -72,7 +73,6 @@ class VideoGenerator:
 
         Priority level: Default pipeline config < User's pipeline config < User's kwargs
         """
-
         config = None
         # 1. If users provide a pipeline config, it will override the default pipeline config
         if isinstance(pipeline_config, PipelineConfig):
@@ -181,12 +181,46 @@ class VideoGenerator:
                 f"height={sampling_param.height}, width={sampling_param.width}, "
                 f"num_frames={sampling_param.num_frames}")
 
-        if (
-                sampling_param.num_frames - 1
-        ) % fastvideo_args.vae_config.arch_config.temporal_compression_ratio != 0:
-            raise ValueError(
-                f"num_frames-1 must be a multiple of {fastvideo_args.vae_config.arch_config.temporal_compression_ratio}, got {sampling_param.num_frames}"
-            )
+        temporal_scale_factor = fastvideo_args.vae_config.arch_config.temporal_compression_ratio
+        num_frames = sampling_param.num_frames
+        num_gpus = fastvideo_args.num_gpus
+        use_temporal_scaling_frames = fastvideo_args.vae_config.use_temporal_scaling_frames
+
+        # Adjust number of frames based on number of GPUs
+        if use_temporal_scaling_frames:
+            orig_latent_num_frames = (num_frames -
+                                      1) // temporal_scale_factor + 1
+        else:  # stepvideo only
+            orig_latent_num_frames = sampling_param.num_frames // 17 * 3
+
+        if orig_latent_num_frames % fastvideo_args.num_gpus != 0:
+            # Adjust latent frames to be divisible by number of GPUs
+            if sampling_param.num_frames_round_down:
+                # Ensure we have at least 1 batch per GPU
+                new_latent_num_frames = max(
+                    1, (orig_latent_num_frames // num_gpus)) * num_gpus
+            else:
+                new_latent_num_frames = math.ceil(
+                    orig_latent_num_frames / num_gpus) * num_gpus
+
+            if use_temporal_scaling_frames:
+                # Convert back to number of frames, ensuring num_frames-1 is a multiple of temporal_scale_factor
+                new_num_frames = (new_latent_num_frames -
+                                  1) * temporal_scale_factor + 1
+            else:  # stepvideo only
+                # Find the least common multiple of 3 and num_gpus
+                divisor = math.lcm(3, num_gpus)
+                # Round up to the nearest multiple of this LCM
+                new_latent_num_frames = (
+                    (new_latent_num_frames + divisor - 1) // divisor) * divisor
+                # Convert back to actual frames using the StepVideo formula
+                new_num_frames = new_latent_num_frames // 3 * 17
+
+            logger.info(
+                "Adjusting number of frames from %s to %s based on number of GPUs (%s)",
+                sampling_param.num_frames, new_num_frames,
+                fastvideo_args.num_gpus)
+            sampling_param.num_frames = new_num_frames
 
         # Calculate sizes
         target_height = align_to(sampling_param.height, 16)
@@ -210,23 +244,26 @@ class VideoGenerator:
               guidance_scale: {sampling_param.guidance_scale}
                     n_tokens: {n_tokens}
                   flow_shift: {fastvideo_args.flow_shift}
-     embedded_guidance_scale: {fastvideo_args.embedded_cfg_scale}"""
+     embedded_guidance_scale: {fastvideo_args.embedded_cfg_scale}
+                  save_video: {sampling_param.save_video}
+                  output_path: {sampling_param.output_path}
+        """ # type: ignore[attr-defined]
         logger.info(debug_str)
 
         # Prepare batch
         batch = ForwardBatch(
-            **asdict(sampling_param),
+            **shallow_asdict(sampling_param),
             eta=0.0,
             n_tokens=n_tokens,
             extra={},
         )
 
         # Run inference
-        start_time = time.time()
+        start_time = time.perf_counter()
         output_batch = self.executor.execute_forward(batch, fastvideo_args)
         samples = output_batch
 
-        gen_time = time.time() - start_time
+        gen_time = time.perf_counter() - start_time
         logger.info("Generated successfully in %.2f seconds", gen_time)
 
         # Process outputs
@@ -243,7 +280,7 @@ class VideoGenerator:
             if save_path:
                 os.makedirs(os.path.dirname(save_path), exist_ok=True)
                 video_path = os.path.join(save_path, f"{prompt[:100]}.mp4")
-                imageio.mimsave(video_path, frames, fps=batch.fps)
+                imageio.mimsave(video_path, frames, fps=batch.fps, format="mp4")
                 logger.info("Saved video to %s", video_path)
             else:
                 logger.warning("No output path provided, video not saved")
@@ -257,3 +294,12 @@ class VideoGenerator:
                 "size": (target_height, target_width, batch.num_frames),
                 "generation_time": gen_time
             }
+
+    def shutdown(self):
+        """
+        Shutdown the video generator.
+        """
+        self.executor.shutdown()
+        del self.executor
+        gc.collect()
+        torch.cuda.empty_cache()

fastvideo/v1/examples/inference/basic/README.md

@@ -1,28 +0,0 @@
-# Basic Video Generation Tutorial
-The `VideoGenerator` class provides the primary Python interface for doing offline video generation, which is interacting with a diffusion pipeline without using a separate inference api server.
-
-
-## Usage
-The first script in this example shows the most basic usage of FastVideo. If you are new to Python and FastVideo, you should start here.
-
-```bash
-python fastvideo/v1/examples/inference/basic/basic.py
-```
-
-## Basic Walkthrough
-
-All you need to generate videos using multi-gpus from state-of-the-art diffusion pipelines is the following few lines! 
-
-```python
-from fastvideo import VideoGenerator
-
-generator = VideoGenerator.from_pretrained(
-    "FastVideo/FastHunyuan-Diffusers",
-    num_gpus=2,
-)
-
-prompt = "A beautiful woman in a red dress walking down a street"
-video = generator.generate_video(prompt)
-```
-
-More to come! These examples and APIs are still under construction!

fastvideo/v1/examples/inference/basic/basic.py

@@ -1,30 +0,0 @@
-from fastvideo import VideoGenerator
-# from fastvideo.v1.configs.sample import SamplingParam
-
-def main():
-    # FastVideo will automatically use the optimal default arguments for the
-    # model. 
-    # If a local path is provided, FastVideo will make a best effort
-    # attempt to identify the optimal arguments.
-    generator = VideoGenerator.from_pretrained(
-        "FastVideo/FastHunyuan-Diffusers",
-        # if num_gpus > 1, FastVideo will automatically handle distributed setup
-        num_gpus=4,
-    )
-
-    # sampling_param = SamplingParam.from_pretrained("/workspace/data/Wan-AI/Wan2.1-I2V-14B-480P-Diffusers")
-    # sampling_param.num_frames = 45
-    # sampling_param.image_path = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
-    # Generate videos with the same simple API, regardless of GPU count
-    prompt = "A beautiful woman in a red dress walking down a street"
-    video = generator.generate_video(prompt)
-    # video = generator.generate_video(prompt, sampling_param=sampling_param, output_path="wan_t2v_videos/")
-
-    # Generate another video with a different prompt, without reloading the
-    # model!
-    prompt2 = "A beautiful woman in a blue dress walking down a street"
-    video2 = generator.generate_video(prompt2)
-
-
-if __name__ == "__main__":
-    main()

fastvideo/v1/examples/inference/gradio/gradio_demo.py

@@ -1,141 +0,0 @@
-import os
-import gradio as gr
-import torch
-
-from fastvideo.v1.fastvideo_args import FastVideoArgs
-from fastvideo import VideoGenerator
-
-
-if __name__ == "__main__":
-    args = FastVideoArgs(model_path="FastVideo/FastHunyuan-Diffusers", num_gpus=2)
-
-    generator = VideoGenerator.from_pretrained(
-        model_path=args.model_path,
-        num_gpus=args.num_gpus
-    )
-
-    def generate_video(
-        prompt,
-        negative_prompt,
-        use_negative_prompt,
-        seed,
-        guidance_scale,
-        num_frames,
-        height,
-        width,
-        num_inference_steps,
-        randomize_seed=False,
-    ):
-        if randomize_seed:
-            seed = torch.randint(0, 1000000, (1, )).item()
-
-        if not use_negative_prompt:
-            negative_prompt = None
-
-        generator.generate_video(
-            prompt=prompt,
-            negative_prompt=negative_prompt,
-            num_inference_steps=num_inference_steps,
-            num_frames=num_frames,
-            height=height,
-            width=width,
-            guidance_scale=guidance_scale,
-            seed=seed
-        )
-
-        output_path = os.path.join(args.output_path, f"{prompt[:100]}.mp4")
-
-        return output_path, seed
-
-    examples = [
-    "A hand enters the frame, pulling a sheet of plastic wrap over three balls of dough placed on a wooden surface. The plastic wrap is stretched to cover the dough more securely. The hand adjusts the wrap, ensuring that it is tight and smooth over the dough. The scene focuses on the hand’s movements as it secures the edges of the plastic wrap. No new objects appear, and the camera remains stationary, focusing on the action of covering the dough.",
-    "A vintage train snakes through the mountains, its plume of white steam rising dramatically against the jagged peaks. The cars glint in the late afternoon sun, their deep crimson and gold accents lending a touch of elegance. The tracks carve a precarious path along the cliffside, revealing glimpses of a roaring river far below. Inside, passengers peer out the large windows, their faces lit with awe as the landscape unfolds.",
-    "A crowded rooftop bar buzzes with energy, the city skyline twinkling like a field of stars in the background. Strings of fairy lights hang above, casting a warm, golden glow over the scene. Groups of people gather around high tables, their laughter blending with the soft rhythm of live jazz. The aroma of freshly mixed cocktails and charred appetizers wafts through the air, mingling with the cool night breeze.",
-    ]
-
-    with gr.Blocks() as demo:
-        gr.Markdown("# FastVideo Inference Demo")
-
-        with gr.Group():
-            with gr.Row():
-                prompt = gr.Text(
-                    label="Prompt",
-                    show_label=False,
-                    max_lines=1,
-                    placeholder="Enter your prompt",
-                    container=False,
-                )
-                run_button = gr.Button("Run", scale=0)
-            result = gr.Video(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced options", open=False):
-            with gr.Group():
-                with gr.Row():
-                    height = gr.Slider(
-                        label="Height",
-                        minimum=256,
-                        maximum=1024,
-                        step=32,
-                        value=args.height,
-                    )
-                    width = gr.Slider(label="Width", minimum=256, maximum=1024, step=32, value=args.width)
-
-                with gr.Row():
-                    num_frames = gr.Slider(
-                        label="Number of Frames",
-                        minimum=21,
-                        maximum=163,
-                        value=45,
-                    )
-                    guidance_scale = gr.Slider(
-                        label="Guidance Scale",
-                        minimum=1,
-                        maximum=12,
-                        value=args.guidance_scale,
-                    )
-                    num_inference_steps = gr.Slider(
-                        label="Inference Steps",
-                        minimum=4,
-                        maximum=100,
-                        value=6,
-                    )
-
-                with gr.Row():
-                    use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=False)
-                negative_prompt = gr.Text(
-                    label="Negative prompt",
-                    max_lines=1,
-                    placeholder="Enter a negative prompt",
-                    visible=False,
-                )
-
-                seed = gr.Slider(label="Seed", minimum=0, maximum=1000000, step=1, value=args.seed)
-                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-                seed_output = gr.Number(label="Used Seed")
-
-        gr.Examples(examples=examples, inputs=prompt)
-
-        use_negative_prompt.change(
-            fn=lambda x: gr.update(visible=x),
-            inputs=use_negative_prompt,
-            outputs=negative_prompt,
-        )
-
-        run_button.click(
-            fn=generate_video,
-            inputs=[
-                prompt,
-                negative_prompt,
-                use_negative_prompt,
-                seed,
-                guidance_scale,
-                num_frames,
-                height,
-                width,
-                num_inference_steps,
-                randomize_seed,
-            ],
-            outputs=[result, seed_output],
-        )
-
-    demo.queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)

fastvideo/v1/fastvideo_args.py

@@ -5,20 +5,32 @@
 import argparse
 import dataclasses
 from contextlib import contextmanager
-from typing import List, Optional
+from dataclasses import field
+from typing import Any, Callable, List, Optional, Tuple
 
 from fastvideo.v1.configs.models import DiTConfig, EncoderConfig, VAEConfig
 from fastvideo.v1.logger import init_logger
-from fastvideo.v1.utils import FlexibleArgumentParser
+from fastvideo.v1.utils import FlexibleArgumentParser, StoreBoolean
 
 logger = init_logger(__name__)
 
 
+def preprocess_text(prompt: str) -> str:
+    return prompt
+
+
+def postprocess_text(output: Any) -> Any:
+    raise NotImplementedError
+
+
 @dataclasses.dataclass
 class FastVideoArgs:
     # Model and path configuration
     model_path: str
 
+    # Cache strategy
+    cache_strategy: str = "none"
+
     # Distributed executor backend
     distributed_executor_backend: str = "mp"
 
@@ -41,7 +53,7 @@ class FastVideoArgs:
     output_type: str = "pil"
 
     # DiT configuration
-    dit_config: DiTConfig = DiTConfig()
+    dit_config: DiTConfig = field(default_factory=DiTConfig)
     precision: str = "bf16"
 
     # VAE configuration
@@ -49,19 +61,25 @@ class FastVideoArgs:
     vae_tiling: bool = True  # Might change in between forward passes
     vae_sp: bool = False  # Might change in between forward passes
     # vae_scale_factor: Optional[int] = None # Deprecated
-    vae_config: VAEConfig = VAEConfig()
+    vae_config: VAEConfig = field(default_factory=VAEConfig)
 
     # Image encoder configuration
     image_encoder_precision: str = "fp32"
-    image_encoder_config: EncoderConfig = EncoderConfig()
+    image_encoder_config: EncoderConfig = field(default_factory=EncoderConfig)
 
     # Text encoder configuration
-    text_encoder_precision: str = "fp16"
-    text_encoder_config: EncoderConfig = EncoderConfig()
-
-    # Secondary text encoder
-    text_encoder_config_2: EncoderConfig = EncoderConfig()
-    text_encoder_precision_2: str = "fp16"
+    DEFAULT_TEXT_ENCODER_PRECISIONS = (
+        "fp16",
+        "fp16",
+    )
+    text_encoder_precisions: Tuple[str, ...] = field(
+        default_factory=lambda: FastVideoArgs.DEFAULT_TEXT_ENCODER_PRECISIONS)
+    text_encoder_configs: Tuple[EncoderConfig, ...] = field(
+        default_factory=lambda: (EncoderConfig(), ))
+    preprocess_text_funcs: Tuple[Callable[[str], str], ...] = field(
+        default_factory=lambda: (preprocess_text, ))
+    postprocess_text_funcs: Tuple[Callable[[Any], Any], ...] = field(
+        default_factory=lambda: (postprocess_text, ))
 
     # STA (Spatial-Temporal Attention) parameters
     mask_strategy_file_path: Optional[str] = None
@@ -70,6 +88,11 @@ class FastVideoArgs:
     use_cpu_offload: bool = False
     disable_autocast: bool = False
 
+    # StepVideo specific parameters
+    pos_magic: Optional[str] = None
+    neg_magic: Optional[str] = None
+    timesteps_scale: Optional[bool] = None
+
     # Logging
     log_level: str = "info"
 
@@ -86,7 +109,6 @@ class FastVideoArgs:
         parser.add_argument(
             "--model-path",
             type=str,
-            required=True,
             help=
             "The path of the model weights. This can be a local folder or a Hugging Face repo ID.",
         )
@@ -113,7 +135,7 @@ class FastVideoArgs:
         # HuggingFace specific parameters
         parser.add_argument(
             "--trust-remote-code",
-            action="store_true",
+            action=StoreBoolean,
             default=FastVideoArgs.trust_remote_code,
             help="Trust remote code when loading HuggingFace models",
         )
@@ -192,22 +214,23 @@ class FastVideoArgs:
         )
         parser.add_argument(
             "--vae-tiling",
-            action="store_true",
+            action=StoreBoolean,
             default=FastVideoArgs.vae_tiling,
             help="Enable VAE tiling",
         )
         parser.add_argument(
             "--vae-sp",
-            action="store_true",
+            action=StoreBoolean,
             help="Enable VAE spatial parallelism",
         )
 
         parser.add_argument(
-            "--text-encoder-precision",
+            "--text-encoder-precisions",
+            nargs="+",
             type=str,
-            default=FastVideoArgs.text_encoder_precision,
+            default=FastVideoArgs.DEFAULT_TEXT_ENCODER_PRECISIONS,
             choices=["fp32", "fp16", "bf16"],
-            help="Precision for text encoder",
+            help="Precision for each text encoder",
         )
 
         # Image encoder config
@@ -219,16 +242,6 @@ class FastVideoArgs:
             help="Precision for image encoder",
         )
 
-        # Secondary text encoder
-
-        parser.add_argument(
-            "--text-encoder-precision-2",
-            type=str,
-            default=FastVideoArgs.text_encoder_precision_2,
-            choices=["fp32", "fp16", "bf16"],
-            help="Precision for secondary text encoder",
-        )
-
         # STA (Spatial-Temporal Attention) parameters
         parser.add_argument(
             "--mask-strategy-file-path",
@@ -237,23 +250,42 @@ class FastVideoArgs:
         )
         parser.add_argument(
             "--enable-torch-compile",
-            action="store_true",
+            action=StoreBoolean,
             help=
             "Use torch.compile for speeding up STA inference without teacache",
         )
 
         parser.add_argument(
             "--use-cpu-offload",
-            action="store_true",
+            action=StoreBoolean,
             help="Use CPU offload for the model load",
         )
         parser.add_argument(
             "--disable-autocast",
-            action="store_true",
+            action=StoreBoolean,
             help=
             "Disable autocast for denoising loop and vae decoding in pipeline sampling",
         )
 
+        parser.add_argument(
+            "--pos_magic",
+            type=str,
+            default=FastVideoArgs.pos_magic,
+            help="Positive magic prompt for sampling",
+        )
+        parser.add_argument(
+            "--neg_magic",
+            type=str,
+            default=FastVideoArgs.neg_magic,
+            help="Negative magic prompt for sampling",
+        )
+        parser.add_argument(
+            "--timesteps_scale",
+            type=bool,
+            default=FastVideoArgs.timesteps_scale,
+            help="Bool for applying scheduler scale in set_timesteps",
+        )
+
         # Logging
         parser.add_argument(
             "--log-level",
@@ -262,6 +294,14 @@ class FastVideoArgs:
             help="The logging level of all loggers.",
         )
 
+        # Add VAE configuration arguments
+        from fastvideo.v1.configs.models.vaes.base import VAEConfig
+        VAEConfig.add_cli_args(parser)
+
+        # Add DiT configuration arguments
+        from fastvideo.v1.configs.models.dits.base import DiTConfig
+        DiTConfig.add_cli_args(parser)
+
         return parser
 
     @classmethod
@@ -311,6 +351,27 @@ class FastVideoArgs:
                 "Currently enabling vae_sp requires enabling vae_tiling, please set --vae-tiling to True."
             )
 
+        if len(self.text_encoder_configs) != len(self.text_encoder_precisions):
+            raise ValueError(
+                f"Length of text encoder configs ({len(self.text_encoder_configs)}) must be equal to length of text encoder precisions ({len(self.text_encoder_precisions)})"
+            )
+
+        if len(self.text_encoder_configs) != len(self.preprocess_text_funcs):
+            raise ValueError(
+                f"Length of text encoder configs ({len(self.text_encoder_configs)}) must be equal to length of text preprocessing functions ({len(self.preprocess_text_funcs)})"
+            )
+
+        if len(self.preprocess_text_funcs) != len(self.postprocess_text_funcs):
+            raise ValueError(
+                f"Length of text postprocess functions ({len(self.postprocess_text_funcs)}) must be equal to length of text preprocessing functions ({len(self.preprocess_text_funcs)})"
+            )
+
+        if self.enable_torch_compile and self.num_gpus > 1:
+            logger.warning(
+                "Currently torch compile does not work with multi-gpu. Setting enable_torch_compile to False"
+            )
+            self.enable_torch_compile = False
+
 
 _current_fastvideo_args = None
 

fastvideo/v1/forward_context.py

@@ -11,6 +11,7 @@ import torch
 
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
+from fastvideo.v1.pipelines.pipeline_batch_info import ForwardBatch
 
 if TYPE_CHECKING:
     from fastvideo.v1.attention import AttentionMetadata
@@ -30,11 +31,13 @@ batchsize_forward_time: defaultdict = defaultdict(list)
 #
 @dataclass
 class ForwardContext:
+    current_timestep: int
     # TODO(will): check this arg
     # copy from vllm_config.compilation_config.static_forward_context
     # attn_layers: Dict[str, Any]
     # TODO: extend to support per-layer dynamic forward context
     attn_metadata: "AttentionMetadata"  # set dynamically for each forward pass
+    forward_batch: Optional[ForwardBatch] = None
 
 
 _forward_context: Optional[ForwardContext] = None
@@ -52,6 +55,7 @@ def get_forward_context() -> ForwardContext:
 @contextmanager
 def set_forward_context(current_timestep,
                         attn_metadata,
+                        forward_batch: Optional[ForwardBatch] = None,
                         fastvideo_args: Optional[FastVideoArgs] = None):
     """A context manager that stores the current forward context,
     can be attention metadata, etc.
@@ -63,7 +67,9 @@ def set_forward_context(current_timestep,
         forward_start_time = time.perf_counter()
     global _forward_context
     prev_context = _forward_context
-    _forward_context = ForwardContext(attn_metadata=attn_metadata)
+    _forward_context = ForwardContext(current_timestep=current_timestep,
+                                      attn_metadata=attn_metadata,
+                                      forward_batch=forward_batch)
     try:
         yield
     finally:
@@ -76,10 +82,6 @@ def set_forward_context(current_timestep,
             else:
                 # for v1 attention backends
                 batchsize = attn_metadata.num_input_tokens
-            # we use synchronous scheduling right now,
-            # adding a sync point here should not affect
-            # scheduling of the next batch
-            torch.cuda.synchronize()
             now = time.perf_counter()
             # time measurement is in milliseconds
             batchsize_forward_time[batchsize].append(

fastvideo/v1/inference_engine.py

@@ -191,7 +191,7 @@ class InferenceEngine:
         # ========================================================================
         # Pipeline inference
         # ========================================================================
-        start_time = time.time()
+        start_time = time.perf_counter()
         samples = self.pipeline.forward(
             batch=batch,
             fastvideo_args=fastvideo_args,
@@ -201,7 +201,7 @@ class InferenceEngine:
         out_dict["samples"] = samples
         out_dict["prompts"] = prompt
 
-        gen_time = time.time() - start_time
+        gen_time = time.perf_counter() - start_time
         logger.info("Success, time: %s", gen_time)
 
         return out_dict

fastvideo/v1/layers/activation.py

@@ -9,7 +9,6 @@ import torch.nn.functional as F
 
 # TODO (will): remove this dependency
 from fastvideo.v1.layers.custom_op import CustomOp
-from fastvideo.v1.platforms import current_platform
 
 
 @CustomOp.register("silu_and_mul")
@@ -25,21 +24,12 @@ class SiluAndMul(CustomOp):
 
     def __init__(self) -> None:
         super().__init__()
-        if current_platform.is_cuda_alike() or current_platform.is_cpu():
-            self.op = torch.ops._C.silu_and_mul
 
     def forward_native(self, x: torch.Tensor) -> torch.Tensor:
         """PyTorch-native implementation equivalent to forward()."""
         d = x.shape[-1] // 2
         return F.silu(x[..., :d]) * x[..., d:]
 
-    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
-        d = x.shape[-1] // 2
-        output_shape = (x.shape[:-1] + (d, ))
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        self.op(out, x)
-        return out
-
 
 @CustomOp.register("gelu_and_mul")
 class GeluAndMul(CustomOp):
@@ -57,24 +47,12 @@ class GeluAndMul(CustomOp):
         self.approximate = approximate
         if approximate not in ("none", "tanh"):
             raise ValueError(f"Unknown approximate mode: {approximate}")
-        if current_platform.is_cuda_alike() or current_platform.is_cpu():
-            if approximate == "none":
-                self.op = torch.ops._C.gelu_and_mul
-            elif approximate == "tanh":
-                self.op = torch.ops._C.gelu_tanh_and_mul
 
     def forward_native(self, x: torch.Tensor) -> torch.Tensor:
         """PyTorch-native implementation equivalent to forward()."""
         d = x.shape[-1] // 2
         return F.gelu(x[..., :d], approximate=self.approximate) * x[..., d:]
 
-    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
-        d = x.shape[-1] // 2
-        output_shape = (x.shape[:-1] + (d, ))
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        self.op(out, x)
-        return out
-
     def extra_repr(self) -> str:
         return f'approximate={repr(self.approximate)}'
 
@@ -84,8 +62,6 @@ class NewGELU(CustomOp):
 
     def __init__(self):
         super().__init__()
-        if current_platform.is_cuda_alike() or current_platform.is_cpu():
-            self.op = torch.ops._C.gelu_new
 
     def forward_native(self, x: torch.Tensor) -> torch.Tensor:
         """PyTorch-native implementation equivalent to forward()."""
@@ -93,32 +69,17 @@ class NewGELU(CustomOp):
         return 0.5 * x * (1.0 + torch.tanh(c *
                                            (x + 0.044715 * torch.pow(x, 3.0))))
 
-    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        self.op(out, x)
-        return out
-
-    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
-        return self.op(x)
-
 
 @CustomOp.register("quick_gelu")
 class QuickGELU(CustomOp):
     # https://github.com/huggingface/transformers/blob/main/src/transformers/activations.py#L90
     def __init__(self):
         super().__init__()
-        if current_platform.is_cuda_alike() or current_platform.is_cpu():
-            self.op = torch.ops._C.gelu_quick
 
     def forward_native(self, x: torch.Tensor) -> torch.Tensor:
         """PyTorch-native implementation equivalent to forward()."""
         return x * torch.sigmoid(1.702 * x)
 
-    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        self.op(out, x)
-        return out
-
 
 _ACTIVATION_REGISTRY = {
     "gelu": nn.GELU,

fastvideo/v1/layers/custom_op.py

@@ -50,6 +50,10 @@ class CustomOp(nn.Module):
         return self.forward_native(*args, **kwargs)
 
     def dispatch_forward(self) -> Callable:
+        # FIXME(will): for now, we always use the native implementation, since
+        # forward_cuda is using vllm's custom ops and it doesn't support
+        # backwards. We should add our own custom ops that support backwards.
+        return self.forward_native
         # NOTE(woosuk): Here we assume that vLLM was built for only one
         # specific backend. Currently, we do not support dynamic dispatching.
         enabled = self.enabled()