checkpoint

.buildkite/pipeline.yml

@@ -104,18 +104,6 @@ steps:
                   - TEST_TYPE=distillation_dmd
                 agents:
                   queue: "default"
-            - path:
-                - "fastvideo/training/*self_forcing_distillation_pipeline.py"
-                - "fastvideo/tests/training/self-forcing/**"
-                - "pyproject.toml"
-                - "docker/Dockerfile.python3.12"
-              config:
-                command: "timeout 15m .buildkite/scripts/pr_test.sh"
-                label: "Self-Forcing Tests"
-                env:
-                  - TEST_TYPE=self_forcing
-                agents:
-                  queue: "default"
             - path:
                 - "fastvideo/**"
                 - "pyproject.toml"
@@ -210,15 +198,4 @@ steps:
                 env:  
                   - TEST_TYPE=inference_vmoba
                 agents:
-                  queue: "default"
-            - path:
-                - "fastvideo/**"
-                - "pyproject.toml"
-                - "docker/Dockerfile.python3.12"
-              config:
-                command: "timeout 15m .buildkite/scripts/pr_test.sh"
-                label: "Unit Tests"
-                env:
-                  - TEST_TYPE=unit_test
-                agents:
-                  queue: "default"
+                  queue: "default"

.buildkite/scripts/pr_test.sh

@@ -31,9 +31,9 @@ log "Setting up Modal authentication from Buildkite secrets..."
 MODAL_TOKEN_ID=$(buildkite-agent secret get modal_token_id)
 MODAL_TOKEN_SECRET=$(buildkite-agent secret get modal_token_secret)
 
-# Retrieve other secrets
 WANDB_API_KEY=$(buildkite-agent secret get wandb_api_key)
-HF_API_KEY=$(buildkite-agent secret get hf_api_key)
+
+WANDB_API_KEY=$(buildkite-agent secret get wandb_api_key)
 
 if [ -n "$MODAL_TOKEN_ID" ] && [ -n "$MODAL_TOKEN_SECRET" ]; then
     log "Retrieved Modal credentials from Buildkite secrets"
@@ -63,15 +63,15 @@ MODAL_ENV="BUILDKITE_REPO=$BUILDKITE_REPO BUILDKITE_COMMIT=$BUILDKITE_COMMIT BUI
 case "$TEST_TYPE" in
     "encoder")
         log "Running encoder tests..."
-        MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_encoder_tests"
+        MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_encoder_tests"
         ;;
     "vae")
         log "Running VAE tests..."
-        MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_vae_tests"
+        MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_vae_tests"
         ;;
     "transformer")
         log "Running transformer tests..."
-        MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_transformer_tests"
+        MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_transformer_tests"
         ;;
     "ssim")
         log "Running SSIM tests..."
@@ -110,10 +110,6 @@ case "$TEST_TYPE" in
         MODAL_COMMAND="$MODAL_ENV WANDB_API_KEY=$WANDB_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_distill_dmd_tests"
         ;;
         # run_inference_tests_vmoba
-    "self_forcing")
-        log "Running self-forcing tests..."
-        MODAL_COMMAND="$MODAL_ENV WANDB_API_KEY=$WANDB_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_self_forcing_tests"
-        ;;
     "inference_vmoba")
         log "Running V-MoBA inference tests..."
         MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_inference_tests_vmoba"
@@ -122,10 +118,6 @@ case "$TEST_TYPE" in
         log "Running V-MoBA precision tests..."
         MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_precision_tests_vmoba"
         ;;
-    "unit_test")
-        log "Running unit tests..."
-        MODAL_COMMAND="$MODAL_ENV python3 -m modal run $MODAL_TEST_FILE::run_unit_test"
-        ;;
     *)
         log "Error: Unknown test type: $TEST_TYPE"
         exit 1

.github/workflows/docs.yml

@@ -1,55 +1,82 @@
-name: Deploy Documentation
+# Sample workflow for building and deploying a Hugo site to GitHub Pages
+name: Deploy FastVideo Docs to Pages
 
 on:
+  # Runs on pushes targeting the default branch
   push:
-    branches: [ main ]
+    branches:
+      - main
+    paths:
+      - "docs/**/*.md"
+      - "fastvideo/examples/**/*.py"
   pull_request:
-    branches: [ main ]
+    branches: 
+      - main
+    types: [opened, ready_for_review, synchronize, reopened]
+    paths:
+      - "docs/**/*.md"
+      - "fastvideo/examples/**/*.py"
 
+  # Allows you to run this workflow manually from the Actions tab
+  workflow_dispatch:
+
+# Sets permissions of the GITHUB_TOKEN to allow deployment to GitHub Pages
 permissions:
   contents: read
   pages: write
   id-token: write
 
+# Allow only one concurrent deployment, skipping runs queued between the run in-progress and latest queued.
+# However, do NOT cancel in-progress runs as we want to allow these production deployments to complete.
 concurrency:
   group: "pages"
   cancel-in-progress: false
 
+# Default to bash
+defaults:
+  run:
+    shell: bash
+
 jobs:
+  pre-commit:
+    uses: ./.github/workflows/pre-commit.yml
+
+  # Build job
   build:
     runs-on: ubuntu-latest
+    needs: pre-commit
     steps:
       - name: Checkout
         uses: actions/checkout@v4
-
-      - name: Setup Python
+      - name: Setup Pages
+        id: pages
+        uses: actions/configure-pages@v5
+      - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: '3.12'
-
+          python-version: "3.10"
       - name: Install dependencies
         run: |
-          python -m pip install --upgrade pip
-          pip install -r requirements-mkdocs.txt
-
-      - name: Setup Pages
-        uses: actions/configure-pages@v4
-
-      - name: Build documentation
-        run: mkdocs build
-
+          cd docs
+          pip install -r requirements-docs.txt
+      - name: Build docs
+        run: |
+          cd docs
+          make clean
+          make html
       - name: Upload artifact
         uses: actions/upload-pages-artifact@v3
         with:
-          path: ./site
+          path: ./docs/build/html
 
+  # Deployment job
   deploy:
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
+    if: ${{ github.event_name == 'push' }}
     runs-on: ubuntu-latest
     needs: build
-    if: github.ref == 'refs/heads/main'
     steps:
       - name: Deploy to GitHub Pages
         id: deployment

.github/workflows/pr-test.yml

@@ -62,8 +62,8 @@ on:
         required: false
         default: false
         type: boolean
-      run_unit_test:
-        description: "Run unit-test"
+      run_nightly_test:
+        description: "Run nightly-test"
         required: false
         default: false
         type: boolean
@@ -93,7 +93,6 @@ jobs:
       inference-test-STA: ${{ steps.filter.outputs.inference-test-STA }}
       precision-test-STA: ${{ steps.filter.outputs.precision-test-STA }}
       precision-test-VSA: ${{ steps.filter.outputs.precision-test-VSA }}
-      unit-test: ${{ steps.filter.outputs.unit-test }}
     steps:
       - uses: actions/checkout@v4
       - uses: dorny/paths-filter@v3
@@ -103,8 +102,6 @@ jobs:
             # Define reusable path patterns
             common-paths: &common-paths
               - 'pyproject.toml'
-              - 'docker/Dockerfile.python3.10'
-              - 'docker/Dockerfile.python3.11'
               - 'docker/Dockerfile.python3.12'
             sta-kernel-paths: &sta-kernel-paths
               - 'csrc/attn/sliding_tile_attn/**'
@@ -158,9 +155,6 @@ jobs:
             precision-test-VSA:
               - *common-paths
               - *vsa-kernel-paths
-            unit-test:
-              - 'fastvideo/**'
-              - *common-paths
 
   encoder-test:
     needs: change-filter
@@ -339,42 +333,23 @@ jobs:
       RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
       RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
 
-  unit-test:
-    needs: change-filter
+  nightly-test:
     if: >-
-      (github.event_name != 'workflow_dispatch' && needs.change-filter.outputs.unit-test == 'true') || 
-      (github.event_name == 'workflow_dispatch' && github.event.inputs.run_unit_test == 'true')
+      (github.event_name == 'workflow_dispatch' && github.event.inputs.run_nightly_test == 'true')
     uses: ./.github/workflows/runpod-test.yml
     with:
-      job_id: "unit-test"
-      gpu_type: "NVIDIA L40S"
-      gpu_count: 1
+      job_id: "nightly-test"
+      gpu_type: "NVIDIA A40"
+      gpu_count: 4
       volume_size: 100
       disk_size: 100
       image: "ghcr.io/${{ github.repository }}/fastvideo-dev:py3.12-latest"
-      test_command: "uv pip install -e .[test] && pytest ./fastvideo/dataset/ -vs && pytest ./fastvideo/workflow/ -vs && pytest ./fastvideo/entrypoints/ -vs"
+      test_command: "wandb login $WANDB_API_KEY && uv pip install -e .[test] && pytest ./fastvideo/tests/nightly/test_e2e_overfit_single_sample.py -vs"
       timeout_minutes: 30
     secrets:
       RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
       RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
-
-  # nightly-test:
-  #   if: >-
-  #     (github.event_name == 'workflow_dispatch' && github.event.inputs.run_nightly_test == 'true')
-  #   uses: ./.github/workflows/runpod-test.yml
-  #   with:
-  #     job_id: "nightly-test"
-  #     gpu_type: "NVIDIA A40"
-  #     gpu_count: 4
-  #     volume_size: 100
-  #     disk_size: 100
-  #     image: "ghcr.io/${{ github.repository }}/fastvideo-dev:py3.12-latest"
-  #     test_command: "wandb login $WANDB_API_KEY && uv pip install -e .[test] && pytest ./fastvideo/tests/nightly/test_e2e_overfit_single_sample.py -vs"
-  #     timeout_minutes: 30
-  #   secrets:
-  #     RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
-  #     RUNPOD_PRIVATE_KEY: ${{ secrets.RUNPOD_PRIVATE_KEY }}
-  #     WANDB_API_KEY: ${{ secrets.WANDB_API_KEY }}
+      WANDB_API_KEY: ${{ secrets.WANDB_API_KEY }}
 
   runpod-cleanup:
     # Add other jobs to this list as you create them

.gitignore

@@ -14,8 +14,6 @@ wandb/
 *.pt
 cache_dir/
 wandb/
-venv/
-.venv/
 runs/
 samples/
 *validation/
@@ -39,13 +37,12 @@ dist/
 eggs/
 .eggs/
 
-# MkDocs documentation
-site/
-docs/getting_started/examples/
-docs/inference/examples/
-docs/training/examples/
-docs/distillation/examples/
-!requirements-mkdocs.txt
+# Sphinx documentation
+docs/_build/
+docs/source/getting_started/examples/
+docs/source/inference/examples/
+docs/source/training/examples/
+docs/source/distillation/examples/
 
 # VSCode
 .vscode/
@@ -64,9 +61,7 @@ docs/distillation/examples/
 !fastvideo/tests/ssim/reference_videos/**/*.mp4
 
 # Static images
-!docs/assets/images/**/*.png
+!docs/source/_static/images/**/*.png
 !comfyui/assets/**/*.png
 !comfyui/assets/**/*.gif
-
-dmd_t2v_output/
-preprocess_output_text/
+dmd_t2v_output/

.pre-commit-config.yaml

@@ -10,9 +10,11 @@ exclude: |
     demo/.*|
     predict\.py|
     scripts/.*|
-    prompts/.*|
     fastvideo/data_preprocess/.*|
     fastvideo/dataset/.*|
+    fastvideo/distill/.*|
+    fastvideo/distill\.py|
+    fastvideo/distill_adv\.py|
     fastvideo/models/.*|
     fastvideo/sample/.*|
     fastvideo/train\.py|
@@ -42,10 +44,10 @@ repos:
   - id: codespell
     additional_dependencies: ['tomli']
     args: ['--toml', 'pyproject.toml']
-# - repo: https://github.com/PyCQA/isort
-#   rev: 6.0.1
-#   hooks:
-#   - id: isort
+- repo: https://github.com/PyCQA/isort
+  rev: 6.0.1
+  hooks:
+  - id: isort
 - repo: https://github.com/jackdewinter/pymarkdown
   rev: v0.9.30
   hooks:

README.md

@@ -7,7 +7,7 @@
 FastVideo features an end-to-end unified pipeline for accelerating diffusion models, starting from data preprocessing to model training, finetuning, distillation, and inference. FastVideo is designed to be modular and extensible, allowing users to easily add new optimizations and techniques. Whether it is training-free optimizations or post-training optimizations, FastVideo has you covered.
 
 <p align="center">
-    | 🕹️ <a href="https://fastwan.fastvideo.org/"<b>Online Demo</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/"><b> Quick Start</b></a> | 🤗 <a href="https://huggingface.co/collections/FastVideo/fastwan-6886a305d9799c8cd1496408"  target="_blank"><b>FastWan</b></a>  | 🟣💬 <a href="https://join.slack.com/t/fastvideo/shared_invite/zt-3f4lao1uq-u~Ipx6Lt4J27AlD2y~IdLQ" target="_blank"> <b>Slack</b> </a> |  🟣💬 <a href="https://ibb.co/XcY0Cpv" target="_blank"> <b> WeChat </b> </a> |
+    | 🕹️ <a href="https://fastwan.fastvideo.org/"<b>Online Demo</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/collections/FastVideo/fastwan-6886a305d9799c8cd1496408"  target="_blank"><b>FastWan</b></a>  | 🟣💬 <a href="https://join.slack.com/t/fastvideo/shared_invite/zt-3csdw1isz-Euq8_Q8~baewG8hxjXs2gQ" target="_blank"> <b>Slack</b> </a> |  🟣💬 <a href="https://ibb.co/S7HLCSTh" target="_blank"> <b> WeChat </b> </a> |
 </p>
 
 <div align="center">
@@ -49,10 +49,10 @@ conda activate fastvideo
 pip install fastvideo
 ```
 
-Please see our [docs](https://hao-ai-lab.github.io/FastVideo/getting_started/installation/) for more detailed installation instructions.
+Please see our [docs](https://hao-ai-lab.github.io/FastVideo/getting_started/installation.html) for more detailed installation instructions.
 
 ## Sparse Distillation
-For our sparse distillation techniques, please see our [distillation docs](https://hao-ai-lab.github.io/FastVideo/distillation/dmd/) and check out our [blog](https://hao-ai-lab.github.io/blogs/fastvideo_post_training/).
+For our sparse distillation techniques, please see our [distillation docs](https://hao-ai-lab.github.io/FastVideo/distillation/dmd.html) and check out our [blog](https://hao-ai-lab.github.io/blogs/fastvideo_post_training/).
 
 See below for recipes and datasets:
 
@@ -64,7 +64,7 @@ See below for recipes and datasets:
 
 ## Inference
 ### Generating Your First Video
-Here's a minimal example to generate a video using the default settings. Make sure VSA kernels are [installed](https://hao-ai-lab.github.io/FastVideo/video_sparse_attention/installation/). Create a file called `example.py` with the following code:
+Here's a minimal example to generate a video using the default settings. Make sure VSA kernels are [installed](https://hao-ai-lab.github.io/FastVideo/video_sparse_attention/installation.html). Create a file called `example.py` with the following code:
 
 ```python
 import os
@@ -100,15 +100,15 @@ Run the script with:
 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/).
+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:
 
-- [Design Overview](https://hao-ai-lab.github.io/FastVideo/design/overview/)
-- [Contribution Guide](https://hao-ai-lab.github.io/FastVideo/getting_started/installation/)
+- [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)
 
 ## Distillation and Finetuning
-- [Distillation Guide](https://hao-ai-lab.github.io/FastVideo/distillation/dmd/)
+- [Distillation Guide](https://hao-ai-lab.github.io/FastVideo/distillation/dmd.html)
 <!-- - [Finetuning Guide](https://hao-ai-lab.github.io/FastVideo/training/finetune.html) -->
 
 ## 📑 Development Plan
@@ -127,7 +127,7 @@ See details in [development roadmap](https://github.com/hao-ai-lab/FastVideo/iss
 
 ## 🤝 Contributing
 
-We welcome all contributions. Please check out our guide [here](https://hao-ai-lab.github.io/FastVideo/contributing/overview/)
+We welcome all contributions. Please check out our guide [here](https://hao-ai-lab.github.io/FastVideo/contributing/overview.html)
 
 ## Acknowledgement
 We learned and reused code from the following projects:
@@ -155,8 +155,8 @@ If you find FastVideo useful, please considering citing our work:
 }
 
 @article{zhang2025vsa,
-  title={Vsa: Faster video diffusion with trainable sparse attention},
-  author={Zhang, Peiyuan and Chen, Yongqi and Huang, Haofeng and Lin, Will and Liu, Zhengzhong and Stoica, Ion and Xing, Eric and Zhang, Hao},
+  title={VSA: Faster Video Diffusion with Trainable Sparse Attention},
+  author={Zhang, Peiyuan and Huang, Haofeng and Chen, Yongqi and Lin, Will and Liu, Zhengzhong and Stoica, Ion and Xing, Eric and Zhang, Hao},
   journal={arXiv preprint arXiv:2505.13389},
   year={2025}
 }

assets/full.svg

@@ -1,18 +0,0 @@
-<svg width="252" height="105" viewBox="0 0 252 105" fill="none" xmlns="http://www.w3.org/2000/svg">
-<path d="M89.4843 55.5457H101.361L87.7028 101H74.638L89.4843 55.5457Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M96.0167 1.00057H112.645L118.583 48.273H104.924L103.737 39.7882H79.9827L67.5117 55.5457H85.3273L43.1638 101H28.3174L22.3789 55.5457H33.6621L38.4129 91.3031L58.604 68.2729H44.3515L96.0167 1.00057ZM100.768 13.1217L87.7028 29.4852H103.143L100.768 13.1217Z" fill="#356CFF"/>
-<path d="M37.2252 1.00057L22.3789 48.273H36.0375L40.7884 30.6974L62.6727 30.6974L69.6727 21.0005L43.7576 21.0004L46.7269 11.9096L77.6727 11.9096L86 1.00057L37.2252 1.00057Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M108.488 55.5457L94.2351 101C94.2351 101 105.518 101 120.959 101C136.399 101 144.078 93.0133 148.276 79.788C152.432 68.0157 153.027 55.5457 136.399 55.5457C119.771 55.5457 108.488 55.5457 108.488 55.5457ZM109.081 90.697L116.802 65.8487C116.802 65.8487 120.959 65.8487 132.242 65.8487C143.525 65.8487 137.586 78.5759 135.211 84.0304C133.307 88.4021 127.491 90.697 122.74 90.697C117.989 90.697 109.081 90.697 109.081 90.697Z" fill="#356CFF"/>
-<path d="M173.188 1.00056L168.625 11.9096C168.625 11.9096 149.386 11.9092 142.525 11.9095C135.664 11.9098 136.586 20.3944 141.337 20.3944H159.747C168.654 20.3944 166.961 33.6899 163.904 38.5761C160.467 44.0675 157.371 48.273 148.463 48.273L125.188 48.273L124 37.97L147.87 37.97C153.808 37.97 156.184 29.4852 151.433 29.4852H131.836C120.142 29.4852 125.897 1.00043 141.337 1.00043L173.188 1.00056Z" fill="#356CFF"/>
-<path d="M179.938 1.00056L175.688 11.9096L191.221 11.9096L179.938 48.273H192.409L203.692 11.9096L219.132 11.9095L223.289 1.00043L179.938 1.00056Z" fill="#356CFF"/>
-<path d="M161.341 55.5457H202.845L198.5 65.8487H169.654L167.279 73.7268H188.5L184.749 82.8177H164.31L161.934 90.697H190.251L186.624 101H146.494L161.341 55.5457Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M230.821 54.9391C255.169 54.9391 251.776 67.0602 249.231 77.9692C246.686 88.8783 240.917 101 217.757 101C194.596 101 195.606 88.8783 199.347 77.9692C203.089 67.0602 206.473 54.9391 230.821 54.9391ZM237.948 77.9692C239.984 70.6965 240.917 65.242 228.446 65.242C215.975 65.242 211.818 71.9087 210.037 77.9692C208.255 84.0298 208.255 91.3025 219.538 91.3025C230.821 91.3025 235.911 85.2419 237.948 77.9692Z" fill="#356CFF"/>
-<path d="M173.188 1.00056L168.625 11.9096C168.625 11.9096 149.386 11.9092 142.525 11.9095C135.664 11.9098 136.586 20.3944 141.337 20.3944M173.188 1.00056C173.188 1.00056 156.777 1.00043 141.337 1.00043M173.188 1.00056L141.337 1.00043M141.337 20.3944C146.088 20.3944 150.839 20.3944 159.747 20.3944M141.337 20.3944H159.747M159.747 20.3944C168.654 20.3944 166.961 33.6899 163.904 38.5761C160.467 44.0675 157.371 48.273 148.463 48.273M148.463 48.273C139.556 48.273 125.188 48.273 125.188 48.273M148.463 48.273L125.188 48.273M125.188 48.273L124 37.97M124 37.97C124 37.97 141.931 37.97 147.87 37.97M124 37.97L147.87 37.97M147.87 37.97C153.808 37.97 156.184 29.4852 151.433 29.4852M151.433 29.4852C146.682 29.4852 138.962 29.4852 131.836 29.4852M151.433 29.4852H131.836M131.836 29.4852C120.142 29.4852 125.897 1.00043 141.337 1.00043M37.2252 1.00057L22.3789 48.273H36.0375L40.7884 30.6974L62.6727 30.6974L69.6727 21.0005L43.7576 21.0004L46.7269 11.9096L77.6727 11.9096L86 1.00057L37.2252 1.00057ZM96.0167 1.00057H112.645L118.583 48.273H104.924L103.737 39.7882H79.9827L67.5117 55.5457H85.3273L43.1638 101H28.3174L22.3789 55.5457H33.6621L38.4129 91.3031L58.604 68.2729H44.3515L96.0167 1.00057ZM87.7028 29.4852L100.768 13.1217L103.143 29.4852H87.7028ZM89.4843 55.5457H101.361L87.7028 101H74.638L89.4843 55.5457ZM108.488 55.5457L94.2351 101C94.2351 101 105.518 101 120.959 101C136.399 101 144.078 93.0133 148.276 79.788C152.432 68.0157 153.027 55.5457 136.399 55.5457C119.771 55.5457 108.488 55.5457 108.488 55.5457ZM116.802 65.8487L109.081 90.697C109.081 90.697 117.989 90.697 122.74 90.697C127.491 90.697 133.307 88.4021 135.211 84.0304C137.586 78.5759 143.525 65.8487 132.242 65.8487C120.959 65.8487 116.802 65.8487 116.802 65.8487ZM179.938 1.00056L175.688 11.9096L191.221 11.9096L179.938 48.273H192.409L203.692 11.9096L219.132 11.9095L223.289 1.00043L179.938 1.00056ZM161.341 55.5457H202.845L198.5 65.8487H169.654L167.279 73.7268H188.5L184.749 82.8177H164.31L161.934 90.697H190.251L186.624 101H146.494L161.341 55.5457ZM230.821 54.9391C255.169 54.9391 251.776 67.0602 249.231 77.9692C246.686 88.8783 240.917 101 217.757 101C194.596 101 195.606 88.8783 199.347 77.9692C203.089 67.0602 206.473 54.9391 230.821 54.9391ZM228.446 65.242C240.917 65.242 239.984 70.6965 237.948 77.9692C235.911 85.2419 230.821 91.3025 219.538 91.3025C208.255 91.3025 208.255 84.0298 210.037 77.9692C211.818 71.9087 215.975 65.242 228.446 65.242Z" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M15.2524 55.5451L21.191 100.999L24.7541 100.999L18.8156 55.5451L15.2524 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M8.12646 55.5451L14.065 100.999L15.2527 100.999L9.31417 55.5451L8.12646 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M1 55.5451L6.93853 100.999L7.53239 100.999L1.59385 55.5451L1 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="0.593853"/>
-<path d="M15.2524 48.2724L30.0988 1H33.6619L18.8156 48.2724H15.2524Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M8.12646 48.2724L22.9728 1H24.1605L9.31417 48.2724H8.12646Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M1 48.2724L15.8463 1H16.4402L1.59385 48.2724H1Z" fill="#356CFF" stroke="#356CFF" stroke-width="0.593853"/>
-<path d="M85.3271 55.5457H67.5116L87 12.7363L44.3513 68.2729H58.6038L43.1636 101L85.3271 55.5457Z" fill="#FDC717" stroke="#FDC717" stroke-width="1.18771" stroke-miterlimit="16"/>
-</svg>

assets/icon-simple.svg

@@ -1,6 +0,0 @@
-<svg width="160" height="93" viewBox="0 0 160 93" fill="none" xmlns="http://www.w3.org/2000/svg">
-<path d="M28.8511 91.66L57.6319 1.86368H64.5394L35.7585 91.66H28.8511Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-<path d="M15.0376 91.66L43.8185 1.86368H46.1209L17.3401 91.66H15.0376Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-<path d="M1.22217 91.66L30.003 1.86366H31.1543L2.3734 91.66H1.22217Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.15122"/>
-<path d="M71.4465 1.86483L42.666 91.6599H69.144L78.3538 58.2746H123.251L129.007 39.855H84.1099L89.866 22.5868H152.032L157.788 1.86483H71.4465Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-</svg>

csrc/attn/sliding_tile_attn/README.md

@@ -72,15 +72,6 @@ We give a demo for 2D STA with window size (6,6) operating on a (10, 10) image.
 
 https://github.com/user-attachments/assets/f3b6dd79-7b43-4b60-a0fa-3d6495ec5747
 
-
-## STA Configuration Logic
-Here is a diagram of how the window is configured and passed through the FastVideo pipeline:
-
-<div align="center">
-<img src="../../../docs/assets/images/STA_configuration.png" width="80%"/>
-</div>
-
-
 ## Why is STA Fast?
 2D/3D Sliding Window Attention (SWA) creates many mixed blocks in the attention map. Even though mixed blocks have less output value,a mixed block is significantly slower than a dense block due to the GPU-unfriendly masking operation. 
 

csrc/attn/video_sparse_attn/vsa/block_sparse_attn_triton.py

@@ -247,11 +247,11 @@ def _attn_bwd_dq(dq, q, K, V,  #
 
     kv_blocks = tl.load(q2k_num  + meta_base)                 # int32
     kv_ptr    = q2k_index + meta_base * max_kv_blks           # ptr to list
-    block_size = tl.load(variable_block_sizes + q_blk)
     
     
     for blk_idx in range(kv_blocks*2):
         block_sparse_offset = (tl.load(kv_ptr + blk_idx//2).to(tl.int32)*2 + blk_idx%2) *step_n * stride_tok
+        block_size = tl.load(variable_block_sizes + blk_idx//2) - (blk_idx%2) * step_n
         kT = tl.load(kT_ptrs + block_sparse_offset)
         vT = tl.load(vT_ptrs + block_sparse_offset)
         qk = tl.dot(q, kT)

csrc/attn/vmoba_attn/setup.py

@@ -20,7 +20,5 @@ setup(
         "License :: OSI Approved :: Apache Software License",
     ],
     python_requires='>=3.12',
-    install_requires=[
-        "flash-attn >= 2.7.1",
-    ]
+    install_requires=[]
 )

csrc/attn/vmoba_attn/vmoba/vmoba.py

@@ -6,16 +6,8 @@ import time
 import os
 import torch
 from typing import Tuple
-try:
-    from flash_attn import flash_attn_varlen_func  # Use the new flash attention function
-    from flash_attn.flash_attn_interface import _flash_attn_varlen_forward, _flash_attn_varlen_backward
-except ImportError:
-    def _unsupported(*args, **kwargs):
-        raise ImportError("flash-attn is not installed. Please install it, e.g., `pip install flash-attn`.")
-    _flash_attn_varlen_forward = _unsupported
-    _flash_attn_varlen_backward = _unsupported
-    flash_attn_varlen_func = _unsupported
-    
+from flash_attn import flash_attn_varlen_func  # Use the new flash attention function
+from flash_attn.flash_attn_interface import _flash_attn_varlen_forward, _flash_attn_varlen_backward
 from functools import lru_cache
 from einops import rearrange
 

docs/Makefile

@@ -0,0 +1,26 @@
+# Minimal makefile for Sphinx documentation
+#
+
+# You can set these variables from the command line, and also
+# from the environment for the first two.
+SPHINXOPTS    ?=
+SPHINXBUILD   ?= sphinx-build
+SOURCEDIR     = source
+BUILDDIR      = build
+
+# Put it first so that "make" without argument is like "make help".
+help:
+	@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+.PHONY: help Makefile
+
+# Catch-all target: route all unknown targets to Sphinx using the new
+# "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
+%: Makefile
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+clean:
+	@$(SPHINXBUILD) -M clean "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+	rm -rf "$(SOURCEDIR)/getting_started/examples"
+	rm -rf "$(SOURCEDIR)/inference/examples"
+	rm -rf "$(SOURCEDIR)/training/examples"

docs/README.md

@@ -1,39 +1,20 @@
-# FastVideo Documentation
+# FastVideo documents
 
-This directory contains the FastVideo documentation built with MkDocs.
-
-## Build the docs locally
+## Build the docs
 
 ```bash
-# Install dependencies
-pip install -r docs/requirements-mkdocs.txt
+# Install dependencies.
+pip install -r requirements-docs.txt
 
-# Serve docs with live reload (recommended for development)
-mkdocs serve
-
-# Or build static site
-mkdocs build
+# Build the docs.
+make clean
+make html
 ```
 
-## View the docs
-
-### Development server (with live reload)
+## Open the docs with your browser
 
 ```bash
-mkdocs serve
+python -m http.server -d build/html/
 ```
 
-Then open your browser to: http://127.0.0.1:8000
-
-### Static build
-
-```bash
-mkdocs build
-python -m http.server -d site/
-```
-
-Then open your browser to: http://localhost:8000
-
-## Automatic Deployment
-
-Documentation is automatically built and deployed to GitHub Pages when changes are pushed to the `main` branch via the `.github/workflows/docs.yml` workflow.
+Launch your browser and open localhost:8000.

docs/api/fastvideo.md

@@ -1,248 +0,0 @@
-# FastVideo API Reference
-
-This page contains the complete API reference for the FastVideo library.
-
-## fastvideo
-
-### Modules
-
-| Name | Description |
-|------|-------------|
-| [attention](#fastvideoattention) | Attention mechanisms and backends for video generation |
-| [configs](#fastvideoconfigs) | Configuration classes for pipelines, models, and sampling |
-| [distributed](#fastvideodistributed) | Distributed execution and communication utilities |
-| [entrypoints](#fastvideoentrypoints) | Main API entry points for video generation |
-| [models](#fastvideomodels) | Model implementations (transformers, VAEs, schedulers) |
-| [pipelines](#fastvideopipelines) | Core pipeline classes for video diffusion |
-| [training](#fastvideotraining) | Training utilities and helpers |
-| [workflow](#fastvideoworkflow) | Workflow management and orchestration |
-| [dataset](#fastvideodataset) | Dataset handling and preprocessing |
-| [layers](#fastvideolayers) | Custom neural network layers |
-| [platforms](#fastvideoplatforms) | Platform-specific implementations |
-| [utils](#fastvideoutils) | Utility functions and helpers |
-| [worker](#fastvideoworker) | Execution workers for video generation |
-
-## fastvideo.attention
-
-::: fastvideo.attention
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.configs
-
-::: fastvideo.configs
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-### Submodules
-
-#### fastvideo.configs.pipelines
-
-::: fastvideo.configs.pipelines
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.configs.models
-
-::: fastvideo.configs.models
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.configs.sample
-
-::: fastvideo.configs.sample
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-## fastvideo.distributed
-
-::: fastvideo.distributed
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.entrypoints
-
-::: fastvideo.entrypoints
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.models
-
-::: fastvideo.models
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-### Submodules
-
-#### fastvideo.models.registry
-
-::: fastvideo.models.registry
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.models.loader
-
-::: fastvideo.models.loader
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-## fastvideo.pipelines
-
-::: fastvideo.pipelines
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-### Submodules
-
-#### fastvideo.pipelines.composed_pipeline_base
-
-::: fastvideo.pipelines.composed_pipeline_base
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.pipelines.lora_pipeline
-
-::: fastvideo.pipelines.lora_pipeline
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.pipelines.pipeline_batch_info
-
-::: fastvideo.pipelines.pipeline_batch_info
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.pipelines.pipeline_registry
-
-::: fastvideo.pipelines.pipeline_registry
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-#### fastvideo.pipelines.stages
-
-::: fastvideo.pipelines.stages
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 4
-
-## fastvideo.training
-
-::: fastvideo.training
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.workflow
-
-::: fastvideo.workflow
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.dataset
-
-::: fastvideo.dataset
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.layers
-
-::: fastvideo.layers
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.platforms
-
-::: fastvideo.platforms
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3
-
-## fastvideo.utils
-
-::: fastvideo.utils
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      heading_level: 3
-
-## fastvideo.worker
-
-::: fastvideo.worker
-    options:
-      show_source: true
-      show_root_heading: true
-      show_root_toc_entry: true
-      show_submodules: true
-      heading_level: 3

docs/api/summary.md

@@ -1,27 +0,0 @@
-# API Summary
-
-This page provides a quick overview of the main FastVideo API components.
-
-## Video Generator
-
-::: fastvideo.VideoGenerator
-    options:
-      show_root_heading: false
-      show_source: false
-      heading_level: 3
-
-## Initialization Configuration
-
-::: fastvideo.PipelineConfig
-    options:
-      show_root_heading: false
-      show_source: false
-      heading_level: 3
-
-## Sampling Configuration
-
-::: fastvideo.SamplingParam
-    options:
-      show_root_heading: false
-      show_source: false
-      heading_level: 3

docs/assets/custom.css

@@ -1,41 +0,0 @@
-.vertical-table-header th.head:not(.stub) {
-    writing-mode: sideways-lr;
-    white-space: nowrap;
-    max-width: 0;
-    p {
-       margin: 0;
-    }
-}
-
-/* Image sizing classes */
-.image-small {
-    max-width: 200px;
-    height: auto;
-}
-
-.image-medium {
-    max-width: 400px;
-    height: auto;
-}
-
-.image-large {
-    max-width: 600px;
-    height: auto;
-}
-
-.image-full {
-    max-width: 100%;
-    height: auto;
-}
-
-/* Responsive images */
-img {
-    max-width: 100%;
-    height: auto;
-}
-
-/* Center images */
-.image-center {
-    display: block;
-    margin: 0 auto;
-}

docs/assets/logos/icon_simple.svg

@@ -1,6 +0,0 @@
-<svg width="160" height="93" viewBox="0 0 160 93" fill="none" xmlns="http://www.w3.org/2000/svg">
-<path d="M28.8511 91.66L57.6319 1.86368H64.5394L35.7585 91.66H28.8511Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-<path d="M15.0376 91.66L43.8185 1.86368H46.1209L17.3401 91.66H15.0376Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-<path d="M1.22217 91.66L30.003 1.86366H31.1543L2.3734 91.66H1.22217Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.15122"/>
-<path d="M71.4465 1.86483L42.666 91.6599H69.144L78.3538 58.2746H123.251L129.007 39.855H84.1099L89.866 22.5868H152.032L157.788 1.86483H71.4465Z" fill="#356CFF" stroke="#356CFF" stroke-width="2.30244"/>
-</svg>

docs/assets/logos/logo.svg

@@ -1,18 +0,0 @@
-<svg width="252" height="105" viewBox="0 0 252 105" fill="none" xmlns="http://www.w3.org/2000/svg">
-<path d="M89.4843 55.5457H101.361L87.7028 101H74.638L89.4843 55.5457Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M96.0167 1.00057H112.645L118.583 48.273H104.924L103.737 39.7882H79.9827L67.5117 55.5457H85.3273L43.1638 101H28.3174L22.3789 55.5457H33.6621L38.4129 91.3031L58.604 68.2729H44.3515L96.0167 1.00057ZM100.768 13.1217L87.7028 29.4852H103.143L100.768 13.1217Z" fill="#356CFF"/>
-<path d="M37.2252 1.00057L22.3789 48.273H36.0375L40.7884 30.6974L62.6727 30.6974L69.6727 21.0005L43.7576 21.0004L46.7269 11.9096L77.6727 11.9096L86 1.00057L37.2252 1.00057Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M108.488 55.5457L94.2351 101C94.2351 101 105.518 101 120.959 101C136.399 101 144.078 93.0133 148.276 79.788C152.432 68.0157 153.027 55.5457 136.399 55.5457C119.771 55.5457 108.488 55.5457 108.488 55.5457ZM109.081 90.697L116.802 65.8487C116.802 65.8487 120.959 65.8487 132.242 65.8487C143.525 65.8487 137.586 78.5759 135.211 84.0304C133.307 88.4021 127.491 90.697 122.74 90.697C117.989 90.697 109.081 90.697 109.081 90.697Z" fill="#356CFF"/>
-<path d="M173.188 1.00056L168.625 11.9096C168.625 11.9096 149.386 11.9092 142.525 11.9095C135.664 11.9098 136.586 20.3944 141.337 20.3944H159.747C168.654 20.3944 166.961 33.6899 163.904 38.5761C160.467 44.0675 157.371 48.273 148.463 48.273L125.188 48.273L124 37.97L147.87 37.97C153.808 37.97 156.184 29.4852 151.433 29.4852H131.836C120.142 29.4852 125.897 1.00043 141.337 1.00043L173.188 1.00056Z" fill="#356CFF"/>
-<path d="M179.938 1.00056L175.688 11.9096L191.221 11.9096L179.938 48.273H192.409L203.692 11.9096L219.132 11.9095L223.289 1.00043L179.938 1.00056Z" fill="#356CFF"/>
-<path d="M161.341 55.5457H202.845L198.5 65.8487H169.654L167.279 73.7268H188.5L184.749 82.8177H164.31L161.934 90.697H190.251L186.624 101H146.494L161.341 55.5457Z" fill="#356CFF"/>
-<path fill-rule="evenodd" clip-rule="evenodd" d="M230.821 54.9391C255.169 54.9391 251.776 67.0602 249.231 77.9692C246.686 88.8783 240.917 101 217.757 101C194.596 101 195.606 88.8783 199.347 77.9692C203.089 67.0602 206.473 54.9391 230.821 54.9391ZM237.948 77.9692C239.984 70.6965 240.917 65.242 228.446 65.242C215.975 65.242 211.818 71.9087 210.037 77.9692C208.255 84.0298 208.255 91.3025 219.538 91.3025C230.821 91.3025 235.911 85.2419 237.948 77.9692Z" fill="#356CFF"/>
-<path d="M173.188 1.00056L168.625 11.9096C168.625 11.9096 149.386 11.9092 142.525 11.9095C135.664 11.9098 136.586 20.3944 141.337 20.3944M173.188 1.00056C173.188 1.00056 156.777 1.00043 141.337 1.00043M173.188 1.00056L141.337 1.00043M141.337 20.3944C146.088 20.3944 150.839 20.3944 159.747 20.3944M141.337 20.3944H159.747M159.747 20.3944C168.654 20.3944 166.961 33.6899 163.904 38.5761C160.467 44.0675 157.371 48.273 148.463 48.273M148.463 48.273C139.556 48.273 125.188 48.273 125.188 48.273M148.463 48.273L125.188 48.273M125.188 48.273L124 37.97M124 37.97C124 37.97 141.931 37.97 147.87 37.97M124 37.97L147.87 37.97M147.87 37.97C153.808 37.97 156.184 29.4852 151.433 29.4852M151.433 29.4852C146.682 29.4852 138.962 29.4852 131.836 29.4852M151.433 29.4852H131.836M131.836 29.4852C120.142 29.4852 125.897 1.00043 141.337 1.00043M37.2252 1.00057L22.3789 48.273H36.0375L40.7884 30.6974L62.6727 30.6974L69.6727 21.0005L43.7576 21.0004L46.7269 11.9096L77.6727 11.9096L86 1.00057L37.2252 1.00057ZM96.0167 1.00057H112.645L118.583 48.273H104.924L103.737 39.7882H79.9827L67.5117 55.5457H85.3273L43.1638 101H28.3174L22.3789 55.5457H33.6621L38.4129 91.3031L58.604 68.2729H44.3515L96.0167 1.00057ZM87.7028 29.4852L100.768 13.1217L103.143 29.4852H87.7028ZM89.4843 55.5457H101.361L87.7028 101H74.638L89.4843 55.5457ZM108.488 55.5457L94.2351 101C94.2351 101 105.518 101 120.959 101C136.399 101 144.078 93.0133 148.276 79.788C152.432 68.0157 153.027 55.5457 136.399 55.5457C119.771 55.5457 108.488 55.5457 108.488 55.5457ZM116.802 65.8487L109.081 90.697C109.081 90.697 117.989 90.697 122.74 90.697C127.491 90.697 133.307 88.4021 135.211 84.0304C137.586 78.5759 143.525 65.8487 132.242 65.8487C120.959 65.8487 116.802 65.8487 116.802 65.8487ZM179.938 1.00056L175.688 11.9096L191.221 11.9096L179.938 48.273H192.409L203.692 11.9096L219.132 11.9095L223.289 1.00043L179.938 1.00056ZM161.341 55.5457H202.845L198.5 65.8487H169.654L167.279 73.7268H188.5L184.749 82.8177H164.31L161.934 90.697H190.251L186.624 101H146.494L161.341 55.5457ZM230.821 54.9391C255.169 54.9391 251.776 67.0602 249.231 77.9692C246.686 88.8783 240.917 101 217.757 101C194.596 101 195.606 88.8783 199.347 77.9692C203.089 67.0602 206.473 54.9391 230.821 54.9391ZM228.446 65.242C240.917 65.242 239.984 70.6965 237.948 77.9692C235.911 85.2419 230.821 91.3025 219.538 91.3025C208.255 91.3025 208.255 84.0298 210.037 77.9692C211.818 71.9087 215.975 65.242 228.446 65.242Z" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M15.2524 55.5451L21.191 100.999L24.7541 100.999L18.8156 55.5451L15.2524 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M8.12646 55.5451L14.065 100.999L15.2527 100.999L9.31417 55.5451L8.12646 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M1 55.5451L6.93853 100.999L7.53239 100.999L1.59385 55.5451L1 55.5451Z" fill="#356CFF" stroke="#356CFF" stroke-width="0.593853"/>
-<path d="M15.2524 48.2724L30.0988 1H33.6619L18.8156 48.2724H15.2524Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M8.12646 48.2724L22.9728 1H24.1605L9.31417 48.2724H8.12646Z" fill="#356CFF" stroke="#356CFF" stroke-width="1.18771"/>
-<path d="M1 48.2724L15.8463 1H16.4402L1.59385 48.2724H1Z" fill="#356CFF" stroke="#356CFF" stroke-width="0.593853"/>
-<path d="M85.3271 55.5457H67.5116L87 12.7363L44.3513 68.2729H58.6038L43.1636 101L85.3271 55.5457Z" fill="#FDC717" stroke="#FDC717" stroke-width="1.18771" stroke-miterlimit="16"/>
-</svg>

docs/contributing/profiling.md

@@ -1,53 +0,0 @@
-# Profiling FastVideo
-
-!!! warning
-    Profiling is only intended for FastVideo developers and maintainers to understand the proportion of time spent in different parts of the codebase. **FastVideo end-users should never turn on profiling** as it will significantly slow down the inference.
-
-## Profiling with PyTorch
-
-FastVideo exposes a process-wide torch profiler that you can enable via environment variables. Set `FASTVIDEO_TORCH_PROFILER_DIR` to an absolute directory path to start collecting traces, and specify the regions you want recorded with `FASTVIDEO_TORCH_PROFILE_REGIONS`:
-
-```bash
-FASTVIDEO_TORCH_PROFILER_DIR=/mnt/traces/fastvideo \
-FASTVIDEO_TORCH_PROFILE_REGIONS="profiler_region_model_loading,profiler_region_training_step"
-```
-
-All profiled regions must be registered in `fastvideo.profiler`; the current list includes:
-
-- `profiler_region_model_loading` — pipeline/module loading
-- `profiler_region_inference_pre_denoising`
-- `profiler_region_inference_denoising`
-- `profiler_region_inference_post_denoising`
-- `profiler_region_training_checkpoint_saving`
-- `profiler_region_training_dit`
-- `profiler_region_training_validation`
-- `profiler_region_training_epoch`
-- `profiler_region_training_step`
-- `profiler_region_training_backward`
-- `profiler_region_training_optimizer`
-- `profiler_region_distillation_teacher_forward`
-- `profiler_region_distillation_student_forward`
-- `profiler_region_distillation_loss`
-- `profiler_region_distillation_update`
-
-While profiling is enabled, FastVideo records additional annotations:
-
-- `fastvideo.region::<name>` spans are emitted when entering a region.
-- `fastvideo.profiler.enable_collection` / `fastvideo.profiler.disable_collection` events mark when torch profiler collection is toggled on or off.
-
-Only one profiler instance is created per process; subsequent pipelines reuse the same controller. If you set `FASTVIDEO_TORCH_PROFILE_REGIONS` incorrectly (e.g. misspelled name), FastVideo logs a warning and ignores that entry.
-
-Additional knobs:
-
-- `FASTVIDEO_TORCH_PROFILER_RECORD_SHAPES`
-- `FASTVIDEO_TORCH_PROFILER_WITH_PROFILE_MEMORY`
-- `FASTVIDEO_TORCH_PROFILER_WITH_STACK`
-- `FASTVIDEO_TORCH_PROFILER_WITH_FLOPS`
-
-Traces can be visualized using <https://ui.perfetto.dev/>.
-
-### Best Practices
-
-- Keep the profiled step count small; traces can be large and slow down job shutdown while the profiler flushes data.
-- After profiling, clean up trace directories to avoid filling disks.
-- When adding new regions, register them in `fastvideo.profiler` and wrap the corresponding code block with `with self.profiler_controller.region("your_region"):` or the `@profile_region` decorator.

docs/examples/examples_index.md

@@ -1,12 +0,0 @@
-# 💡 Examples
-
-A collection of examples demonstrating usage of FastVideo.
-
-All documented examples are autogenerated using [generate_examples.py](https://github.com/hao-ai-lab/FastVideo/blob/main/docs/generate_examples.py) from examples found in the [examples](https://github.com/hao-ai-lab/FastVideo/tree/main/examples) directory.
-
-## Examples
-
-- [Examples Distillation Index](distillation/examples/examples_distillation_index.md)
-- [Examples Training Index](training/examples/examples_training_index.md)
-- [Examples Inference Index](inference/examples/examples_inference_index.md)
-

docs/getting_started/installation.md

@@ -1,41 +0,0 @@
-
-# 🔧 Installation
-
-FastVideo supports the following hardware platforms:
-
-- [NVIDIA CUDA](installation/gpu.md)
-- [Apple silicon](installation/mps.md)
-
-## Quick Installation
-
-### Using pip
-
-```bash
-pip install fastvideo
-```
-
-### Using conda
-
-```bash
-conda install -c conda-forge fastvideo
-```
-
-### From source
-
-```bash
-git clone https://github.com/hao-ai-lab/FastVideo.git
-cd FastVideo
-pip install -e .
-```
-
-## Hardware Requirements
-
-- **NVIDIA GPUs**: CUDA 11.8+ with compute capability 7.0+
-- **Apple Silicon**: macOS 12.0+ with M1/M2/M3 chips
-- **CPU**: x86_64 architecture (for CPU-only inference)
-
-## Next Steps
-
-- [Quick Start Guide](quick_start.md) - Get started with your first video generation
-- [Configuration](../inference/configuration.md) - Learn about configuration options
-- [Examples](../inference/examples/) - Explore example scripts and notebooks

docs/getting_started/quick_start.md

@@ -1,55 +0,0 @@
-# 🚀 Quick Start
-
-Get up and running with FastVideo in minutes!
-
-## Installation
-
-First, install FastVideo:
-
-```bash
-pip install fastvideo
-```
-
-## Basic Usage
-
-### Text-to-Video Generation
-
-```python
-from fastvideo import FastVideoPipeline
-
-# Initialize the pipeline
-pipe = FastVideoPipeline.from_pretrained("wan2.1-t2v-1.3B")
-
-# Generate a video
-prompt = "A cat playing with a ball of yarn"
-video = pipe(prompt, num_frames=16, height=512, width=512)
-
-# Save the video
-video.save("output.mp4")
-```
-
-### Image-to-Video Generation
-
-```python
-from fastvideo import FastVideoPipeline
-from PIL import Image
-
-# Load an image
-image = Image.open("input.jpg")
-
-# Initialize the pipeline
-pipe = FastVideoPipeline.from_pretrained("wan2.1-i2v-14B-480p")
-
-# Generate a video from the image
-video = pipe(image, num_frames=16, height=480, width=480)
-
-# Save the video
-video.save("output.mp4")
-```
-
-## Next Steps
-
-- [Installation Guide](installation.md) - Detailed installation instructions
-- [Configuration](../inference/configuration.md) - Learn about configuration options
-- [Examples](../inference/examples/) - Explore more examples
-- [Optimizations](../inference/optimizations.md) - Performance optimization tips

docs/getting_started/v1_api.md

@@ -1,42 +0,0 @@
-# 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.
-
-::: fastvideo.entrypoints.video_generator.VideoGenerator
-    options:
-      show_root_heading: true
-      show_source: false
-      members:
-        - from_pretrained
-      heading_level: 3
-
-`VideoGenerator.from_pretrained()` should be the primary way of creating a new video generator.
-
-## Configuring FastVideo
-
-The following two classes `PipelineConfig` and `SamplingParam` are used to configure initialization and sampling parameters, respectively.
-
-### PipelineConfig
-
-::: fastvideo.configs.pipelines.base.PipelineConfig
-    options:
-      show_root_heading: true
-      show_source: false
-      members:
-        - from_pretrained
-        - dump_to_json
-      heading_level: 4
-
-### SamplingParam
-
-::: fastvideo.configs.sample.base.SamplingParam
-    options:
-      show_root_heading: true
-      show_source: false
-      members:
-        - from_pretrained
-      heading_level: 4

docs/inference/support_matrix.md

@@ -1,66 +0,0 @@
-# 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
-- ⭕ = Does not apply to this model
-
-## 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.
-
-<style>
-  /* Target tables in this section */
-  #models-x-optimization + p + table {
-    display: block;
-    overflow-x: auto;
-    width: 100%;
-    font-size: 0.85rem;
-  }
-  
-  #models-x-optimization + p + table td,
-  #models-x-optimization + p + table th {
-    text-align: center;
-    white-space: nowrap;
-    padding: 0.5em;
-  }
-  
-  /* First two columns can wrap */
-  #models-x-optimization + p + table td:nth-child(1),
-  #models-x-optimization + p + table td:nth-child(2) {
-    white-space: normal;
-    min-width: 120px;
-  }
-  
-  #models-x-optimization + p + table td:nth-child(2) code {
-    font-size: 0.75rem;
-  }
-</style>
-
-| Model Name | HuggingFace Model ID | Resolutions | TeaCache | Sliding Tile Attn | Sage Attn | VSA |
-|------------|---------------------|-------------|----------|-------------------|-----------|-----|
-| FastWan2.1 T2V 1.3B | `FastVideo/FastWan2.1-T2V-1.3B-Diffusers` | 480P | ⭕ | ⭕ | ⭕ | ✅ |
-| FastWan2.2 TI2V 5B Full Attn* | `FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers` | 720P | ⭕ | ⭕ | ⭕ | ✅ |
-| Wan2.2 TI2V 5B | `Wan-AI/Wan2.2-TI2V-5B-Diffusers` | 720P | ⭕ | ⭕ | ✅ | ⭕ |
-| Wan2.2 T2V A14B | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 480P<br>720P | ❌ | ❌ | ✅ | ⭕ |
-| Wan2.2 I2V A14B | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | 480P<br>720P | ❌ | ❌ | ✅ | ⭕ |
-| HunyuanVideo | `hunyuanvideo-community/HunyuanVideo` | 720px1280p<br>544px960p | ❌ | ✅ | ✅ | ⭕ |
-| FastHunyuan | `FastVideo/FastHunyuan-diffusers` | 720px1280p<br>544px960p | ❌ | ✅ | ✅ | ⭕ |
-| Wan2.1 T2V 1.3B | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers` | 480P | ✅ | ✅* | ✅ | ⭕ |
-| Wan2.1 T2V 14B | `Wan-AI/Wan2.1-T2V-14B-Diffusers` | 480P, 720P | ✅ | ✅* | ✅ | ⭕ |
-| Wan2.1 I2V 480P | `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers` | 480P | ✅ | ✅* | ✅ | ⭕ |
-| Wan2.1 I2V 720P | `Wan-AI/Wan2.1-I2V-14B-720P-Diffusers` | 720P | ✅ | ✅ | ✅ | ⭕ |
-| StepVideo T2V | `FastVideo/stepvideo-t2v-diffusers` | 768px768px204f<br>544px992px204f<br>544px992px136f | ❌ | ❌ | ✅ | ⭕ |
-
-**Note**: Wan2.2 TI2V 5B has some quality issues when performing I2V generation. 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/make.bat

@@ -0,0 +1,35 @@
+@ECHO OFF
+
+pushd %~dp0
+
+REM Command file for Sphinx documentation
+
+if "%SPHINXBUILD%" == "" (
+	set SPHINXBUILD=sphinx-build
+)
+set SOURCEDIR=source
+set BUILDDIR=build
+
+%SPHINXBUILD% >NUL 2>NUL
+if errorlevel 9009 (
+	echo.
+	echo.The 'sphinx-build' command was not found. Make sure you have Sphinx
+	echo.installed, then set the SPHINXBUILD environment variable to point
+	echo.to the full path of the 'sphinx-build' executable. Alternatively you
+	echo.may add the Sphinx directory to PATH.
+	echo.
+	echo.If you don't have Sphinx installed, grab it from
+	echo.https://www.sphinx-doc.org/
+	exit /b 1
+)
+
+if "%1" == "" goto help
+
+%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+goto end
+
+:help
+%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+
+:end
+popd

docs/requirements-docs.txt

@@ -0,0 +1,15 @@
+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
+commonmark # Required by sphinx-argparse when using :markdownhelp:
+
+# packages to install to build the documentation
+cachetools
+# -f https://download.pytorch.org/whl/cpu
+torch

docs/source/_static/custom.css

@@ -0,0 +1,8 @@
+.vertical-table-header th.head:not(.stub) {
+    writing-mode: sideways-lr;
+    white-space: nowrap;
+    max-width: 0;
+    p {
+       margin: 0;
+    }
+}

docs/source/_templates/sections/header.html

@@ -0,0 +1,39 @@
+<style>
+  .notification-bar {
+    width: 100vw;
+    display: flex;
+    justify-content: center;
+    align-items: center;
+    font-size: 16px;
+    padding: 0 6px 0 6px;
+  }
+  .notification-bar p {
+    margin: 0;
+  }
+  .notification-bar a {
+    font-weight: bold;
+    text-decoration: none;
+  }
+
+  /* Light mode styles (default) */
+  .notification-bar {
+    background-color: #fff3cd;
+    color: #856404;
+  }
+  .notification-bar a {
+    color: #d97706;
+  }
+
+  /* Dark mode styles */
+  html[data-theme=dark] .notification-bar {
+    background-color: #333;
+    color: #ddd;
+  }
+  html[data-theme=dark] .notification-bar a {
+    color: #ffa500; /* Brighter color for visibility */
+  }
+</style>
+
+<!-- <div class="notification-bar">
+  <p>You are viewing the latest developer preview docs. <a href="https://docs.vllm.ai/en/stable/">Click here</a> to view docs for the latest stable release.</p>
+</div> -->

docs/source/api/summary.md

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

@@ -0,0 +1,275 @@
+# SPDX-License-Identifier: Apache-2.0
+
+# Configuration file for the Sphinx documentation builder.
+#
+# This file only contains a selection of the most common options. For a full
+# list see the documentation:
+# https://www.sphinx-doc.org/en/master/usage/configuration.html
+
+# -- Path setup --------------------------------------------------------------
+
+# If extensions (or modules to document with autodoc) are in another directory,
+# add these directories to sys.path here. If the directory is relative to the
+# documentation root, use os.path.abspath to make it absolute, like shown here.
+
+import datetime
+import logging
+import os
+import re
+import sys
+from pathlib import Path
+
+import requests
+
+logger = logging.getLogger(__name__)
+REPO_ROOT = Path(__file__).resolve().parent.parent.parent
+print(os.path.abspath(REPO_ROOT))
+sys.path.append(os.path.abspath(REPO_ROOT))
+
+# -- Project information -----------------------------------------------------
+
+project = 'FastVideo'
+copyright = f'{datetime.datetime.now().year}, FastVideo Team'
+author = 'the FastVideo Team'
+
+# -- General configuration ---------------------------------------------------
+
+# Add any Sphinx extension module names here, as strings. They can be
+# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
+# ones.
+extensions = [
+    "sphinx.ext.napoleon",
+    "sphinx.ext.linkcode",
+    "sphinx.ext.intersphinx",
+    "sphinx_copybutton",
+    "autodoc2",
+    "myst_parser",
+    "sphinxarg.ext",
+    "sphinx_design",
+    "sphinx_togglebutton",
+]
+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.
+templates_path = ['_templates']
+
+# List of patterns, relative to source directory, that match files and
+# directories to ignore when looking for source files.
+# This pattern also affects html_static_path and html_extra_path.
+exclude_patterns: list[str] = ["**/*.template.md", "**/*.inc.md"]
+
+# Exclude the prompt "$" when copying code
+copybutton_prompt_text = r"\$ "
+copybutton_prompt_is_regexp = True
+
+# -- Options for HTML output -------------------------------------------------
+
+# The theme to use for HTML and HTML Help pages.  See the documentation for
+# a list of builtin themes.
+#
+html_title = project
+html_theme = 'sphinx_book_theme'
+html_logo = '../../assets/logos/icon_simple.svg'
+html_theme_options = {
+    'path_to_docs': 'docs/source',
+    '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,
+# so a file named "default.css" will overwrite the builtin "default.css".
+html_static_path = ["_static"]
+html_js_files = ["custom.js"]
+html_css_files = ["custom.css"]
+
+myst_url_schemes = {
+    'http': None,
+    'https': None,
+    'mailto': None,
+    'ftp': None,
+    "gh-issue": {
+        "url":
+        "https://github.com/hao-ai-lab/FastVideo/issues/{{path}}#{{fragment}}",
+        "title": "Issue #{{path}}",
+        "classes": ["github"],
+    },
+    "gh-pr": {
+        "url":
+        "https://github.com/hao-ai-lab/FastVideo/pull/{{path}}#{{fragment}}",
+        "title": "Pull Request #{{path}}",
+        "classes": ["github"],
+    },
+    "gh-dir": {
+        "url": "https://github.com/hao-ai-lab/FastVideo/tree/main/{{path}}",
+        "title": "{{path}}",
+        "classes": ["github"],
+    },
+    "gh-file": {
+        "url": "https://github.com/hao-ai-lab/FastVideo/blob/main/{{path}}",
+        "title": "{{path}}",
+        "classes": ["github"],
+    },
+}
+
+# see https://docs.readthedocs.io/en/stable/reference/environment-variables.html # noqa
+READTHEDOCS_VERSION_TYPE = os.environ.get('READTHEDOCS_VERSION_TYPE')
+if READTHEDOCS_VERSION_TYPE == "tag":
+    # remove the warning banner if the version is a tagged release
+    header_file = os.path.join(os.path.dirname(__file__),
+                               "_templates/sections/header.html")
+    # The file might be removed already if the build is triggered multiple times
+    # (readthedocs build both HTML and PDF versions separately)
+    if os.path.exists(header_file):
+        os.remove(header_file)
+
+
+# Generate additional rst documentation here.
+def setup(app):
+    from docs.source.generate_examples import generate_examples
+    generate_examples()
+
+
+_cached_base: str = ""
+_cached_branch: str = ""
+
+
+def get_repo_base_and_branch(pr_number: str) -> tuple[str | None, str | None]:
+    global _cached_base, _cached_branch
+    if _cached_base and _cached_branch:
+        return _cached_base, _cached_branch
+
+    url = f"https://api.github.com/repos/hao-ai-lab/FastVideo/pulls/{pr_number}"
+    response = requests.get(url)
+    if response.status_code == 200:
+        data = response.json()
+        _cached_base = data['head']['repo']['full_name']
+        _cached_branch = data['head']['ref']
+        return _cached_base, _cached_branch
+    else:
+        logger.error("Failed to fetch PR details: %s", response)
+        return None, None
+
+
+def linkcode_resolve(domain, info):
+    if domain != 'py':
+        return None
+    if not info['module']:
+        return None
+
+    # 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 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 the line number is not found, return None
+    if lineno == len(lines):
+        return None
+
+    # 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 = 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}"
+
+
+# Mock out external dependencies here, otherwise the autodoc pages may be blank.
+autodoc_mock_imports = [
+    "blake3",
+    "compressed_tensors",
+    "cpuinfo",
+    "cv2",
+    "torch",
+    "huggingface_hub",
+    "torchvision",
+    "transformers",
+    "psutil",
+    "prometheus_client",
+    "sentencepiece",
+    "vllm._C",
+    "PIL",
+    "numpy",
+    'triton',
+    "tqdm",
+    "tensorizer",
+    "pynvml",
+    "outlines",
+    "xgrammar",
+    "librosa",
+    "soundfile",
+    "gguf",
+    "lark",
+    "decord",
+]
+
+for mock_target in autodoc_mock_imports:
+    if mock_target in sys.modules:
+        logger.info(
+            "Potentially problematic mock target (%s) found; "
+            "autodoc_mock_imports cannot mock modules that have already "
+            "been loaded into sys.modules when the sphinx build starts.",
+            mock_target)
+
+intersphinx_mapping = {
+    "python": ("https://docs.python.org/3", None),
+    "typing_extensions":
+    ("https://typing-extensions.readthedocs.io/en/latest", None),
+    "aiohttp": ("https://docs.aiohttp.org/en/stable", None),
+    "pillow": ("https://pillow.readthedocs.io/en/stable", None),
+    "numpy": ("https://numpy.org/doc/stable", None),
+    "torch": ("https://pytorch.org/docs/stable", None),
+    "psutil": ("https://psutil.readthedocs.io/en/stable", None),
+}
+
+navigation_with_keys = False

docs/source/contributing/developer_env/docker.md

@@ -1,4 +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/developer_env/index.md

@@ -3,3 +3,11 @@
 # 🧰 Developer Environment
 
 Accelerate your FastVideo development workflow by leveraging Docker images and cloud GPUs for efficient experimentation and reproducible environments.
+
+:::{toctree}
+:caption: Contents
+:maxdepth: 1
+
+docker
+runpod
+:::

docs/source/contributing/developer_env/runpod.md

@@ -1,3 +1,4 @@
+(runpod)=
 
 # 📦 Developing FastVideo on RunPod
 
@@ -9,7 +10,7 @@ Choose a GPU that supports CUDA 12.8
 
 Pick 1 or 2 L40S GPU(s)
 
-![RunPod CUDA selection](../../assets/images/runpod_cuda.png)
+![RunPod CUDA selection](../../_static/images/runpod_cuda.png)
 
 When creating your pod template, use this image:
 
@@ -23,11 +24,11 @@ Paste Container Start Command to support SSH ([RunPod Docs](https://docs.runpod.
 bash -c "apt update;DEBIAN_FRONTEND=noninteractive apt-get install openssh-server -y;mkdir -p ~/.ssh;cd $_;chmod 700 ~/.ssh;echo \"$PUBLIC_KEY\" >> authorized_keys;chmod 700 authorized_keys;service ssh start;sleep infinity"
 ```
 
-![RunPod template configuration](../../assets/images/runpod_template.png)
+![RunPod template configuration](../../_static/images/runpod_template.png)
 
 After deploying, the pod will take a few minutes to pull the image and start the SSH service.
 
-![RunPod ssh](../../assets/images/runpod_ssh.png)
+![RunPod ssh](../../_static/images/runpod_ssh.png)
 
 ## Working with the pod
 

docs/source/contributing/overview.md

@@ -1,3 +1,4 @@
+(developer-overview)=
 
 # 🛠️ Contributing to FastVideo
 

docs/source/design/overview.md

@@ -29,6 +29,7 @@ FastVideo separates model components from execution logic with these principles:
 - **Custom Attention Backends**: Components can support and use different Attention implementations
 - **Pipeline Abstraction**: Consistent interface across diffusion models
 
+(design-fastvideo-args)=
 ## FastVideoArgs
 
 The `FastVideoArgs` class in `fastvideo/fastvideo_args.py` serves as the central configuration system for FastVideo. It contains all parameters needed to control model loading, inference configuration, performance optimization settings, and more.
@@ -60,6 +61,7 @@ with set_current_fastvideo_args(fastvideo_args):
     result = generate_video()
 ```
 
+(design-pipeline-system)=
 ## Pipeline System
 
 ### `ComposedPipelineBase`
@@ -106,8 +108,7 @@ def forward(self, batch: ForwardBatch, fastvideo_args: FastVideoArgs) -> Forward
     return batch
 ```
 
-![Pipeline execution and data flow](../assets/images/pipeline.png)
-
+(design-forwardbatch)=
 ### ForwardBatch
 
 Defined in `fastvideo/pipelines/pipeline_batch_info.py`, `ForwardBatch` encapsulates the data payload passed between pipeline stages. It typically holds:
@@ -119,10 +120,12 @@ Defined in `fastvideo/pipelines/pipeline_batch_info.py`, `ForwardBatch` encapsul
 
 This structure facilitates clear state transitions between stages.
 
+(design-model-components)=
 ## Model Components
 
 The `fastvideo/models/` directory contains implementations of the core neural network models used in video diffusion:
 
+(design-transformer-models)=
 ### Transformer Models
 
 Transformer networks perform the actual denoising during diffusion:
@@ -149,6 +152,7 @@ def forward(
     return noise_pred  # Predicted noise residual
 ```
 
+(design-vae-variational-auto-encoder)=
 ### VAE (Variational Auto-Encoder)
 
 VAEs handle conversion between pixel space and latent space:
@@ -166,6 +170,7 @@ FastVideo's VAE implementations include:
 - Optional tiling for large frames
 - Distributed weight support
 
+(design-text-and-image-encoders)=
 ### Text and Image Encoders
 
 Encoders process conditioning inputs into embeddings:
@@ -183,6 +188,7 @@ FastVideo implements optimizations such as:
 - Caching for common prompts
 - Precision-tuned computation
 
+(design-schedulers)=
 ### Schedulers
 
 Schedulers manage the diffusion sampling process:
@@ -210,10 +216,7 @@ def step(
     return prev_sample
 ```
 
-This diagram shows how models are discovered, validated, and loaded across entrypoints, executors, pipelines, and model loaders.
-
-![Model loading flow](../assets/images/load_models.png)
-
+(design-optimized-attention)=
 ## Optimized Attention
 
 The `fastvideo/attention/` directory contains optimized attention implementations crucial for efficient video diffusion:
@@ -237,17 +240,17 @@ self.attn = LocalAttention(
 # export FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
 ```
 
-![Attention backend selector design](../assets/images/attention_backend.png)
-
 ### Attention Patterns
 Supports various patterns with memory optimization techniques:
 - **Cross/Self/Temporal/Global-Local Attention**
 - Chunking, progressive computation, optimized masking
 
+(design-distributed-processing)=
 ## Distributed Processing
 
 The `fastvideo/distributed/` directory contains implementations for distributed model execution:
 
+(design-tensor-parallelism)=
 ### Tensor Parallelism
 
 Tensor parallelism splits model weights across devices:
@@ -304,6 +307,7 @@ Efficient communication primitives minimize distributed overhead:
 - **Tensor-Parallel AllReduce**: Combines partial results
 - **Distributed Synchronization**: Coordinates execution
 
+(design-forwardcontext)=
 ## Forward Context Management
 
 ### ForwardContext
@@ -326,6 +330,7 @@ with set_forward_context(current_timestep, attn_metadata, fastvideo_args):
     output = model(inputs)
 ```
 
+(design-executor-and-worker-abstractions)=
 ## Executor and Worker System
 
 The `fastvideo/worker/` directory contains the distributed execution framework:
@@ -352,6 +357,7 @@ Each GPU worker:
 
 This design allows FastVideo to efficiently utilize multiple GPUs while providing a simple, unified interface for model execution.
 
+(design-platforms)=
 ## Platforms
 
 The `fastvideo/platforms/` directory provides hardware platform abstractions that enable FastVideo to run efficiently on different hardware configurations:
@@ -382,6 +388,7 @@ else:
 
 The platform system is designed to be extensible for future hardware targets.
 
+(design-logger)=
 ## Logger
 See [PR](https://github.com/hao-ai-lab/FastVideo/pull/356)
 

docs/source/distillation/data_preprocess.md

@@ -1,3 +1,4 @@
+(v0-data-preprocess)=
 
 # 🧱 Data Preprocess for Distillation
 

docs/source/generate_examples.py

@@ -6,11 +6,10 @@ import re
 from dataclasses import dataclass, field
 from pathlib import Path
 
-ROOT_DIR = Path(__file__).parent.parent.resolve()
-ROOT_DIR_RELATIVE = '../..'
+ROOT_DIR = Path(__file__).parent.parent.parent.resolve()
+ROOT_DIR_RELATIVE = '../../../..'
 EXAMPLE_DIR = ROOT_DIR / "examples"
-EXAMPLE_DOC_DIR = ROOT_DIR / "docs/getting_started/examples"
-GITHUB_REPO = "hao-ai-lab/FastVideo"  # Update this to your repo
+EXAMPLE_DOC_DIR = ROOT_DIR / "docs/source/getting_started/examples"
 
 
 def fix_case(text: str) -> str:
@@ -72,16 +71,9 @@ class Index:
 
     def generate(self) -> str:
         content = f"# {self.title}\n\n{self.description}\n\n"
-        if self.caption:
-            content += f"## {self.caption}\n\n"
-        # Generate a simple list of links for MkDocs
-        for doc in self.documents:
-            # Convert document path to proper link
-            doc_link = doc.replace("\\", "/")
-            # Get just the filename for the link text
-            doc_title = fix_case(Path(doc).stem.replace("_", " ").title())
-            content += f"- [{doc_title}]({doc_link}.md)\n"
-        content += "\n"
+        content += ":::{toctree}\n"
+        content += f":caption: {self.caption}\n:maxdepth: {self.maxdepth}\n"
+        content += "\n".join(self.documents) + "\n:::\n"
         return content
 
 
@@ -150,66 +142,30 @@ class Example:
         return fix_case(self.path.stem.replace("_", " ").title())
 
     def generate(self) -> str:
-        # Create GitHub link to source
-        github_path = str(self.path.relative_to(ROOT_DIR)).replace("\\", "/")
-        github_url = f"https://github.com/{GITHUB_REPO}/blob/main/{github_path}"
-        content = f"**Source:** [{github_path}]({github_url})\n\n"
+        # Convert the path to a relative path from __file__
+        make_relative = lambda path: ROOT_DIR_RELATIVE / path.relative_to(
+            ROOT_DIR)
 
-        # Add title for code files
-        if self.main_file.suffix != ".md":
+        content = f"Source <gh-file:{self.path.relative_to(ROOT_DIR)}>.\n\n"
+        include = "include" if self.main_file.suffix == ".md" else \
+            "literalinclude"
+        if include == "literalinclude":
             content += f"# {self.title}\n\n"
-
-        # Include main file content
-        if self.main_file.suffix == ".md":
-            # For markdown files, include the content directly
-            with open(self.main_file, encoding='utf-8') as f:
-                content += f.read() + "\n\n"
-        else:
-            # For code files, use code blocks
-            language = self.main_file.suffix[1:] if self.main_file.suffix else ""
-            with open(self.main_file, encoding='utf-8') as f:
-                file_content = f.read()
-            content += f"```{language}\n{file_content}\n```\n\n"
+        content += f":::{{{include}}} {make_relative(self.main_file)}\n"  # type: ignore[no-untyped-call]
+        if include == "literalinclude":
+            content += f":language: {self.main_file.suffix[1:]}\n"
+        content += ":::\n\n"
 
         if not self.other_files:
             return content
 
-        content += "## Additional Files\n\n"
-        # Define binary/non-text file extensions to skip
-        binary_extensions = {
-            '.mp4', '.avi', '.mov', '.mkv', '.gif', '.jpg', '.jpeg', '.png',
-            '.webp', '.bmp', '.pdf', '.zip', '.tar', '.gz', '.mp3', '.wav'
-        }
-
+        content += "## Example materials\n\n"
         for file in sorted(self.other_files):
-            # Skip binary files
-            if file.suffix.lower() in binary_extensions:
-                continue
-
-            file_rel_path = file.relative_to(self.path)
-            # Use collapsible admonition syntax for MkDocs
-            content += f"??? note \"{file_rel_path}\"\n\n"
-
-            try:
-                if file.suffix == ".md":
-                    # Include markdown content with indentation
-                    with open(file, encoding='utf-8') as f:
-                        for line in f:
-                            content += f"    {line}"
-                else:
-                    # Include code with proper formatting
-                    language = file.suffix[1:] if file.suffix else ""
-                    with open(file, encoding='utf-8') as f:
-                        file_content = f.read()
-                    # Indent the code block for the admonition
-                    content += f"    ```{language}\n"
-                    for line in file_content.split('\n'):
-                        content += f"    {line}\n"
-                    content += "    ```\n"
-                content += "\n"
-            except UnicodeDecodeError:
-                # Skip files that can't be decoded as UTF-8
-                continue
+            include = "include" if file.suffix == ".md" else "literalinclude"
+            content += f":::{{admonition}} {file.relative_to(self.path)}\n"
+            content += ":class: dropdown\n\n"
+            content += f":::{{{include}}} {make_relative(file)}\n:::\n"  # type: ignore[no-untyped-call]
+            content += ":::\n\n"
 
         return content
 
@@ -239,7 +195,7 @@ class NestedStructure:
 
 def create_category_indices() -> dict[str, Index]:
     """Create category indices with their respective configurations."""
-    main_index_dir = ROOT_DIR / "docs/examples"
+    main_index_dir = ROOT_DIR / "docs/source/examples"
     if not main_index_dir.exists():
         main_index_dir.mkdir(parents=True)
 
@@ -247,16 +203,17 @@ def create_category_indices() -> dict[str, Index]:
         "inference":
         Index(
             path=ROOT_DIR /
-            "docs/inference/examples/examples_inference_index.md",
+            "docs/source/inference/examples/examples_inference_index.md",
             title="🚀 Examples",
             description=
-            "Inference examples demonstrate how to use FastVideo inference. We recommend starting with [basic.md](basic.md).",
+            "Inference examples demonstrate how to use FastVideo inference. We recommend starting with <project:basic.md>.",
             caption="Examples",
             maxdepth=1,
         ),
         "training":
         Index(
-            path=ROOT_DIR / "docs/training/examples/examples_training_index.md",
+            path=ROOT_DIR /
+            "docs/source/training/examples/examples_training_index.md",
             title="🚀 Examples",
             description=
             "Training examples demonstrate how to use FastVideo training.",
@@ -266,7 +223,7 @@ def create_category_indices() -> dict[str, Index]:
         "distillation":
         Index(
             path=ROOT_DIR /
-            "docs/distillation/examples/examples_distillation_index.md",
+            "docs/source/distillation/examples/examples_distillation_index.md",
             title="🚀 Examples",
             description=
             "Distillation examples demonstrate how to use FastVideo distillation.",
@@ -289,21 +246,9 @@ def find_examples(category_indices: dict[str, Index],
     examples = []
     glob_patterns = ["*.py", "*.md", "*.sh"]
 
-    # Map category names to actual directory names
-    category_dir_mapping = {
-        "distillation": "distill",  # examples/distill/ -> distillation category
-    }
-
     # Find categorised examples
     for category in category_indices:
-        # Use mapped directory name if available, otherwise use category name
-        dir_name = category_dir_mapping.get(category, category)
-        category_dir = EXAMPLE_DIR / dir_name
-
-        # Skip if directory doesn't exist
-        if not category_dir.exists():
-            continue
-
+        category_dir = EXAMPLE_DIR / category
         globs = [category_dir.glob(pattern) for pattern in glob_patterns]
         for path in itertools.chain(*globs):
             examples.append(Example(path, category))
@@ -334,18 +279,11 @@ def create_nested_structures(
                                                 dict[str,
                                                      NestedStructure]]]] = {}
 
-    # Map category names to actual directory names
-    category_dir_mapping = {
-        "distillation": "distill",
-    }
-
     for example in examples:
         if example.category not in ["training", "distillation"]:
             continue
 
-        # Use mapped directory name if available
-        dir_name = category_dir_mapping.get(example.category, example.category)
-        category_dir = EXAMPLE_DIR / dir_name
+        category_dir = EXAMPLE_DIR / example.category
         relative_path = example.path.relative_to(category_dir)
         path_parts = relative_path.parts
 
@@ -477,7 +415,7 @@ def generate_nested_examples(nested_structures: dict[str, dict[str, dict[
             category_index.documents.append(method)
 
 
-def generate_examples(generate_main_index: bool = False) -> None:
+def generate_examples(generate_main_index=False):
     """
     Generate example documentation.
     
@@ -491,14 +429,12 @@ def generate_examples(generate_main_index: bool = False) -> None:
     # Create the main examples index only if requested
     examples_index = None
     if generate_main_index:
-        main_index_dir = ROOT_DIR / "docs/examples"
+        main_index_dir = ROOT_DIR / "docs/source/examples"
         examples_index = Index(
             path=main_index_dir / "examples_index.md",
             title="💡 Examples",
             description=
-            "A collection of examples demonstrating usage of FastVideo.\n\n"
-            f"All documented examples are autogenerated using [generate_examples.py](https://github.com/{GITHUB_REPO}/blob/main/docs/generate_examples.py) "
-            f"from examples found in the [examples](https://github.com/{GITHUB_REPO}/tree/main/examples) directory.",
+            "A collection of examples demonstrating usage of FastVideo.\nAll documented examples are autogenerated using <gh-file:docs/source/generate_examples.py> from examples found in <gh-file:examples>.",
             caption="Examples",
             maxdepth=2)
 
@@ -535,19 +471,3 @@ def generate_examples(generate_main_index: bool = False) -> None:
     if generate_main_index and examples_index:
         with open(examples_index.path, "w+") as f:
             f.write(examples_index.generate())
-
-
-def on_pre_build_hook(config, **kwargs):
-    """
-    MkDocs hook to generate examples before building the documentation.
-    This function is called automatically by the mkdocs-simple-hooks plugin.
-    """
-    print("Generating example documentation...")
-    generate_examples(generate_main_index=True)
-    print("Example documentation generated successfully!")
-
-
-if __name__ == "__main__":
-    print("Generating example documentation...")
-    generate_examples(generate_main_index=True)
-    print("Example documentation generated successfully!")

docs/source/getting_started/installation.md

@@ -0,0 +1,18 @@
+(installation-index)=
+
+# 🔧 Installation
+
+FastVideo supports the following hardware platforms:
+
+:::{toctree}
+:maxdepth: 1
+:hidden:
+
+installation/gpu
+installation/mps
+:::
+
+- <project:installation/gpu.md>
+  - NVIDIA CUDA
+- <project:installation/mps.md>
+  - Apple silicon

docs/source/getting_started/installation/gpu.md

@@ -30,8 +30,17 @@ conda create -n fastvideo python=3.12 -y
 conda activate fastvideo
 ```
 
+:::{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.
+Note that you can also use `uv` to install FastVideo in a Conda environment.
+:::
+
 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

docs/source/getting_started/installation/mps.md

@@ -31,8 +31,17 @@ conda create -n fastvideo python=3.12.4 -y
 conda activate fastvideo
 ```
 
+:::{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.
+Note that you can also use `uv` to install FastVideo in a Conda environment.
+:::
+
 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

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.fastvideo_args.FastVideoArgs, executor_class: type[fastvideo.worker.executor.Executor], log_stats: bool)
+:canonical: fastvideo.entrypoints.video_generator.VideoGenerator
+
+```{autodoc2-docstring} fastvideo.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.configs.pipelines.PipelineConfig]] = None, **kwargs) -> fastvideo.entrypoints.video_generator.VideoGenerator
+:canonical: fastvideo.entrypoints.video_generator.VideoGenerator.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.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.configs.pipelines.base.PipelineConfig
+
+```{autodoc2-docstring} fastvideo.configs.pipelines.base.PipelineConfig
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+````{py:method} from_pretrained(model_path: str) -> fastvideo.configs.pipelines.base.PipelineConfig
+:canonical: fastvideo.configs.pipelines.base.PipelineConfig.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.configs.pipelines.base.PipelineConfig.from_pretrained
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+
+````{py:method} dump_to_json(file_path: str)
+:canonical: fastvideo.configs.pipelines.base.PipelineConfig.dump_to_json
+
+```{autodoc2-docstring} fastvideo.configs.pipelines.base.PipelineConfig.dump_to_json
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+
+### SamplingParam
+
+```{autodoc2-summary}
+    fastvideo.SamplingParam
+```
+
+`````{py:class} SamplingParam
+:canonical: fastvideo.configs.sample.base.SamplingParam
+
+```{autodoc2-docstring} fastvideo.configs.sample.base.SamplingParam
+:parser: docs.source.autodoc2_docstring_parser
+```
+
+````{py:method} from_pretrained(model_path: str) -> fastvideo.configs.sample.base.SamplingParam
+:canonical: fastvideo.configs.sample.base.SamplingParam.from_pretrained
+:classmethod:
+
+```{autodoc2-docstring} fastvideo.configs.sample.base.SamplingParam.from_pretrained
+:parser: docs.source.autodoc2_docstring_parser
+```

docs/source/index.md

@@ -1,24 +1,30 @@
 # Welcome to FastVideo
 
-<div style="text-align: center;">
-  <img src="assets/logos/logo.svg" alt="FastVideo" style="width: 60%;" />
-</div>
+:::{figure} ../../assets/logos/logo.svg
+:align: center
+:alt: FastVideo
+:class: no-scaled-link
+:width: 60%
+:::
 
-<div style="text-align: center;">
-<strong>FastVideo is a unified inference and post-training framework for accelerated video generation.</strong>
-</div>
+:::{raw} html
+<p style="text-align:center">
+<strong>FastVideo is a unified inference and post-training framework for accelerated video generation.
+</strong>
+</p>
 
-<div style="text-align: center;">
+<p style="text-align:center">
 <script async defer src="https://buttons.github.io/buttons.js"></script>
 <a class="github-button" href="https://github.com/hao-ai-lab/FastVideo/" data-show-count="true" data-size="large" aria-label="Star">Star</a>
 <a class="github-button" href="https://github.com/hao-ai-lab/FastVideo/subscription" data-icon="octicon-eye" data-size="large" aria-label="Watch">Watch</a>
 <a class="github-button" href="https://github.com/hao-ai-lab/FastVideo/fork" data-icon="octicon-repo-forked" data-size="large" aria-label="Fork">Fork</a>
-</div>
+</p>
+:::
 
 FastVideo is an inference and post-training framework for diffusion models. It features an end-to-end unified pipeline for accelerating diffusion models, starting from data preprocessing to model training, finetuning, distillation, and inference. FastVideo is designed to be modular and extensible, allowing users to easily add new optimizations and techniques. Whether it is training-free optimizations or post-training optimizations, FastVideo has you covered.
 
 <div style="text-align: center;">
-  <img src="assets/images/fastwan.png" style="width: 100%;"/>
+  <img src=_static/images/fastwan.png width="100%"/>
 </div>
 
 ## Key Features
@@ -36,16 +42,90 @@ FastVideo has the following features:
 
 ## Documentation
 
-Welcome to FastVideo! This documentation will help you get started with our unified inference and post-training framework for accelerated video generation.
+% How to start using FastVideo?
 
-Use the navigation menu on the left to explore different sections:
+:::{toctree}
+:caption: Getting Started
+:maxdepth: 1
 
-- **Getting Started**: Installation and quick start guides
-- **Inference**: Learn how to use FastVideo for video generation
-- **Training**: Data preprocessing and fine-tuning workflows  
-- **Distillation**: Post-training optimization techniques
-- **Sliding Tile Attention**: Advanced attention mechanisms
-- **Video Sparse Attention**: Efficient attention for video models
-- **Design**: Framework architecture and design principles
-- **Developer Guide**: Contributing and development setup
-- **API Reference**: Complete API documentation
+getting_started/installation
+<!-- getting_started/v1_api -->
+:::
+
+:::{toctree}
+:caption: Inference
+:maxdepth: 1
+
+inference/inference_quick_start
+inference/examples/examples_inference_index
+inference/configuration
+inference/optimizations
+inference/comfyui
+inference/support_matrix
+inference/cli
+inference/add_pipeline
+:::
+
+:::{toctree}
+:caption: Training
+:maxdepth: 1
+
+training/examples/examples_training_index
+training/data_preprocess
+<!-- training/finetune -->
+:::
+
+:::{toctree}
+:caption: Distillation
+:maxdepth: 1
+
+distillation/examples/examples_distillation_index
+distillation/data_preprocess
+distillation/dmd
+:::
+
+% What is STA Kernel?
+
+:::{toctree}
+:caption: Sliding Tile Attention
+:maxdepth: 1
+
+sliding_tile_attention/installation
+sliding_tile_attention/demo
+:::
+
+% What is VSA Kernel?
+
+:::{toctree}
+:caption: Video Sparse Attention
+:maxdepth: 1
+
+video_sparse_attention/installation
+:::
+
+:::{toctree}
+:caption: Design
+:maxdepth: 1
+design/overview
+:::
+
+:::{toctree}
+:caption: Developer Guide
+:maxdepth: 2
+
+contributing/overview
+contributing/developer_env/index
+:::
+
+:::{toctree}
+:caption: API Reference
+:maxdepth: 2
+
+<!-- api/summary -->
+api/fastvideo/fastvideo
+:::
+
+## Indices and tables
+
+- {ref}`genindex`
+- {ref}`modindex`

docs/source/inference/add_pipeline.md

@@ -1,3 +1,4 @@
+(add-pipeline)=
 
 # 🏗️ Adding a New Pipeline
 

docs/source/inference/configuration.md

@@ -1,4 +1,4 @@
-
+(inference-configuration)=
 # Configuration
 
 ## Multi-GPU Setup

docs/source/inference/optimizations.md

@@ -1,37 +1,32 @@
-
+(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)
-  - [Sage Attention 3](#optimizations-sage3)
 
 - 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`
 - Video Sparse Attention: `FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN`
 - Sage Attention: `FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN`
-- Sage Attention 3: `FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN_THREE`
 
 ### 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
@@ -39,13 +34,13 @@ 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`**
@@ -62,12 +57,16 @@ And if using a Hopper+ GPU (ie H100), installing [Flash Attention 3](https://git
 git clone https://github.com/Dao-AILab/flash-attention.git && cd flash-attention
 
 cd hopper
-pip install ninja
+pip install ninja 
 python setup.py install
 ```
 
-### Sliding Tile Attention
+:::{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
@@ -76,8 +75,8 @@ pip install st_attn==0.0.4
 
 Please see [this page](#sta-installation) for more installation instructions.
 
+(optimizations-vsa)=
 ### Video Sparse Attention
-
 **`VIDEO_SPARSE_ATTN`**
 
 ```bash
@@ -87,40 +86,20 @@ python setup_vsa.py install
 
 Please see [this page](#vsa-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
+cd sageattention 
 python setup.py install  # or pip install -e .
 ```
 
-### Sage Attention 3
-
-**`SAGE_ATTN_THREE`**
-
-[SageAttention 3](https://huggingface.co/jt-zhang/SageAttention3) is an advanced attention mechanism that leverages FP4 quantization and Blackwell GPU Tensor Cores for significant performance improvements.
-
-#### Hardware Requirements
-
-- RTX5090
-
-#### Installation
-
-Note that Sage Attention 3 requires `python>=3.13`, `torch>=2.8.0`, `CUDA >=12.8`. If you are using `uv` and using `torch==2.8.0` make sure that `sentencepiece==0.2.1` in the pyproject.toml file.
-
-To use Sage Attention 3 in FastVideo, first get access to the SageAttention3 code, then move `sageattn/` and `setup.py` to the directory `fastvideo/attention/backends`, then install from using:
-
-```bash
-python setup.py install
-```
-
+(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?

docs/source/inference/support_matrix.md

@@ -0,0 +1,136 @@
+(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
+- ⭕ = Does not apply to this model
+
+## 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
+  * Video Sparse Attention (VSA)
+- * FastWan2.1 T2V 1.3B
+  * `FastVideo/FastWan2.1-T2V-1.3B-Diffusers`
+  * 480P
+  * ⭕
+  * ⭕
+  * ⭕
+  * ✅
+- * FastWan2.2 TI2V 5B Full Attn*
+  * `FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers`
+  * 720P
+  * ⭕
+  * ⭕
+  * ⭕
+  * ✅
+- * Wan2.2 TI2V 5B 
+  * `Wan-AI/Wan2.2-TI2V-5B-Diffusers`
+  * 720P
+  * ⭕
+  * ⭕
+  * ✅
+  * ⭕
+- * Wan2.2 T2V A14B
+  * `Wan-AI/Wan2.2-T2V-A14B-Diffusers`
+  * 480P<br>720P
+  * ❌
+  * ❌
+  * ✅
+  * ⭕
+- * Wan2.2 I2V A14B
+  * `Wan-AI/Wan2.2-I2V-A14B-Diffusers`
+  * 480P<br>720P
+  * ❌
+  * ❌
+  * ✅
+  * ⭕
+- * HunyuanVideo
+  * `hunyuanvideo-community/HunyuanVideo`
+  * 720px1280p<br>544px960p
+  * ❌
+  * ✅
+  * ✅
+  * ⭕
+- * FastHunyuan
+  * `FastVideo/FastHunyuan-diffusers`
+  * 720px1280p<br>544px960p
+  * ❌
+  * ✅
+  * ✅
+  * ⭕
+- * Wan2.1 T2V 1.3B
+  * `Wan-AI/Wan2.1-T2V-1.3B-Diffusers`
+  * 480P
+  * ✅
+  * ✅*
+  * ✅
+  * ⭕
+- * Wan2.1 T2V 14B
+  * `Wan-AI/Wan2.1-T2V-14B-Diffusers`
+  * 480P, 720P
+  * ✅
+  * ✅*
+  * ✅
+  * ⭕
+- * Wan2.1 I2V 480P
+  * `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers`
+  * 480P
+  * ✅
+  * ✅*
+  * ✅
+  * ⭕
+- * Wan2.1 I2V 720P
+  * `Wan-AI/Wan2.1-I2V-14B-720P-Diffusers`
+  * 720P
+  * ✅
+  * ✅
+  * ✅
+  * ⭕
+- * StepVideo T2V
+  * `FastVideo/stepvideo-t2v-diffusers`
+  * 768px768px204f<br>544px992px204f<br>544px992px136f
+  * ❌
+  * ❌
+  * ✅
+  * ⭕
+:::
+
+**Note**: Wan2.2 TI2V 5B has some quality issues when performing I2V generation. 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/sliding_tile_attention/demo.md

@@ -1,3 +1,4 @@
+(sta-demo)=
 
 # 🔍 Demo
 This is is a demo for 2D STA with window size (6,6) operating on a (10, 10) image.

docs/source/sliding_tile_attention/installation.md

@@ -1,3 +1,4 @@
+(sta-installation)=
 
 # 🔧 Installation
 You can install the Sliding Tile Attention package using

docs/source/training/data_preprocess.md

@@ -1,3 +1,4 @@
+(v0-data-preprocess)=
 
 # 🧱 Data Preprocess
 

docs/source/training/finetune.md

@@ -1,4 +1,4 @@
-
+(v0-finetune)=
 # 🧠 Finetune
 ## ⚡ Full Finetune
 Ensure your data is prepared and preprocessed in the format specified in [data_preprocess.md](#v0-data-preprocess). For convenience, we also provide a mochi preprocessed Black Myth Wukong data that can be downloaded directly:

docs/source/video_sparse_attention/installation.md

@@ -1,3 +1,4 @@
+(vsa-installation)=
 
 # 🔧 Installation
 You can install the Video Sparse Attention package using

examples/dataset/vidprom/README.md

@@ -1,9 +0,0 @@
-# VidProm Dataset
-
-From [Self-Forcing](https://github.com/gdhe17/Self-Forcing) repository.
-
-## Download the dataset
-
-```bash
-./download_dataset.sh
-```

examples/dataset/vidprom/download_dataset.sh

@@ -1,3 +0,0 @@
-#! /bin/bash
-
-huggingface-cli download gdhe17/Self-Forcing vidprom_filtered_extended.txt --local-dir prompts

examples/distill/SFWan2.1-T2V/distill_dmd_t2v_1.3B.sh

@@ -1,15 +1,21 @@
 #!/bin/bash
-#SBATCH --job-name=t2v
+#SBATCH --job-name=wl_t2v
 #SBATCH --partition=main
-#SBATCH --nodes=1
-#SBATCH --ntasks=1
+#SBATCH --nodes=4
+#SBATCH --ntasks=4
 #SBATCH --ntasks-per-node=1
-#SBATCH --gres=gpu:1
+#SBATCH --gres=gpu:8
 #SBATCH --cpus-per-task=128
 #SBATCH --mem=1440G
-#SBATCH --output=dmd_t2v_output/t2v_%j.out
-#SBATCH --error=dmd_t2v_output/t2v_%j.err
+#SBATCH --output=dmd_t2v_output/sf.out
+#SBATCH --error=dmd_t2v_output/sf.err
 #SBATCH --exclusive
+set -e -x
+
+
+# Environment Setup
+source ~/conda/miniconda/bin/activate
+conda activate will-fv
 
 # Basic Info
 export NCCL_P2P_DISABLE=1
@@ -18,71 +24,82 @@ export TORCH_NCCL_ENABLE_MONITORING=0
 export TRITON_CACHE_DIR=/tmp/triton_cache_${SLURM_PROCID}
 export MASTER_PORT=29503
 export TOKENIZERS_PARALLELISM=false
-export WANDB_API_KEY=your_wandb_api_key
+export WANDB_API_KEY="8d9f4b39abd68eb4e29f6fc010b7ee71a2207cde"
 export WANDB_BASE_URL="https://api.wandb.ai"
 export WANDB_MODE=online
 export FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
 
 # Configs
-NUM_GPUS=1
+NUM_GPUS=8
 
 # Model paths for Self-Forcing DMD distillation:
 GENERATOR_MODEL_PATH="wlsaidhi/SFWan2.1-T2V-1.3B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-14B-Diffusers"  # Teacher model
+REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"  # Teacher model
 FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"  # Critic model
 
-DATA_DIR=your_data_dir
-VALIDATION_DATASET_FILE=your_validation_data_dir
+# DATA_DIR="data/test-text-preprocessing/Node_0_GPU_1_File_1/combined_parquet_dataset/"
+DATA_DIR="/mnt/weka/home/hao.zhang/matthew/FastVideo/data/test-text-preprocessing"
+VALIDATION_DATASET_FILE="data/crush-smol-single_processed_t2v/validation.json"
+VALIDATION_DATASET_FILE="/mnt/weka/home/hao.zhang/wl/FastVideo/examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/validation_64.json"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
+# Training arguments
 training_args=(
-  --tracker_project_name SFwan_t2v_distill_self_forcing_dmd  
-  --output_dir your_output_dir
-  --max_train_steps 4000
+  --tracker_project_name SFwan_t2v_distill_self_forcing_dmd  # Updated for self-forcing DMD
+  --output_dir "/mnt/sharefs/users/hao.zhang/SFwan_t2v_finetune"
+  # --use_sf_wan
+  # --sf_ode_init_path "checkpoints/ode_init.pt"
+  --max_train_steps 6000
   --train_batch_size 1
   --train_sp_batch_size 1
   --gradient_accumulation_steps 1
   --num_latent_t 21
   --num_height 480
   --num_width 832
+  --num_frames 81  # Must be divisible by num_frame_per_block (81 % 3 = 0 ✓)
   --enable_gradient_checkpointing_type "full"
   --log_visualization
   --simulate_generator_forward
-  --num_frames 81
   --num_frame_per_block 3  # Frame generation block size for self-forcing
   --enable_gradient_masking
   --gradient_mask_last_n_frames 21
 )
 
+# Parallel arguments
 parallel_args=(
-  --num_gpus $NUM_GPUS # 64
+  --num_gpus 32 # 64
   --sp_size 1
   --tp_size 1
-  --hsdp_replicate_dim 1 # 64
-  --hsdp_shard_dim $NUM_GPUS
+  --hsdp_replicate_dim 32 # 64
+  --hsdp_shard_dim 1
 )
 
+# Model arguments
 model_args=(
   --model_path $GENERATOR_MODEL_PATH  # TODO: check if you can remove this in this script
   --pretrained_model_name_or_path $GENERATOR_MODEL_PATH
+  --generator_model_path $GENERATOR_MODEL_PATH
   --real_score_model_path $REAL_SCORE_MODEL_PATH
   --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
+# Dataset arguments
 dataset_args=(
   --data_path "$DATA_DIR"
   --dataloader_num_workers 4
 )
 
+# Validation arguments
 validation_args=(
   --log_validation
   --validation_dataset_file "$VALIDATION_DATASET_FILE"
-  --validation_steps 50
+  --validation_steps 10
   --validation_sampling_steps "4"
   --validation_guidance_scale "6.0" # not used for dmd inference
 )
 
+# Optimizer arguments
 optimizer_args=(
   --learning_rate 1e-5
   --mixed_precision "bf16"
@@ -93,6 +110,7 @@ optimizer_args=(
   --max_grad_norm 1.0
 )
 
+# Miscellaneous arguments
 miscellaneous_args=(
   --inference_mode False
   --checkpoints_total_limit 3
@@ -103,9 +121,11 @@ miscellaneous_args=(
   --use_ema True
   --ema_decay 0.99
   --ema_start_step 100
-  --init_weights_from_safetensors your_ode_init_weights_path
+  --init_weights_from_safetensors "/mnt/weka/home/hao.zhang/wl/Self-Forcing/diffusers_ode_init/model.safetensors"
+  # --init_weights_from_safetensors "/mnt/weka/home/hao.zhang/wl/FastVideo2/warp_vidprom_8b16k_test_warp_1e-5/checkpoint-2000/transformer/diffusion_pytorch_model.safetensors"
 )
 
+# Self-forcing DMD arguments
 dmd_args=(
   --dmd_denoising_steps '1000,750,500,250'
   --min_timestep_ratio 0.02
@@ -117,16 +137,20 @@ dmd_args=(
   --warp_denoising_step
 )
 
+# Self-forcing specific arguments
 self_forcing_args=(
   --independent_first_frame False  # Whether to treat first frame independently
-  --same_step_across_blocks True  # Whether to use same denoising step across all blocks
+  --same_step_across_blocks False  # Whether to use same denoising step across all blocks
   --last_step_only False  # Whether to only use the last denoising step
   --context_noise 0  # Amount of noise to add during context caching (0 = no noise)
+  --validate_cache_structure False  # Set to True for debugging KV cache issues
 )
 
-torchrun \
---nnodes 1 \
---master_port $MASTER_PORT \
+srun torchrun \
+--nnodes $SLURM_JOB_NUM_NODES \
+--node_rank $SLURM_PROCID \
+--rdzv_backend=c10d \
+--rdzv_endpoint="$MASTER_ADDR:$MASTER_PORT" \
 --nproc_per_node $NUM_GPUS \
     fastvideo/training/wan_self_forcing_distillation_pipeline.py \
     "${parallel_args[@]}" \
@@ -137,4 +161,4 @@ torchrun \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
     "${dmd_args[@]}" \
-    "${self_forcing_args[@]}"
+    "${self_forcing_args[@]}"

examples/distill/SFWan2.1-T2V/download_dataset.sh

@@ -1,3 +1,3 @@
 #!/bin/bash
 
-python scripts/huggingface/download_hf.py --repo_id "wlsaidhi/crush-smol-merged" --local_dir "data/crush-smol" --repo_type "dataset"
+python scripts/huggingface/download_hf.py --repo_id "wlsaidhi/crush-smol-merged" --local_dir "data/crush-smol" --repo_type "dataset"

examples/distill/SFWan2.1-T2V/preprocess_data.sh

@@ -21,4 +21,4 @@ torchrun --nproc_per_node=$GPU_NUM \
     --samples_per_file 8 \
     --flush_frequency 8 \
     --video_length_tolerance_range 5 \
-    --preprocess_task "t2v" 
+    --preprocess_task "t2v" 

examples/distill/SFWan2.2-A14B/distill_dmd.sh

@@ -1,157 +0,0 @@
-#!/bin/bash
-#SBATCH --job-name=t2v
-#SBATCH --partition=main
-#SBATCH --nodes=4
-#SBATCH --ntasks=4
-#SBATCH --ntasks-per-node=1
-#SBATCH --gres=gpu:8
-#SBATCH --cpus-per-task=128
-#SBATCH --mem=1440G
-#SBATCH --output=dmd_t2v_output/t2v_%j.out
-#SBATCH --error=dmd_t2v_output/t2v_%j.err
-#SBATCH --exclusive
-
-# Basic Info
-export NCCL_P2P_DISABLE=1
-export TORCH_NCCL_ENABLE_MONITORING=0
-export NCCL_DEBUG_SUBSYS=INIT,NET
-# different cache dir for different processes
-export TRITON_CACHE_DIR=/tmp/triton_cache_${SLURM_PROCID}
-export MASTER_PORT=29500
-export NODE_RANK=$SLURM_PROCID
-nodes=( $(scontrol show hostnames $SLURM_JOB_NODELIST) )
-export MASTER_ADDR=${nodes[0]}
-export TOKENIZERS_PARALLELISM=false
-export WANDB_API_KEY="2f25ad37933894dbf0966c838c0b8494987f9f2f"
-# export WANDB_API_KEY='your_wandb_api_key_here'
-export WANDB_BASE_URL="https://api.wandb.ai"
-export WANDB_MODE=online
-export FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
-
-# Configs
-NUM_GPUS=8
-
-# Model paths for Self-Forcing DMD distillation with Wan2.2:
-# GENERATOR_MODEL_PATH="Wan-AI/Wan2.2-T2V-A14B-Diffusers"  # Updated to Wan2.2
-# REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.2-T2V-A14B-Diffusers" # Teacher model
-# FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.2-T2V-A14B-Diffusers" # Critic model
-GENERATOR_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"  # Updated to Wan2.2
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"  # Teacher model
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers" # Critic model
-
-# DATA_DIR="data/test-text-preprocessing/Node_0_GPU_1_File_1/combined_parquet_dataset/"
-DATA_DIR="/mnt/weka/home/hao.zhang/matthew/FastVideo/data/test-text-preprocessing"
-# DATA_DIR=data/crush-smol_processed_t2v/combined_parquet_dataset
-# DATA_DIR="/mnt/sharefs/users/hao.zhang/Vchitect-2M/Wan-Syn-upload/latents_i2v/train/"
-# VALIDATION_DATASET_FILE="data/crush-smol-single_processed_t2v/validation.json"
-VALIDATION_DATASET_FILE="/mnt/weka/home/hao.zhang/wl/FastVideo/examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/validation_64.json"
-
-# export CUDA_VISIBLE_DEVICES=4,5
-# IP=[MASTER NODE IP]
-
-training_args=(
-  --tracker_project_name SFwan2.2_t2v_distill_self_forcing_dmd  # Updated for Wan2.2
-  --output_dir "/mnt/sharefs/users/hao.zhang/SFwan2.2_t2v_finetune"
-  --override_transformer_cls_name "CausalWanTransformer3DModel"
-  --max_train_steps 4000
-  --train_batch_size 1
-  --train_sp_batch_size 1
-  --gradient_accumulation_steps 1
-  --num_latent_t 21
-  --num_height 448  # Updated to match Wan2.2 config
-  --num_width 832   # Updated to match Wan2.2 config
-  --enable_gradient_checkpointing_type "full"
-  --simulate_generator_forward
-  # --log_visualization
-  --num_frames 81
-  --num_frame_per_block 3  # Frame generation block size for self-forcing
-  --enable_gradient_masking
-  --gradient_mask_last_n_frames 21
-  # --init_weights_from_safetensors /mnt/sharefs/users/hao.zhang/wl/models/sf_ode_init_wan22_checkpoints/high/3k/
-  # --init_weights_from_safetensors_2 /mnt/sharefs/users/hao.zhang/wl/models/sf_ode_init_wan22_checkpoints/low/3k/
-)
-
-parallel_args=(
-  --num_gpus 32 # 64
-  --sp_size 1
-  --tp_size 1
-  --hsdp_replicate_dim 1 # 64
-  --hsdp_shard_dim 32
-)
-
-model_args=(
-  --model_path $GENERATOR_MODEL_PATH
-  --pretrained_model_name_or_path $GENERATOR_MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
-)
-
-dataset_args=(
-  --data_path "$DATA_DIR"
-  --dataloader_num_workers 4
-)
-
-validation_args=(
-  --log_validation
-  --validation_dataset_file "$VALIDATION_DATASET_FILE"
-  --validation_steps 20
-  --validation_sampling_steps "4"
-  --validation_guidance_scale "6.0" # not used for dmd inference
-)
-
-optimizer_args=(
-  --learning_rate 1e-5
-  --mixed_precision "bf16"
-  --training_state_checkpointing_steps 500
-  --weight_only_checkpointing_steps 500
-  --weight_decay 0.01
-  --betas '0.0,0.999'
-  --max_grad_norm 1.0
-)
-
-miscellaneous_args=(
-  --inference_mode False
-  --checkpoints_total_limit 3
-  --training_cfg_rate 0.0
-  --dit_precision "fp32"
-  --flow_shift 5
-  --seed 1000
-  --use_ema True
-  --ema_decay 0.99
-  --ema_start_step 100
-)
-
-dmd_args=(
-  --dmd_denoising_steps '1000,750,500,250'
-  --min_timestep_ratio 0.02
-  --max_timestep_ratio 0.98
-  --dfake_gen_update_ratio 5
-  --real_score_guidance_scale 3.0
-  --fake_score_learning_rate 8e-6
-  --fake_score_betas '0.0,0.999'
-  --warp_denoising_step
-)
-
-self_forcing_args=(
-  --independent_first_frame False  # Whether to treat first frame independently
-  --same_step_across_blocks True  # Whether to use same denoising step across all blocks
-  --last_step_only False  # Whether to only use the last denoising step
-  --context_noise 0  # Amount of noise to add during context caching (0 = no noise)
-)
-
-srun torchrun \
---nnodes $SLURM_JOB_NUM_NODES \
---nproc_per_node $NUM_GPUS \
---node_rank $SLURM_PROCID \
---rdzv_backend=c10d \
---rdzv_endpoint="$MASTER_ADDR:$MASTER_PORT" \
-    fastvideo/training/wan_self_forcing_distillation_pipeline.py \
-    "${parallel_args[@]}" \
-    "${model_args[@]}" \
-    "${dataset_args[@]}" \
-    "${training_args[@]}" \
-    "${optimizer_args[@]}" \
-    "${validation_args[@]}" \
-    "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}" \
-    "${self_forcing_args[@]}"

examples/distill/Wan2.1-T2V/Wan-Syn-Data-480P/distill_dmd_VSA_t2v_1.3B.slurm

@@ -39,18 +39,15 @@ echo "NODE_RANK: $NODE_RANK"
 # Configs
 NUM_GPUS=8
 MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
 DATA_DIR=your_data_dir
 VALIDATION_DATASET_FILE=your_validation_dataset_file
-OUTPUT_DIR="checkpoints/wan_t2v_finetune"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
 # Training arguments
 training_args=(
   --tracker_project_name wan_t2v_distill_dmd_VSA
-  --output_dir $OUTPUT_DIR
+  --output_dir"checkpoints/wan_t2v_finetune"
   --max_train_steps 4000
   --train_batch_size 1
   --train_sp_batch_size 1
@@ -75,8 +72,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -96,7 +91,7 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 2e-6
+  --learning_rate 1e-5
   --mixed_precision "bf16"
   --training_state_checkpointing_steps 500
   --weight_only_checkpointing_steps 500
@@ -139,4 +134,4 @@ srun torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.1-T2V/Wan-Syn-Data-480P/distill_dmd_VSA_t2v_14B.slurm

@@ -39,18 +39,15 @@ echo "NODE_RANK: $NODE_RANK"
 # Configs
 NUM_GPUS=8
 MODEL_PATH="Wan-AI/Wan2.1-T2V-14B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-14B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-14B-Diffusers"
 DATA_DIR=your_data_dir
 VALIDATION_DATASET_FILE=your_validation_dataset_file
-OUTPUT_DIR="checkpoints/wan_t2v_finetune"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
 # Training arguments
 training_args=(
   --tracker_project_name wan_t2v_distill_dmd_VSA
-  --output_dir "$OUTPUT_DIR"
+  --output_dir "checkpoints/wan_t2v_finetune"
   --max_train_steps 4000
   --train_batch_size 1
   --train_sp_batch_size 1
@@ -75,8 +72,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -96,7 +91,7 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 2e-6
+  --learning_rate 1e-5
   --mixed_precision "bf16"
   --training_state_checkpointing_steps 500
   --weight_only_checkpointing_steps 500
@@ -139,4 +134,4 @@ srun torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.1-T2V/Wan-Syn-Data-480P/distill_dmd_t2v_1.3B.slurm

@@ -39,18 +39,15 @@ echo "NODE_RANK: $NODE_RANK"
 # Configs
 NUM_GPUS=8
 MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
 DATA_DIR=your_data_dir
 VALIDATION_DATASET_FILE=your_validation_dataset_file
-OUTPUT_DIR="checkpoints/wan_t2v_finetune"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
 # Training arguments
 training_args=(
   --tracker_project_name wan_t2v_distill_dmd
-  --output_dir "$OUTPUT_DIR"
+  --output_dir "checkpoints/wan_t2v_finetune"
   --max_train_steps 4000
   --train_batch_size 1
   --train_sp_batch_size 1
@@ -75,8 +72,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -96,7 +91,7 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 2e-6
+  --learning_rate 1e-5
   --mixed_precision "bf16"
   --training_state_checkpointing_steps 500
   --weight_only_checkpointing_steps 500
@@ -138,4 +133,4 @@ srun torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.1-T2V/Wan-Syn-Data-480P/download_dataset.sh

@@ -1,3 +1,3 @@
 #!/bin/bash
 
-python scripts/huggingface/download_hf.py --repo_id "FastVideo/Wan-Syn_77x448x832_600k" --local_dir "FastVideo/Wan-Syn_77x448x832_600k" --repo_type "dataset"
+python scripts/huggingface/download_hf.py --repo_id "FastVideo/Wan-Syn_77x448x832_600k" --local_dir "FastVideo/Wan-Syn_77x448x832_600k" --repo_type "dataset"

examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/distill_dmd_t2v_5B.sh

@@ -40,18 +40,15 @@ echo "NODE_RANK: $NODE_RANK"
 # Configs
 NUM_GPUS=8
 MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
 DATA_DIR=your_data_dir
 VALIDATION_DIR=your_validation_path  #(example:validation_64.json)
-OUTPUT_DIR="checkpoints/wan_t2v_finetune"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
 # Training arguments
 training_args=(
   --tracker_project_name Wan_distillation
-  --output_dir "$OUTPUT_DIR"
+  --output_dir "your_output_dir"
   --max_train_steps 4000
   --train_batch_size 1
   --train_sp_batch_size 1
@@ -76,8 +73,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -97,11 +92,11 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 4e-6
+  --learning_rate 2e-5
   --lr_scheduler "cosine_with_min_lr"
   --min_lr_ratio 0.5
   --lr_warmup_steps 100
-  --fake_score_learning_rate 2e-6
+  --fake_score_learning_rate 1e-5
   --fake_score_lr_scheduler "cosine_with_min_lr"
   --mixed_precision "bf16"
   --training_state_checkpointing_steps 500
@@ -146,4 +141,4 @@ srun torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/distill_dmd_t2v_5B_VSA.sh

@@ -40,8 +40,6 @@ echo "NODE_RANK: $NODE_RANK"
 # Configs
 NUM_GPUS=8
 MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
 DATA_DIR=your_data_dir
 VALIDATION_DIR=your_validation_path  #(example:validation_64.json)
 # export CUDA_VISIBLE_DEVICES=4,5
@@ -75,8 +73,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -96,11 +92,11 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 4e-6
+  --learning_rate 2e-5
   --lr_scheduler "cosine_with_min_lr"
   --min_lr_ratio 0.5
   --lr_warmup_steps 100
-  --fake_score_learning_rate 2e-6
+  --fake_score_learning_rate 1e-5
   --fake_score_lr_scheduler "cosine_with_min_lr"
   --mixed_precision "bf16"
   --training_state_checkpointing_steps 500
@@ -146,4 +142,4 @@ srun torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/validation_64.json

@@ -55,462 +55,6 @@
       "height": 448,
       "width": 832,
       "num_frames": 61
-    },
-    {
-      "caption": "In an industrial setting, a person leans casually against a railing, exuding a sense of confidence and composure. They are wearing a striking outfit, consisting of a vibrant, patterned jacket over a simple white crop top, creating a bold contrast. The atmosphere is infused with warm, ambient lighting that casts soft shadows on the concrete walls and metallic surfaces. Intricate wiring and pipes form an intricate backdrop, enhancing the urban aesthetic. Their relaxed posture and direct, engaging gaze suggest a sense of ease in this industrial environment. This scene encapsulates a blend of modern fashion and gritty, urban architecture, creating a visually compelling narrative.",
-      "video_path": "Fashion/mixkit-portrait-of-a-hipster-woman-walking-down-a-stairs-1297_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A man is energetically stretching in an open-air setting, surrounded by rows of vibrant red seats that suggest an amphitheater or outdoor venue. He wears a sleeveless black shirt layered with a hooded vest, emphasizing his athletic build as he engages in a warm-up routine. Behind him, the striking modern architecture of the building features geometric panels, with large sections of glass and overlapping metallic beams creating a dynamic backdrop. The scene captures the contrast between his focused movements and the static, bold design of the structure, while the surrounding greenery adds a touch of nature to the environment. The overall atmosphere is one of preparation and anticipation, with the man appearing determined and ready for an upcoming event or performance.",
-      "video_path": "Sport/mixkit-man-doing-arm-stretches-595_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A young woman is seated on the floor in front of a plush, beige tufted couch, fully engrossed in sorting through a stack of papers. Her dark hair falls loosely past her shoulders, and she wears a green plaid shirt, contributing to the casual yet focused atmosphere. She gently places the papers onto a small round white table, occasionally lifting individual sheets to examine them more closely. Her expression shifts subtly, reflecting concentration and contemplation as she processes the information on the pages. Two small, round nested tables hold her documents, along with a small plant in a gray pot, adding a touch of greenery to the scene. The background features a dark paneled wall, creating a contrasting backdrop for the light-colored furniture. The setting is tranquil and organized, the couch and tables arranged symmetrically, conveying a sense of harmony. A calculator rests on the smaller table, hinting at a task involving calculations or budgeting.",
-      "video_path": "Woman/mixkit-frustrated-woman-throws-paperwork-on-the-floor-4526_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A heavily rusted metal gate stands firmly locked, with two vertical bars joined by a thick, old chain that loops elegantly around them. The chain's texture is coarse and rugged, its surface reflecting varying shades of orange and brown, indicative of years exposed to the elements. At the heart of the chain, a black iron padlock, slightly worn yet imposing, secures the gate, its curves and edges smooth against the aged links. The gate's metalwork is outlined by a backdrop of soft, blurred greenery, suggesting a serene and isolated location beyond the barrier. Tall trees rise in the distance, their trunks and leaves creating a lush, forest-like setting that contrasts with the gate's severe rust. A pathway leads away from the gate, its surface uneven with patches of moss and weathered stone visible in the soft focus, inviting yet inaccessible. The ambiance is quiet and mysterious, with a sense of abandonment hanging subtly in the air, evoking curiosity about what lies beyond. Shadows play across the gate, cast by branches swaying gently in the breeze, adding to the dynamic interaction of light and texture. This scene, rich in detail and atmosphere, captures the viewer's imagination, evoking both the allure of the forbidden and the beauty of decay.",
-      "video_path": "forest/mixkit-rusty-fence-with-a-chain-of-a-property-in-nature-5294_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a serene and softly lit yoga studio, three individuals engage in a yoga session, each performing an upward-facing stretch. The central figure is a woman with shoulder-length brown hair, dressed in a light cropped top and green leggings, her posture reflecting grace and concentration. To her right, another participant, a woman in a purple outfit, mirrors the pose with equal poise. On her left, a person with a bun focuses intently, supported slightly by yoga blocks beneath their hands. The warm-colored wooden floor contrasts soothingly with the soft pastel mural on the back wall, featuring an abstract design and partial visage of a serene face. Natural light floods the space from a large window on the right, where lush greens peek through, adding an element of tranquility. In the corner of the room, a collection of meditation instruments, including a gong and a Buddha statue, subtly frame the peaceful setting. The mood is calm yet focused, as all three participants are deeply engaged in their practice. The scene combines elements of balance, harmony, and a shared journey towards mindfulness. This depiction captures the essence of a yoga session that blends personal growth with collective experience.",
-      "video_path": "People/mixkit-small-group-of-people-doing-yoga-together-43730_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the deep blue expanse of the ocean, two dolphins glide effortlessly, their sleek bodies reflecting the sunlight filtering through the water. The prominent shadows and caustics create a shimmering effect on their skin, capturing the beauty of their natural habitat. Each dolphin moves with a fluid grace, occasionally interacting with gentle nudges, showcasing their playful and social nature. The scene is vibrant and dynamic, with the clear blue background accentuating the dolphins' movements, making it an ideal subject for AI recreation.",
-      "video_path": "sea/mixkit-dolphins-underwater-4133_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, a young woman stands against a vibrant graffiti-covered wall, deeply engrossed in her smartphone. Her expression reflects a mix of focus and subtle satisfaction as she interacts with the screen. She wears a black floral-patterned top, which contrasts with the bright, abstract shapes and bold colors of the mural behind her. As she continues to engage with her phone, a series of like count notifications appear on the screen, indicating a growing online appreciation. The wall behind her features a striking mix of geometric and organic shapes, including swirls of teal, orange, and black, with large humanoid figures in a pop-art style. Her long, light-brown hair frames her face, adding a calm, composed aura amidst the lively backdrop. The video captures a blend of contemporary digital interaction and expressive urban art, creating a dynamic yet harmonious scene.",
-      "video_path": "Girl/mixkit-girl-looking-at-the-likes-in-her-post-4914_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A young mother and her baby sit comfortably on a bed, surrounded by an inviting, cozy atmosphere. The woman, wearing a sleeveless top and jeans, is gently engaging with the baby, who is dressed in an adorable animal-print onesie. The child is seated on the bed with colorful toys scattered around, including a plush toy and a board book. The warm glow from a hanging lamp casts a soft light on them, enhancing the serene environment. Pillows are propped up against the headboard, providing a cushioned backdrop as the mother leans slightly over to interact with the baby. A small bottle is visible beside her, suggesting a nurturing setting. Her hand gestures animatedly as she holds up a soft, white cushion with red and blue accents, likely stimulating the baby\u2019s curiosity. Their shared moment is filled with affection and joy, a perfect snapshot of familial bonding.",
-      "video_path": "Baby/mixkit-loving-mother-and-her-baby-playing-with-soft-toys-49966_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A young girl with long brown hair sits at a round wooden table, engrossed in working on her laptop. The laptop screen is a vivid green, suggesting a green screen effect is in use. To her left, a doll dressed in a yellow and white outfit is casually laid on top of some books, adding a playful and innocent touch to the scene. The setting is cozy, with sheer curtains in the background allowing soft natural light to spill into the room. The girl's posture and focused attention on the laptop suggest she is either playing a game or learning something new. This serene and domestic atmosphere is complemented by the slight blur of a dark couch in the foreground, framing the focused activity of the child.",
-      "video_path": "Girl/mixkit-little-girl-doing-homework-on-a-laptop-4757_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "An expansive view of a calm bay reveals a fleet of sailboats, each anchored in a regimented line stretching toward the horizon. The water is a serene blue, reflecting the soft hues of the early morning sky. A gentle breeze is indicated by the subtle ripples trailing behind the boats, while a single, larger vessel cuts a distinct path, leaving a graceful wake in its journey to the open sea. On one side, a cluster of modern high-rise buildings stands, contrasting against the natural simplicity of the water, suggesting a blend of urban and marine life. The distant shoreline is barely visible, softened by the atmospheric perspective, giving a sense of endless waters meeting the sky. The overall mood is peaceful and orderly, with the boats appearing almost as sentinels guarding the expanse of the tranquil bay.",
-      "video_path": "beach/mixkit-flying-backwards-over-the-sea-near-a-coast-50187_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, a person is standing in the center of a dark, featureless space, illuminated by a spotlight that emphasizes their presence. The individual is dressed in a traditional martial arts uniform, known as a gi, which is predominantly white with a black belt tied around the waist, indicating a high level of expertise. The background remains pitch black, creating a stark contrast with the brightly lit figure, ensuring complete focus on them. The person's expression is serious and focused, reflecting a deep sense of discipline and concentration. Their hands move gracefully, transitioning through various martial arts stances, demonstrating practiced skill and fluidity. The uniform's crisp fabric folds and subtly reflects the light, further highlighting each precise movement. Despite the simplicity of the environment, the scene is dynamic, with each motion capturing the essence of martial arts practice. The video effectively conveys a sense of calm strength and mastery, making it ideal for an AI to recreate with attention to posture, lighting, and attire.",
-      "video_path": "Sport/mixkit-karate-fighter-bowing-to-the-front-49706_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a dimly lit room bathed in a mix of neon purple and blue lights, a focused individual is seated in a gaming chair. She wears a white hoodie and large headphones with cat ears that glow softly, creating a striking silhouette. Her hands rest on a keyboard, typing swiftly as she concentrates intently on the screen in front of her. The atmosphere exudes a sense of intensity and immersion, with the soft-colored lighting enhancing the futuristic vibe. Her long hair cascades down her shoulders, adding a touch of elegance to the otherwise tech-centric setting. The overall scene captures the essence of a dedicated gamer deeply engaged in her virtual world.",
-      "video_path": "earth/mixkit-a-young-woman-wearing-headphones-with-rgb-lights-suddenly-gets-51621_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "Inside a dimly-lit bus, five individuals are seated along the rows of worn seats, each subtly illuminated by the colorful lights emanating from overhead. On the left, a woman sits with a relaxed posture, her curly hair accented by a patterned scarf, wearing a plaid outfit paired with bright neon socks. Next to her, a person clad in a denim jacket appears deep in thought, resting their head on a hand. Further back, another figure in a bucket hat and oversized yellow attire gazes across the aisle, evoking a sense of introspection. The atmosphere is enriched by the soft glow of red and green lights, bathing the bus interior in an almost surreal ambiance, creating a compelling tableau of urban life.",
-      "video_path": "Music/mixkit-conceptual-urban-fashion-42581_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "An aerial view captures two tennis players on a court, with one dressed in white on the left and another in red on the right. They are mid-game, each poised for action with rackets in hand, accentuated by their strategic positioning at opposite baselines. The court itself is a stark, deep blue, bordered by the vibrant green of the surrounding area, with a dark central net dividing the space. Long shadows stretch dramatically across the ground, suggesting a late afternoon setting. The subtly textured surface of the court contrasts with the crisp, white lines marking its boundaries and sections. This scene creates a vivid, balanced composition, highlighting both the competitive tension and serene atmosphere of the game.",
-      "video_path": "People/mixkit-two-people-playing-tennis-aerial-view-880_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a vibrant, dreamlike setting, a lone figure moves energetically against a backdrop of deep blue and purple hues, casting emotive shadows that ripple with dynamic motion. The figure, almost obscured by a smeared effect, suggests a rhythmic dance or a passionate performance, arms blurred as they sweep through colorful, streaked lighting. A neon glow accentuates their form, particularly highlighting the face which is abstractly illuminated in bursts of orange and red, suggesting intense emotional expression. The scene is dominated by two primary elements \u2013 the figure\u2019s motion and the dramatic lighting, creating a synergy of human emotion and visual spectacle. Swirling trails of light seem to intertwine with the figure, like a visual symphony of movement and color that floods the space. The lighting changes, casting intricate patterns on the figure and the surrounding space, giving the impression of a kaleidoscope in motion. Despite the blurred and abstract portrayal, there is a sense of focus conveyed through the figure\u2019s intent movements, akin to a conductor orchestrating a visual and auditory performance. The environment resonates with an electric energy, suggesting a seamless fusion of art and technology. As the visual drama unfolds, the scene invites viewers to lose themselves in the abstract dance and the play of vivid luminance.",
-      "video_path": "Music/mixkit-dancer-dancing-with-a-light-bar-in-his-hands-42221_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a brightly lit studio, a photographer wearing a denim jacket focuses intently, capturing shots with a professional camera. Facing him, a model stands gracefully, adjusting her long, flowing hair with delicate movements. The scene is characterized by strong contrasts; the model's soft pink attire and gentle gestures complement the rugged, precise demeanor of the photographer. Positioned against a minimalist backdrop, the pair work seamlessly, with the camera\u2019s lens pointed directly at the model, capturing her elegance. The soft, diffused lighting casts a gentle glow on both subjects, creating an airy and ethereal atmosphere perfect for a high-fashion photo shoot.",
-      "video_path": "Fashion/mixkit-professional-photo-session-with-a-young-female-model-41621_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The video showcases a serene, expansive landscape covered with a variety of trees dotting the hills. The hills gently slope across the frame, with patches of dry grass contrasting against the lush green foliage. Tall trees with dense canopies stand elegantly, casting soft shadows on the ground below. The sunlight bathes the entire scene, highlighting the varied textures of the leaves and terrain. Gaps between the trees reveal a narrow dirt path meandering through the hills, suggesting a sense of quiet solitude. The undulating hills extend into the distance, creating depth and a calming sense of vast space. The verdant hues of the leaves contrast with the earthy tones of the hills, enhancing the visual richness. In the background, a faint outline of distant hills can be seen, blurred softly by the atmospheric perspective. This tranquil setting could be efficiently recreated in a virtual environment by focusing on its layered composition, color palette, and natural textures.",
-      "video_path": "forest/mixkit-aerial-panorama-of-a-sunny-mountain-landscape-40846_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A bustling ski slope comes alive with skiers descending a pristine, snow-covered hill, surrounded by towering, snow-draped evergreens. Several figures stand atop the slope, silhouetted against a clear blue sky, preparing to embark on their ski run. The chair lift on the right continuously drops off eager adventurers, adding to the excitement at the hilltop. Each skier, clad in colorful winter gear, carves distinct paths into the textured snow as they weave their way down. The interplay of sunlight and shadows accentuates the myriad tracks etched into the slope, creating a dynamic visual rhythm. The scene captures a vibrant winter wonderland, full of action and the thrill of a perfect ski day.",
-      "video_path": "Car/mixkit-skiers-on-a-snowy-slope-3327_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The scene unfolds within a dimly lit bus, where three young individuals are seated, each absorbed in their unique world. To the left, a person with tied-back hair rests their head on their hand, dressed casually in a jacket and jeans, projecting a relaxed demeanor. Central to the frame is another individual, sitting upright with intense focus, donning a plaid blazer and oversize hoops, enhancing their confident presence. The muted green and red lighting casts an atmospheric glow, adding depth and intrigue to the setting. On the right, a person in a bucket hat and striped shirt leans back, appearing contemplative as they adjust their hat with a nonchalant gesture. The interplay of light and shadow highlights their expressions, creating an intimate and cinematic ambiance. Together, these figures form a cohesive tableau, capturing a moment of introspection amid a bustling yet serene urban environment.",
-      "video_path": "City/mixkit-three-models-posing-to-the-lens-while-on-board-a-42575_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A determined climber is scaling a massive rock face, showcasing exceptional strength and skill. The person, clad in a teal shirt and dark pants, climbs with precision, their movements measured and deliberate. They are secured by climbing gear, which includes ropes and a harness, emphasizing their commitment to safety. The rugged texture of the sandy-colored rock provides an imposing backdrop, adding drama and scale to the climb. In the distance, other large rock formations and sparse vegetation can be seen under a bright, overcast sky, contributing to the natural and adventurous atmosphere. The scene captures a moment of focus and challenge, highlighting the climber's tenacity and the breathtaking environment.",
-      "video_path": "Sport/mixkit-alpinist-climbing-a-huge-rock-in-a-desert-43306_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A woman stands confidently in front of a large array of solar panels, her navy blue jumpsuit contrasting against the lush green grass beneath her feet. Her expression is calm and focused, eyes facing directly ahead, suggesting a deep connection to the subject matter\u2014renewable energy. The sunlight bathes the scene in warm hues, casting gentle shadows and highlighting the geometric precision of the solar panels' grid-like structure. The background reveals a blend of nature and technology, as the panels are anchored on a grassy slope with foliage on the left side of the frame. This composition captures a harmonious blend of human innovation and environmental consciousness, accentuated by the serene outdoor setting.",
-      "video_path": "Business/mixkit-woman-standing-in-front-of-a-solar-panel-4880_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, two people are working at a wooden desk, using an iMac computer. One person, wearing a white knit sweater, is using the apple wireless mouse with their right hand, while their left hand rests on the sleek white keyboard. Their movements are smooth yet intentional, suggesting they are focused on a task on the computer screen. The monitor displays a well-organized array of files and folders, hinting at a task that involves detailed organization or detailed data navigation. The second person, only subtly visible, sits closely by and appears to observe or assist, creating a collaborative atmosphere. Their presence adds a quiet dynamic to the scene, as if they are ready to provide input or guidance. Sticky notes with handwritten notes are attached to the monitor\u2019s stand, adding a touch of personal organization amidst the digital workspace. The focus on the keyboard and mouse emphasizes a streamlined workflow, indicative of a productive work environment. The overall ambiance is calm and focuses on teamwork, technology, and efficient workspace management.",
-      "video_path": "People/mixkit-person-with-glasses-working-on-a-desktop-computer-3248_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A man stands in front of a modern glass facade, taking off a dark hoodie to reveal his gray tank top underneath. His arms are lifted high as he maneuvers the hoodie over his head, showcasing a fluid motion that conveys a sense of calm and routine. The lighting highlights the contours of his muscles, emphasizing a combination of strength and quiet determination. Behind him, the reflective surface of the glass panels provides a subtle backdrop, enhancing the focus on his focused and serene demeanor.",
-      "video_path": "Sport/mixkit-man-puts-on-sleeveless-hoodie-603_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The video displays a captivating dance of fiery orange flames against a stark black background, creating an intense visual contrast. The flames twist and intertwine, forming symmetrical, swirling patterns that expand and contract rhythmically across the frame. Each fiery tendril seems to be alive, moving with an almost hypnotic fluidity that captures the viewer's attention. The illumination from the flames casts subtle shadows, enhancing the depth and texture of the scene. Overall, the dynamic movement and vibrant color palette create an atmosphere of both beauty and power.",
-      "video_path": "fire/mixkit-two-orange-flames-on-black-background-685_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In this scene, a person is seated in a dimly lit room, possibly a recording studio, holding several drumsticks in their hands. The individual's face is partially obscured by sunglasses, adding a touch of mystery to their demeanor. They are wearing a colorful, patterned shirt with a mix of orange and blue tones that stands out against the darker background. The person appears focused and engaged with the drumsticks, their hands prominently displayed. The ambient light casts warm, soft shadows, emphasizing the texture and colors of their shirt and the wooden drumsticks. The room features wooden paneling, which complements the overall cozy, music-centric setting of the scene. The use of perspective centers on the drumsticks, highlighting the importance of rhythm and music in the captured moment.",
-      "video_path": "Music/mixkit-drummer-stretching-before-playing-42783_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A man is casually sitting on a sofa, engrossed in his meal and entertainment. He is holding a TV remote in one hand while reaching for food with the other, indicating a laid-back, comfortable evening. The table before him is filled with takeout containers, revealing a variety of appetizers and dishes, suggestive of a casual dining experience at home. The background is defined by colorful patterned cushions, adding a cozy, homey feel to the scene. Warm, ambient lighting highlights the relaxed atmosphere, casting soft shadows that contribute to the intimate setting. In this moment, he takes a bite of a sandwich, comfortably balancing his attention between food and whatever is playing on the screen.",
-      "video_path": "Man/mixkit-man-watching-tv-and-eating-fast-food-26089_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The scene opens to a breathtaking view of a tranquil ocean horizon at dusk, displaying a vibrant tapestry of oranges, pinks, and purples as the sun sets. In the foreground, tall, swaying palm trees frame the scene, their silhouettes stark against the colorful sky. The ocean itself shimmers with reflections of the sunset, creating a peaceful, almost ethereal atmosphere. A small boat can be seen in the distance, centered on the horizon, adding a sense of scale and solitude to the scene. The waves gently lap the shore, creating faint patterns on the sandy beach, which stretches across the foreground. Above, the sky is dotted with scattered clouds that catch the last light of the day, enhancing the drama and beauty of the scene. The overall mood is serene and contemplative, capturing a perfect moment of nature\u2019s grandeur.",
-      "video_path": "beach/mixkit-sunset-with-sailing-boats-2166_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A man sits hunched on a couch, the weight of emotions clearly visible on his posture. He wears a simple, gray t-shirt, and his head is bowed, resting in his hands, which cover most of his face, obscuring his features. The gentle light filtering through sheer curtains in the background casts a soft glow upon him, emphasizing the contrast between his static form and the hazy brightness behind. His elbows rest upon his knees, suggesting a posture of deep contemplation or distress. The simplicity of the room, with its muted colors, highlights the focus on the man's internal struggle. Delicate detailing on the fabric of his shirt adds texture, enhancing the scene's realism. Subtle changes in the natural light indicate the passage of time, as the man remains unmoving, absorbed in thought. This intimate moment captures a profound vulnerability, making the scene universally relatable and poignant.",
-      "video_path": "Man/mixkit-worried-and-sad-man-with-his-head-down-4701_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A pair of hands, belonging to an unseen figure, carefully unrolls a large sheet of crisp, white paper on a dark wooden table. The lighting is warm, casting a gentle glow that highlights the textures of the paper and the wood grain of the table. As the paper unfurls, the edges reveal the faint beginnings of a colorful map printed on its surface. The arms, clad in a casual gray T-shirt, suggest a relaxed and focused task at hand. Each motion is deliberate, with fingers deftly guiding the paper, ensuring it lays flat without creases. In the background, a hint of a red curtain can be seen, adding a touch of color and depth to the setting. The composition of the scene emphasizes the contrast between the bright paper and the rich tones of the surroundings. This serene and methodical action evokes a sense of exploration and preparation.",
-      "video_path": "Man/mixkit-unrolling-a-world-map-on-a-table-21626_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A young woman sits on a vibrant green seat inside a bus, illuminated by the soft glow of pink and blue lights. Her outfit is a striking mix of colors: a neon pink top paired with a jacket featuring dark sleeves, and jeans that provide a neutral contrast. She wears large, hoop earrings that catch the light as she moves slightly, exuding an air of cool confidence. Her gaze is directed thoughtfully to the side, suggesting contemplation or daydreaming during her commute. The metallic pole beside her adds a geometric element to the composition, reflecting the kaleidoscope of neon hues. The background is a clean, futuristic white, serving as a blank canvas that amplifies the neon atmosphere. Her relaxed posture and the modern bus setting create a scene that captures a blend of urban life and personal introspection.",
-      "video_path": "City/mixkit-fashion-model-posing-on-a-bus-42578_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A silver SUV drives along a winding, snow-covered mountain road, with dense pine trees blanketed in snow lining both sides. The scene is serene, with the vehicle moving smoothly, possibly on a winter journey or vacation. As the SUV disappears around the bend, another, darker SUV follows, creating a sense of motion and perspective on the snow-dusted asphalt. The towering, snow-laden rock formation to the right contrasts with the dark green of the pines, highlighting the peacefulness of the wintry landscape.",
-      "video_path": "Car/mixkit-curve-on-a-snowy-forest-road-3317_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The video showcases a vibrant urban skyline during twilight, with towering buildings reflecting the warm hues of the setting sun. A series of tall, cylindrical structures dominate the foreground, adjacent to a complex of industrial equipment and grids. The scene includes modern high-rise buildings with glass exteriors, capturing the evolving architecture of a bustling cityscape. A prominent structure labeled \"CITY OF AUSTIN POWER PLANT\" stands out, highlighting the industrial theme amidst the urban backdrop. The soft glow of city lights begins to pierce the approaching dusk, creating an inviting yet dynamic atmosphere. Shadows cast by the buildings add depth and contrast, emphasizing their massive scale and intricate designs. The overall composition is balanced between the natural light of the sunset and the artificial illumination of the city, offering a compelling visual narrative.",
-      "video_path": "Car/mixkit-slow-air-travel-in-reverse-over-a-big-city-49841_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the scene, a striking architectural structure dominates the view, bathed in a soft, ambient light. The enormous yellow arches serve as the centerpiece, drawing the eye upwards with their majestic curves and towering presence. The smooth, clean surfaces of the structure reflect the light, highlighting the texture and depth of the architecture. In the foreground, blurred streaks of headlights and taillights suggest the motion of vehicles passing by, adding dynamic energy to the otherwise still scene. The contrast between the fast-moving lights and the static arches creates a balanced composition. To the left, a lone streetlamp and a small tree provide a touch of nature and urban elements against the monumental backdrop. The night sky subtly peeks through the gaps in the structure, hinting at a clear, calm evening. Shadows from the arches create patterns on the ground, adding an intricate detail to the scene. Overall, the combination of light, shadow, and movement makes for a dramatic and visually captivating moment.",
-      "video_path": "Car/mixkit-a-fast-timelapse-of-the-street-with-a-monumental-yellow-50993_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A tranquil marina comes into full view under the golden hues of a setting sun. A collection of gleaming yachts and boats are neatly moored, their reflections shimmering softly on the gentle water. The sun's low position casts elongated shadows over the bustling harbor scene, while rolling hillsides surround the distant cityscape. The skyline is interspersed with modern buildings and clusters of residences, adding layers to the vibrant community. At the center, a broad wooden pier juts confidently into the harbor, extending an invitation for leisurely strolls. To the left, various shops and colorful structures line the waterfront, indicating a vibrant coastal economy. The entire atmosphere exudes a serene yet lively charm, balancing the hustle of maritime activity with the peacefulness of the encroaching dusk. It's a scene of calm anticipation, as if the whole place holds its breath before the night's events unfold.",
-      "video_path": "beach/mixkit-harbor-on-a-tourist-coast-with-many-boats-and-yachts-40077_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The video features a confident individual standing atop a structure against a clear blue sky, exuding a sense of freedom and style. The person is clad in a striking yellow button-up shirt tied at the waist, and beneath it, they wear a simple white top that adds to their relaxed yet stylish appearance. Completing the ensemble are high-waisted white jeans paired with a black belt, adding a touch of contrast. Around their neck is a bold red scarf, providing a splash of color and an air of vintage flair. The person's sunglasses, tinted in yellow, reflect the sunlight and contribute to the overall cool and composed demeanor. Their hair is styled elegantly, pulled back with headphones resting over the ears, suggesting they are immersed in music. One hand casually grazes the headphones, while the other rests gently on the railing, grounding the individual in the moment. The scene is an effortless blend of fashion and tranquility, capturing the spirit of sunny, carefree days.",
-      "video_path": "Music/mixkit-standing-woman-listening-to-music-460_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A ballerina gracefully spins and moves across a pink-hued studio, her poised figure accentuated by a shimmering white tutu and bodice. The background, a continuous wash of soft pink, provides a serene and ethereal atmosphere, emphasizing her fluid movements. Her arms extend with elegance, highlighting the delicacy and precision of her ballet pose, while her focused expression adds intensity to the scene. The subtle details of her costume, combined with the pink monochromatic ambiance, create a dreamlike spectacle, ideal for an AI to envision a oneiric dance setting.",
-      "video_path": "Dance/mixkit-portrait-of-a-ballerina-spinning-with-pink-background-40163_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "The scene unfolds with two human figures in the distance, making their way through a serene meadow, thick with tall golden grass swaying gently in the breeze. The sun hangs low in the sky, casting a soft, diffused glow that illuminates the landscape with a warm, ethereal light. These figures, clad in hiking gear, move deliberately, suggesting they're either embarking on or concluding a journey. Their silhouettes contrast against the lush greenery of the surrounding trees, whose branches reach out, framing the horizon. The play of light and shadow among the trees creates a quilt of textures, with each leaf catching a hint of the sun's dying rays. This tranquil setting evokes a sense of calm and adventure, capturing the quintessential beauty of nature\u2019s landscape.",
-      "video_path": "People/mixkit-landscape-in-nature-while-two-people-are-jogging-44348_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A large cargo ship is docked at an industrial port, its white superstructure contrasting with the deep green and yellow of its deck. The foreground is dominated by the calm, deep blue waters of the harbor, which reflect the vessel\u2019s imposing presence. Surrounding the ship, a series of industrial buildings and storage facilities are visible, hinting at the bustling activity of the port. The deck is intricately detailed, featuring an array of pipes, equipment, and railings, showcasing the ship's functionality and purpose. In the background, a paved area with green patches and a few parked vehicles adds to the busy, industrious atmosphere of the scene.",
-      "video_path": "sea/mixkit-empty-cargo-ship-waiting-at-the-port-4209_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A lone climber ascends a towering rock face, clad in a pink shirt and gray pants, displaying a determined and focused expression. The climber navigates the rugged surface, where the texture of the rock is peppered with natural pockets and crevices that offer handholds and footholds. Sunlight casts soft shadows across the cliff, highlighting the intricate patterns and the climber\u2019s strategic movements. The cliff looms high, with sparse vegetation breaking the monotony of the stone, while distant rocky formations form a dramatic backdrop against the clear blue sky. The climber\u2019s gear, including a harness and chalk bag, underscores the adventure and challenge woven into this majestic, vertical journey.",
-      "video_path": "Sport/mixkit-mountaineer-girl-climbing-a-steep-rocky-mountain-41089_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A person is seen in a close-up shot, skillfully adjusting the tuning pegs of a guitar, showcasing a focused and practiced hand. The image is in black and white, highlighting the contrast between the textures of the instrument and the clothing. The individual's shirt, visible in the background, adds a soft, subtle texture, while the dark tones of the guitar neck create depth in the scene. This composition captures a moment of concentration and finesse, perfect for recreating an intimate musical setting.",
-      "video_path": "Music/mixkit-guitarist-playing-so-inspired-black-and-white-shot-44178_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A musician is playing a large brass instrument with the words \"Brass Band\" clearly visible on its bell. The scene is set against a vibrant yellow backdrop, casting a warm glow on the subject. The musician wears a dark cap and a matching suit, adding a formal touch to his attire. He is deeply focused on his performance, with the instrument's intricate tubing adding complexity to the visual composition. The lighting creates dramatic shadows and highlights, emphasizing the musician's expression and the instrument's metallic sheen. This harmonious blend of color and form captures the essence of a live brass band performance.",
-      "video_path": "Music/mixkit-musician-playing-the-trombone-while-dancing-43752_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, a lone musician stands gracefully in front of a grand cathedral, playing an accordion while surrounded by the lively water display of a central fountain. Dressed in a casual ensemble, he wears a light-colored shirt, dark pants, and a flat cap that gives him a vintage charm. His posture is relaxed, yet engaged, as he sways gently in rhythm with the music, casting soft shadows on the cobblestone steps beneath him. The backdrop features the cathedral's towering twin spires, with intricate stonework that casts a rich, historical aura around the scene. Sunlight bathes the entire setting, enhancing the golden hues of the cathedral facade and creating a halo-like effect around the musician. The fountain's water jets splash playfully, catching glimmers of light and adding a dynamic element to the tranquil atmosphere. The scene captures a harmonious blend of architectural majesty and human creativity, framed by the clear, azure sky that extends infinitely above. It's a vivid depiction of solitude and artistry, set against a timeless urban landscape.",
-      "video_path": "Music/mixkit-man-plays-an-accordion-in-front-of-a-fountain-630_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the tranquil video, a person sits in a meditative pose on a gentle hillside, silhouetted against the dawning sky. The person is facing the breathtaking sunrise, with their back slightly turned to the viewer, wearing a simple, light-colored shirt. Their right hand rests on their knee, fingers relaxed in a common meditation mudra, symbolizing calmness and peace. The sky, a stunning blend of soft oranges and deep purples, gradually brightens, casting a warm glow over the lush, green landscape. To the left, the outlines of distant urban buildings can be seen against the horizon, adding a contrast between nature and city life. A river reflecting the sky's colors meanders through the scene, lending a serene, flowing dynamic to the landscape. Trees rise and fall gently across the terrain, their leaves rustling only faintly in the morning breeze. The person remains still and focused, embodying a moment of mindfulness and connection with nature. This visual captures a harmonious balance, evoking a sense of tranquility and introspection.",
-      "video_path": "City/mixkit-girl-meditating-in-yoga-pose-at-sunset-4803_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A serene landscape video captures a breathtaking panoramic view of a vast valley covered in a gentle mist. The undulating hills are lush with dense greenery, their rich foliage creating a vibrant border on the left side of the frame. The mist weaves through the landscape like a soft, ethereal blanket, lending a dream-like quality to the scene. In the distance, several mountain peaks emerge, their dark outlines contrasting against the pale blue sky. A few faint, wispy clouds drift lazily across the horizon, complementing the tranquil atmosphere. The sunlight filters through the haze, casting a warm glow and highlighting different textures of the flora. The overall mood is calm and contemplative, inviting the viewer to pause and appreciate nature's untouched beauty. The composition emphasizes depth and expansiveness, drawing attention to the harmony between earth and sky. This captivating scene embodies tranquility, offering a perfect backdrop for meditation or relaxation.",
-      "video_path": "forest/mixkit-flying-over-a-hill-with-a-view-of-the-surrounding-49743_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In this scene, a bearded individual is intently focused on their smartphone, with the sun setting in the background, casting a warm glow across the cityscape. The person, partially visible, is wearing a dark, buttoned shirt that contrasts with the golden hue of the sunset. Their hands are holding the smartphone delicately but purposefully, reflecting a sense of engagement and focus on the screen. The sunlight creates a striking lens flare effect, enhancing the dramatic atmosphere of the moment as it glimmers off the phone\u2019s surface. The surrounding environment hints at an elevated vantage point, providing a panoramic view of the urban landscape below.",
-      "video_path": "City/mixkit-guy-texting-at-sunset-265_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In an expansive, industrial space defined by towering columns and high ceilings, a solitary figure takes center stage. The person, dressed in dark, fitted clothing, assumes a powerful, dynamic stance with one leg bent forward and both arms outstretched in a horizontal arc. Framing this pose are intense flames that engulf their arms, creating a striking visual contrast against the muted tones of the room. The fire forms a brilliant halo of orange and yellow, casting flickering shadows on the weathered walls and worn, tiled floor. This interplay between light and dark showcases the dancer's poise and agility, as they maintain balance amidst the intense heat. Windows line the background, their panes dimly illuminated by the daylight filtering in, adding depth and perspective to the scene. The entire performance evokes a sense of raw energy and elemental mastery, as the figure continues to manipulate the fire in a seamless, mesmerizing display.",
-      "video_path": "fire/mixkit-expert-juggler-doing-tricks-with-a-stick-with-fire-43663_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A man is playing the violin, focused intently on his music. His fingers gracefully dance along the strings, flawlessly executing each note. He holds the violin close to his chin with a sense of familiarity and expertise. The rich, warm tones of the violin reflect in the soft lighting of the room. He wears a dark shirt, and a subtle necklace rests against his chest, adding a personal touch to his attire. The bow moves smoothly across the strings, producing a melody that seems to fill the space with emotion. His expression is one of concentration and passion, immersing himself fully in the performance. The background is softly blurred, bringing the violin's intricate craftsmanship and his precise movements into sharp focus. This serene and intimate moment captures the essence of his musical artistry.",
-      "video_path": "Music/mixkit-fiddler-playing-a-song-639_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the dimly lit parking garage, two figures engage in an impromptu game of soccer. The first person, wearing a light grey shirt and black pants with three white stripes, skillfully maneuvers the ball with precise footwork. The ground is slick with patches of water, reflecting the vibrant neon lights above. A second figure, clad in dark clothing, stands poised in the background, ready to intercept. The space is defined by stark yellow lines and orange safety bollards, adding structure to the chaotic energy of the scene. The soccer ball glides smoothly across the wet floor, kicking up droplets as it passes. Despite the muted colors of the environment, the players' movements are dynamic and full of life. Their shadowy silhouettes dance with the reflecting light, creating a mesmerizing visual interplay. The atmosphere is charged with focus and camaraderie, encapsulating the essence of a late-night urban soccer experience.",
-      "video_path": "Sport/mixkit-player-making-skillful-play-in-a-street-soccer-game-43504_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A lone climber is seen scaling a towering vertical rock face, demonstrating remarkable strength and focus. Dressed in a light-colored shirt and jeans, the climber grips the stone tightly, navigating the rough textures and crevices with precision. The sheer cliff is massive, exhibiting a range of natural hues from light tan to deep gray, accentuating the climber's figure against the vast rocky backdrop. Surrounding the cliff, scattered greenery and rugged terrain provide a sense of wilderness and isolation. The scene portrays a daring ascension requiring concentration and skill, capturing the essence of human endeavor against nature's formidable beauty.",
-      "video_path": "Sport/mixkit-skilled-mountaineer-climbing-a-gigantic-mountain-41083_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In this serene landscape, a lush meadow stretches across the foreground, dotted with vibrant yellow wildflowers swaying gently in the breeze. A towering tree stands majestically on the right side, its branches reaching wide under the bright blue sky filled with fluffy white clouds. On the left, dense trees form a natural corridor leading to the horizon, suggesting a sense of journey and possibility. The richness of the green grass contrasts beautifully with the golden hue of the distant fields, creating a harmonious palette of nature\u2019s colors. The play of light and shadow adds depth and dimension, evoking a tranquil, inviting atmosphere. It's a scene where nature\u2019s beauty simply commands attention, offering a perfect escape into tranquility.",
-      "video_path": "sky/mixkit-countryside-meadow-4075_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A solitary boat glides across the expansive, tranquil expanse of a serene lake. The vessel leaves a gentle wake behind, creating delicate ripples across the mirror-like surface. The water appears a rich shade of teal, seamlessly blending with the sky at the horizon. Silhouettes of distant trees are faintly visible, creating a picturesque backdrop that enhances the solitary journey of the boat. The sky is a calm gradient, shifting from soft oranges near the shore to the pale blues above. In the distance, a few slender poles emerge from the water, remnants of an old structure or natural formation. The mood of the scene is one of peace and solitude, with the boat journeying steadily through the quiet landscape. There is a sense of endless possibilities as the boat moves toward the unseen beyond the frame. The simplicity and stillness of the scene invite contemplation and reflection, encapsulating a perfect moment of quietude on the water.",
-      "video_path": "mountain/mixkit-motorboat-on-a-large-lake-with-turquoise-blue-waters-4996_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a cozy, dimly lit caf\u00e9, a woman sits alone at a rustic wooden table, fully engrossed in her reading. Her dark, wavy hair frames her face as she leans forward over an open book, suggesting deep focus and contemplation. The caf\u00e9\u2019s ambiance is warm, with hanging pendant lights casting a soft glow over the wooden shelves lined with jars and coffee paraphernalia in the background. A small cup of coffee rests just within her reach, alongside a glass dome encasing a solitary pastry, adding a touch of tranquility to the scene. Her casual attire, a denim jacket over a simple shirt, complements the laid-back, comfortable setting of the caf\u00e9. The contrast between her concentrated expression and the bustling, yet subdued caf\u00e9 atmosphere creates a harmonious, serene visual. The overall composition captures a quiet moment of introspection amidst the gentle hum of caf\u00e9 life.",
-      "video_path": "Woman/mixkit-woman-drinking-coffee-in-a-cafe-223_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a vast, deserted landscape under the night sky, a solitary figure stands at a small music setup, illuminated by strategically placed lights. The person is engrossed in playing a keyboard, with various electronic equipment surrounding them, casting soft glows of orange and blue hues across the scene. To the left, a large circular light adds a dramatic focal point, highlighting the intense contrast between the darkness and the lit performance area. This setup, with its minimalistic design and strategic lighting, creates a captivating and easily recognizable scene that merges the serene, expansive backdrop with an intimate, focused music performance.",
-      "video_path": "Music/mixkit-talented-dj-playing-in-a-lonely-desert-42414_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In a bustling urban scene, cars zoom past a weathered building, their blurred motion a testament to the city\u2019s lively pace. The building, with its faded yellow and brown facade, boasts graffiti that speaks of both art and decay, framing the scene with an air of urban grit. A solitary figure stands slightly to the side, clad casually in a gray top and mustard trousers, gazing into the street, seemingly detached from the surrounding flurry. The motion of the traffic creates a dynamic contrast against the static backdrop, emphasizing the relentless movement of the city. As the video progresses, a bright yellow taxi appears, slowing down as it approaches the figure, adding a pop of color to the desaturated hues of the environment. The interaction suggests a routine, a possibly daily exchange between the driver and the pedestrian, hinting at the rhythms of city life. Overhead, a soft, overcast sky casts a diffused light, lending the scene a subdued, timeless quality. Small elements, like the vertical pole cutting through the frame and the distant chatter of urban sounds, complete this vivid tableau of urban existence.",
-      "video_path": "Car/mixkit-morning-in-the-street-time-lapse-1648_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "A young woman sits on a curb in a tranquil park, basking in the golden hue of the setting sun. Beside her, a collie dog rests calmly, its fur illuminated by the warm sunlight, creating a serene glow. The woman's hand gently strokes the dog's back, highlighting the bond and affection between them. Tall trees surround the pair, casting elongated shadows on the leaf-laden ground, adding to the peaceful and intimate ambiance of the scene.",
-      "video_path": "Pets/mixkit-a-woman-pets-a-dog-in-a-park-1562_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, a grand, majestic elephant stands in an open, sunlit field, its massive form dominating the scene. The elephant's skin is a tapestry of earthy tones, with rough, textured wrinkles that add character to its already imposing presence. Its trunk, a powerful and flexible appendage, moves gently, swaying as the elephant possibly enjoys the warmth of the day. The background is a blur of greenery, suggesting a lively environment filled with trees and shrubs that provide a natural habitat. Light plays on the elephant's skin, highlighting patches of dust and dirt that give it an authentic wilderness look. The scene captures the tranquility and majesty of this gentle giant in its natural surroundings.",
-      "video_path": "Zoo/mixkit-wet-elephant-in-the-savanna-3663_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
-    },
-    {
-      "caption": "In the video, a fluffy dog with brown patches is intently engaged with a bright red toy shaped like a fire hydrant, which has a yellow and orange rope attached. The dog's body is relaxed as it lies on a plain white background, concentrating on nudging and playfully biting the toy. Its ears perk up slightly with curiosity, and its eyes are fixated on the toy, suggesting a scene of focused playfulness. The neutral tones of the dog's fur contrast starkly against the vivid red of the toy, creating a visually striking moment.",
-      "video_path": "Pets/mixkit-a-cute-border-collie-dog-play-with-a-fire-street-50662_clip_1.mp4",
-      "num_inference_steps":  3,
-      "height": 448,
-      "width": 832,
-      "num_frames": 61
     }
   ]
 }

examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/distill_dmd_VSA_t2v_5B.sh

@@ -14,29 +14,26 @@ export FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN
 # Configs
 NUM_GPUS=1
 MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers" 
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
 DATA_DIR="data/crush-smol_processed_ti2v/combined_parquet_dataset/"
 VALIDATION_DATASET_FILE="examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/validation.json"
-OUTPUT_DIR="checkpoints/wan_t2v_finetune"
 # export CUDA_VISIBLE_DEVICES=4,5
 # IP=[MASTER NODE IP]
 
 # Training arguments
 training_args=(
   --tracker_project_name wan_t2v_distill_dmd_VSA
-  --output_dir "$OUTPUT_DIR"
-  --max_train_steps 4000
-  --train_batch_size 1
+  --output_dir="checkpoints/wan_t2v_finetune"
+  --max_train_steps=4000
+  --train_batch_size=1
   --train_sp_batch_size 1
-  --gradient_accumulation_steps 1
+  --gradient_accumulation_steps=1
   --num_latent_t 31
   --num_height 704
   --num_width 1280
   --num_frames 121
   --enable_gradient_checkpointing_type "full"
-  --training_state_checkpointing_steps 500
-  --weight_only_checkpointing_steps 500
+  --training_state_checkpointing_steps=500
+  --weight_only_checkpointing_steps=500
 )
 
 # Parallel arguments
@@ -52,8 +49,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -73,8 +68,8 @@ validation_args=(
 
 # Optimizer arguments
 optimizer_args=(
-  --learning_rate 2e-6
-  --mixed_precision "bf16"
+  --learning_rate=1e-5
+  --mixed_precision="bf16"
   --weight_decay 0.01
   --max_grad_norm 1.0
 )
@@ -112,4 +107,4 @@ torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/distill_dmd_VSA_t2v_5B_lora.sh

@@ -14,8 +14,6 @@ export FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN
 # Configs
 NUM_GPUS=1
 MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-REAL_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
-FAKE_SCORE_MODEL_PATH="Wan-AI/Wan2.2-TI2V-5B-Diffusers"
 DATA_DIR="data/crush-smol_processed_ti2v/combined_parquet_dataset/"
 VALIDATION_DATASET_FILE="examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/validation.json"
 # export CUDA_VISIBLE_DEVICES=4,5
@@ -53,8 +51,6 @@ parallel_args=(
 model_args=(
   --model_path $MODEL_PATH
   --pretrained_model_name_or_path $MODEL_PATH
-  --real_score_model_path $REAL_SCORE_MODEL_PATH
-  --fake_score_model_path $FAKE_SCORE_MODEL_PATH
 )
 
 # Dataset arguments
@@ -113,4 +109,4 @@ torchrun \
     "${optimizer_args[@]}" \
     "${validation_args[@]}" \
     "${miscellaneous_args[@]}" \
-    "${dmd_args[@]}"
+    "${dmd_args[@]}"

examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/download_dataset.sh

@@ -1,3 +1,3 @@
 #!/bin/bash
 
-python scripts/huggingface/download_hf.py --repo_id "wlsaidhi/crush-smol-merged" --local_dir "data/crush-smol" --repo_type "dataset"
+python scripts/huggingface/download_hf.py --repo_id "wlsaidhi/crush-smol-merged" --local_dir "data/crush-smol" --repo_type "dataset"

examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/preprocess_wan_data_ti2v_5b.sh

@@ -21,4 +21,4 @@ torchrun --nproc_per_node=$GPU_NUM \
     --samples_per_file 8 \
     --flush_frequency 8 \
     --video_length_tolerance_range 5 \
-    --preprocess_task "t2v" 
+    --preprocess_task "t2v" 

examples/inference/basic/basic.py

@@ -9,9 +9,9 @@ def main():
     # 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",
+        "Wan-AI/Wan2.1-T2V-14B-Diffusers",
         # FastVideo will automatically handle distributed setup
-        num_gpus=1,
+        num_gpus=4,
         use_fsdp_inference=True,
         dit_cpu_offload=False,
         vae_cpu_offload=False,
@@ -25,9 +25,7 @@ def main():
     # 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."
+        "A watermelon wearing a helmet is crushed by a hydraulic press, causing it to flatten and burst open."
     )
     video = generator.generate_video(prompt, output_path=OUTPUT_PATH, save_video=True)
     # video = generator.generate_video(prompt, sampling_param=sampling_param, output_path="wan_t2v_videos/")
@@ -35,11 +33,7 @@ def main():
     # 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.")
+        "The video shows a green and orange object being flattened as if it were under a hydraulic press, with the press moving down and compressing the object")
     video2 = generator.generate_video(prompt2, output_path=OUTPUT_PATH, save_video=True)
 
 

examples/inference/basic/basic_ray.py

@@ -1,44 +0,0 @@
-from fastvideo import VideoGenerator
-
-# from fastvideo.configs.sample import SamplingParam
-
-OUTPUT_PATH = "video_samples"
-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",
-        # FastVideo will automatically handle distributed setup
-        num_gpus=2,
-        use_fsdp_inference=True,
-        dit_cpu_offload=False,
-        vae_cpu_offload=False,
-        text_encoder_cpu_offload=True,
-        pin_cpu_memory=True, # set to false if low CPU RAM or hit obscure "CUDA error: Invalid argument"
-        distributed_executor_backend="ray",
-        # image_encoder_cpu_offload=False,
-    )
-
-    # 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, output_path=OUTPUT_PATH, save_video=True)
-
-    # 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, output_path=OUTPUT_PATH, save_video=True)
-
-
-if __name__ == "__main__":
-    main()

examples/inference/basic/basic_self_forcing_causal.py

@@ -16,6 +16,7 @@ def main():
         use_fsdp_inference=True,
         text_encoder_cpu_offload=False,
         dit_cpu_offload=False,
+        use_sf_wan=True,
     )
 
     sampling_param = SamplingParam.from_pretrained(model_name)

examples/inference/basic/basic_self_forcing_causal_wan2_2_i2v.py

@@ -1,44 +0,0 @@
-# NOTE: This is still a work in progress, and the checkpoints are not released yet.
-
-from fastvideo import VideoGenerator, SamplingParam
-import json
-# from fastvideo.configs.sample import SamplingParam
-
-OUTPUT_PATH = "video_samples_self_forcing_causal_wan2_2_14B_i2v"
-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/SFWan2.2-I2V-A14B-Preview-Diffusers",
-        # FastVideo will automatically handle distributed setup
-        num_gpus=1,
-        use_fsdp_inference=True,
-        dit_cpu_offload=True, # DiT need to be offloaded for MoE
-        dit_precision="fp32",
-        vae_cpu_offload=False,
-        text_encoder_cpu_offload=True,
-        dmd_denoising_steps=[1000, 850, 700, 550, 350, 275, 200, 125],
-        # Set pin_cpu_memory to false if CPU RAM is limited and there're no frequent CPU-GPU transfer
-        pin_cpu_memory=True,
-        # image_encoder_cpu_offload=False,
-    )
-
-    sampling_param = SamplingParam.from_pretrained("FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers")
-    sampling_param.num_frames = 73
-    sampling_param.width = 832
-    sampling_param.height = 480
-    sampling_param.seed = 1000
-
-    with open("prompts/mixkit_i2v.jsonl", "r") as f:
-        prompt_image_pairs = json.load(f)
-
-    for prompt_image_pair in prompt_image_pairs:
-        prompt = prompt_image_pair["prompt"]
-        image_path = prompt_image_pair["image_path"]
-        _ = generator.generate_video(prompt, image_path=image_path, output_path=OUTPUT_PATH, save_video=True, sampling_param=sampling_param)
-
-
-if __name__ == "__main__":
-    main()

examples/inference/basic/basic_self_forcing_causal_wan2_2_t2v.py

@@ -1,43 +0,0 @@
-# NOTE: This is still a work in progress, and the checkpoints are not released yet.
-
-from fastvideo import VideoGenerator
-
-# from fastvideo.configs.sample import SamplingParam
-
-OUTPUT_PATH = "video_samples_self_forcing_causal_wan2_2_14B_t2v"
-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(
-        "rand0nmr/SFWan2.2-T2V-A14B-Diffusers",
-        # FastVideo will automatically handle distributed setup
-        num_gpus=1,
-        use_fsdp_inference=True,
-        dit_cpu_offload=True, # DiT need to be offloaded for MoE
-        vae_cpu_offload=False,
-        text_encoder_cpu_offload=True,
-        dmd_denoising_steps=[1000, 850, 700, 550, 350, 275, 200, 125],
-        # Set pin_cpu_memory to false if CPU RAM is limited and there're no frequent CPU-GPU transfer
-        pin_cpu_memory=True,
-        init_weights_from_safetensors="/mnt/sharefs/users/hao.zhang/wei/SFwan2.2_distill_self_forcing_release_cfg2/checkpoint-246_weight_only/generator_inference_transformer/",
-        init_weights_from_safetensors_2="/mnt/sharefs/users/hao.zhang/wei/SFwan2.2_distill_self_forcing_release_cfg2/checkpoint-246_weight_only/generator_2_inference_transformer/",
-        num_frame_per_block=7,
-        # image_encoder_cpu_offload=False,
-    )
-
-    # 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."
-    )
-    _ = generator.generate_video(prompt, output_path=OUTPUT_PATH, save_video=True, num_frames=81)
-
-
-if __name__ == "__main__":
-    main()

examples/inference/basic/basic_wan2_2_Fun.py

@@ -1,36 +0,0 @@
-from fastvideo import VideoGenerator
-
-# from fastvideo.configs.sample import SamplingParam
-
-OUTPUT_PATH = "video_samples_wan2_1_Fun"
-OUTPUT_NAME = "wan2.1_test"
-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(
-        "IRMChen/Wan2.1-Fun-1.3B-Control-Diffusers",
-        # "alibaba-pai/Wan2.2-Fun-A14B-Control",
-        # FastVideo will automatically handle distributed setup
-        num_gpus=1,
-        use_fsdp_inference=True,
-        dit_cpu_offload=True, # DiT need to be offloaded for MoE
-        vae_cpu_offload=False,
-        text_encoder_cpu_offload=True,
-        # Set pin_cpu_memory to false if CPU RAM is limited and there're no frequent CPU-GPU transfer
-        pin_cpu_memory=True,
-        # image_encoder_cpu_offload=False,
-    )
-
-    prompt = "一位年轻女性穿着一件粉色的连衣裙，裙子上有白色的装饰和粉色的纽扣。她的头发是紫色的，头上戴着一个红色的大蝴蝶结，显得非常可爱和精致。她还戴着一个红色的领结，整体造型充满了少女感和活力。她的表情温柔，双手轻轻交叉放在身前，姿态优雅。背景是简单的灰色，没有任何多余的装饰，使得人物更加突出。她的妆容清淡自然，突显了她的清新气质。整体画面给人一种甜美、梦幻的感觉，仿佛置身于童话世界中。"
-    negative_prompt = "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走"
-    # prompt                  = "A young woman with beautiful, clear eyes and blonde hair stands in the forest, wearing a white dress and a crown. Her expression is serene, reminiscent of a movie star, with fair and youthful skin. Her brown long hair flows in the wind. The video quality is very high, with a clear view. High quality, masterpiece, best quality, high resolution, ultra-fine, fantastical."
-    # negative_prompt         = "Twisted body, limb deformities, text captions, comic, static, ugly, error, messy code."
-    image_path = "https://pai-aigc-photog.oss-cn-hangzhou.aliyuncs.com/wan_fun/asset_Wan2_2/v1.0/8.png"
-    control_video_path = "https://pai-aigc-photog.oss-cn-hangzhou.aliyuncs.com/wan_fun/asset_Wan2_2/v1.0/pose.mp4"
-
-    video = generator.generate_video(prompt, negative_prompt=negative_prompt, image_path=image_path, video_path=control_video_path, output_path=OUTPUT_PATH, output_video_name=OUTPUT_NAME, save_video=True)
-
-if __name__ == "__main__":
-    main()

examples/inference/gradio/gradio_frontend.py

@@ -1,20 +1,87 @@
 import argparse
 import os
+import requests
 import base64
 import time
 
 import gradio as gr
-from fastvideo.entrypoints.video_generator import VideoGenerator
+
 from fastvideo.configs.sample.base import SamplingParam
-from copy import deepcopy
 
 
 MODEL_PATH_MAPPING = {
     "FastWan2.1-T2V-1.3B": "FastVideo/FastWan2.1-T2V-1.3B-Diffusers",
-    # "FastWan2.2-TI2V-5B-FullAttn": "FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
+    "FastWan2.2-TI2V-5B-FullAttn": "FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
 }
 
-def create_timing_display(inference_time, total_time, stage_execution_times, num_frames):
+
+class RayServeClient:
+    def __init__(self, backend_url: str):
+        self.backend_url = backend_url
+        self.session = requests.Session()
+    
+    def check_health(self) -> bool:
+        try:
+            response = self.session.get(f"{self.backend_url}/health", timeout=5)
+            return response.status_code == 200
+        except requests.exceptions.RequestException:
+            return False
+    
+    def generate_video(self, request_data: dict) -> dict:
+        start_time = time.time()
+        
+        try:
+            headers = {"Content-Type": "application/json"}
+            
+            response = self.session.post(
+                f"{self.backend_url}/generate_video",
+                json=request_data,
+                headers=headers,
+                timeout=300
+            )
+            
+            round_trip_time = time.time() - start_time
+            
+            if response.status_code == 200:
+                result = response.json()
+                backend_total = result.get("total_time", 0)
+                network_time = round_trip_time - backend_total
+                result["network_time"] = network_time
+                return result
+            else:
+                return {"success": False, "error_message": f"HTTP {response.status_code}: {response.text}"}
+                
+        except requests.exceptions.RequestException as e:
+            return {"success": False, "error_message": f"Request failed: {str(e)}"}
+
+
+def save_video_from_base64(video_data: str, output_dir: str, prompt: str) -> str:
+    if not video_data:
+        return None
+    
+    try:
+        if video_data.startswith('data:video/'):
+            video_data = video_data.split(',')[1]
+        
+        video_bytes = base64.b64decode(video_data)
+        
+        safe_prompt = prompt[:50].replace(' ', '_').replace('/', '_').replace('\\', '_')
+        video_filename = f"{safe_prompt}.mp4"
+        video_path = os.path.join(output_dir, video_filename)
+        
+        os.makedirs(output_dir, exist_ok=True)
+        
+        with open(video_path, 'wb') as f:
+            f.write(video_bytes)
+        
+        return video_path
+        
+    except Exception as e:
+        print(f"Failed to save video: {e}")
+        return None
+
+
+def create_timing_display(inference_time, encoding_time, network_time, total_time, stage_execution_times, num_frames):
     dit_denoising_time = f"{stage_execution_times[5]:.2f}s" if len(stage_execution_times) > 5 else "N/A"
     
     timing_html = f"""
@@ -34,12 +101,12 @@ def create_timing_display(inference_time, total_time, stage_execution_times, num
             <div class="timing-card">
                 <div style="font-size: 20px;">🎬</div>
                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">Video Encoding</div>
-                <div style="font-size: 16px; color: #dc2626;">N/A</div>
+                <div style="font-size: 16px; color: #dc2626;">{encoding_time:.2f}s</div>
             </div>
             <div class="timing-card">
                 <div style="font-size: 20px;">🌐</div>
                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">Network Transfer</div>
-                <div style="font-size: 16px; color: #059669;">N/A</div>
+                <div style="font-size: 16px; color: #059669;">{network_time:.2f}s</div>
             </div>
             <div class="timing-card">
                 <div style="font-size: 20px;">📊</div>
@@ -57,12 +124,7 @@ def create_timing_display(inference_time, total_time, stage_execution_times, num
         </div>"""
     
     return timing_html + "</div>"
-def setup_model_environment(model_path: str) -> None:
-    if "fullattn" in model_path.lower():
-        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
-    else:
-        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "VIDEO_SPARSE_ATTN"
-    os.environ["FASTVIDEO_STAGE_LOGGING"] = "1"
+
 
 def load_example_prompts():
     def contains_chinese(text):
@@ -82,7 +144,7 @@ def load_example_prompts():
             print(f"Warning: Could not read {filepath}: {e}")
         return prompts, labels
     
-    examples, example_labels = load_from_file("examples/inference/gradio/local/prompts_final.txt")
+    examples, example_labels = load_from_file("prompts/prompts_final.txt")
     
     if not examples:
         examples = ["A crowded rooftop bar buzzes with energy, the city skyline twinkling like a field of stars in the background."]
@@ -91,71 +153,81 @@ def load_example_prompts():
     return examples, example_labels
 
 
-def create_gradio_interface(default_params: dict[str, SamplingParam], generators: dict[str, VideoGenerator]):
+def create_gradio_interface(backend_url: str, default_params: dict[str, SamplingParam]):
+    
+    client = RayServeClient(backend_url)
+    
     def generate_video(
         prompt, negative_prompt, use_negative_prompt, seed, guidance_scale,
         num_frames, height, width, randomize_seed, model_selection, progress
     ):
-        model_path = MODEL_PATH_MAPPING.get(model_selection, "FastVideo/FastWan2.1-T2V-1.3B-Diffusers")
-        setup_model_environment(model_path)
-        try:
-            if progress:
-                progress(0.1, desc="Loading model for local inference...")
+        if not client.check_health():
+            return None, f"Backend is not available. Please check if Ray Serve is running at {backend_url}", ""
+        
+        # Validate dimensions
+        max_pixels = 720 * 1280
+        if height * width > max_pixels:
+            return None, f"Video dimensions too large. Maximum: 720x1280 pixels", ""
+        
+        if progress:
+            progress(0.1, desc="Checking backend health...")
+        
+        request_data = {
+            "prompt": prompt,
+            "negative_prompt": negative_prompt,
+            "use_negative_prompt": use_negative_prompt,
+            "seed": seed,
+            "guidance_scale": guidance_scale,
+            "num_frames": num_frames,
+            "height": height,
+            "width": width,
+            "randomize_seed": randomize_seed,
+            "return_frames": False,
+            "image_path": None,
+            "model_path": MODEL_PATH_MAPPING.get(model_selection, "FastVideo/FastWan2.1-T2V-1.3B-Diffusers")
+        }
+        
+        if progress:
+            progress(0.4, desc="Generating video...")
+        
+        response = client.generate_video(request_data)
+        
+        if progress:
+            progress(0.8, desc="Processing response...")
+        
+        if response.get("success", False):
+            video_data = response.get("video_data", "")
+            used_seed = response.get("seed", seed)
+            inference_time = response.get("inference_time", 0.0)
+            encoding_time = response.get("encoding_time", 0.0)
+            total_time = response.get("total_time", 0.0)
+            network_time = response.get("network_time", 0.0)
+            stage_execution_times = response.get("stage_execution_times", [])
             
-            generator = generators[model_path]
-            params = deepcopy(default_params[model_path])
-            total_start_time = time.time()
-            if progress:
-                progress(0.2, desc="Configuring parameters...")
-
-            params.prompt = prompt
-            params.seed = int(seed)
-            params.guidance_scale = guidance_scale
-            params.num_frames = int(num_frames)
-            params.height = int(height)
-            params.width = int(width)
-
-            if randomize_seed:
-                params.seed = torch.randint(0, 1000000, (1, )).item()
-
-            if use_negative_prompt and negative_prompt:
-                params.negative_prompt = negative_prompt
+            timing_details = create_timing_display(
+                inference_time, encoding_time, network_time, total_time, 
+                stage_execution_times, num_frames
+            )
+            
+            if video_data:
+                if progress:
+                    progress(0.9, desc="Saving video...")
+                
+                video_path = save_video_from_base64(video_data, "outputs", prompt)
+                
+                if progress:
+                    progress(1.0, desc="Generation complete!")
+                
+                if video_path and os.path.exists(video_path):
+                    return video_path, used_seed, timing_details
+                else:
+                    return None, "Failed to save video", ""
             else:
-                params.negative_prompt = default_params[model_path].negative_prompt
-
-            if progress:
-                progress(0.4, desc="Generating video locally...")
-
-            output_dir = "outputs/"
-            os.makedirs(output_dir, exist_ok=True)
-            start_time = time.time()
-            result = generator.generate_video(prompt=prompt, sampling_param=params, save_video=True, return_frames=False)
-            inference_time = time.time() - start_time
-            logging_info = result.get("logging_info", None)
-            if logging_info:
-                stage_names = logging_info.get_execution_order()
-                stage_execution_times = [
-                    logging_info.get_stage_info(stage_name).get("execution_time", 0.0) 
-                    for stage_name in stage_names
-                ]
-            else:
-                stage_names = []
-                stage_execution_times = []
-            total_time = time.time() - total_start_time
-            timing_details=create_timing_display(inference_time=inference_time, total_time=total_time, stage_execution_times=stage_execution_times, num_frames=params.num_frames)
-            safe_prompt = params.prompt[:100].replace(' ', '_').replace('/', '_').replace('\\', '_')
-            video_filename = f"{params.prompt[:100]}.mp4"
-            output_path = os.path.join(output_dir, video_filename)
-
-            if progress:
-                progress(1.0, desc="Generation complete!")
-
-            return output_path, params.seed, timing_details
-
-        except Exception as e:
-            print(f"An error occurred during local generation: {e}")
-            return None, f"Generation failed: {str(e)}", ""
-
+                return None, "No video data received from backend", ""
+        else:
+            error_msg = response.get("error_message", "Unknown error occurred")
+            return None, f"Generation failed: {error_msg}", ""
+    
     examples, example_labels = load_example_prompts()
     
     theme = gr.themes.Base().set(
@@ -189,8 +261,7 @@ def create_gradio_interface(default_params: dict[str, SamplingParam], generators
     initial_values = get_default_values("FastWan2.1-T2V-1.3B")
     
     with gr.Blocks(title="FastWan", theme=theme) as demo:
-        gr.Image("assets/full.svg", show_label=False, container=False, height=80)
-
+        gr.Image("assets/logos/logo.svg", show_label=False, container=False, height=80)
         gr.HTML("""
         <div style="text-align: center; margin-bottom: 10px;">
             <p style="font-size: 18px;"> Make Video Generation Go Blurrrrrrr </p>
@@ -451,7 +522,7 @@ def create_gradio_interface(default_params: dict[str, SamplingParam], generators
         
         gr.HTML("""
         <div style="text-align: center; margin-top: 10px; margin-bottom: 15px;">
-            <p style="font-size: 16px; margin: 0;">Note that this demo is meant to showcase FastWan's quality and that under a large number of requests, generation speed may be affected.</p>
+            <p style="font-size: 16px; margin: 0;">The compute for this demo is generously provided by <a href="https://www.gmicloud.ai/" target="_blank">GMI Cloud</a>. Note that this demo is meant to showcase FastWan's quality and that under a large number of requests, generation speed may be affected. We are also rate-limiting users to 3 requests per minute.</p>
         </div>
         """)
         
@@ -498,6 +569,7 @@ def create_gradio_interface(default_params: dict[str, SamplingParam], generators
                 prompt, negative_prompt, use_negative_prompt, seed, guidance_scale, 
                 num_frames, height, width, randomize_seed, model_selection, progress
             )
+            
             if result_path and os.path.exists(result_path):
                 return (
                     result_path, 
@@ -535,25 +607,28 @@ def create_gradio_interface(default_params: dict[str, SamplingParam], generators
 
 
 def main():
-    parser = argparse.ArgumentParser(description="FastVideo Gradio Local Demo")
+    parser = argparse.ArgumentParser(description="FastVideo Gradio Frontend")
+    parser.add_argument("--backend_url", type=str, default="http://localhost:8000",
+                        help="URL of the Ray Serve backend")
     parser.add_argument("--t2v_model_paths", type=str,
-                        default="FastVideo/FastWan2.1-T2V-1.3B-Diffusers",
+                        default="FastVideo/FastWan2.1-T2V-1.3B-Diffusers,FastVideo/FastWan2.1-T2V-14B-Diffusers",
                         help="Comma separated list of paths to the T2V model(s)")
     parser.add_argument("--host", type=str, default="0.0.0.0",
                         help="Host to bind to")
     parser.add_argument("--port", type=int, default=7860,
                         help="Port to bind to")
+    
     args = parser.parse_args()
-    generators = {}
+    
     default_params = {}
     model_paths = args.t2v_model_paths.split(",")
     for model_path in model_paths:
-        print(f"Loading model: {model_path}")
-        setup_model_environment(model_path)
-        generators[model_path] = VideoGenerator.from_pretrained(model_path)
         default_params[model_path] = SamplingParam.from_pretrained(model_path)
-    demo = create_gradio_interface(default_params, generators)
+    
+    demo = create_gradio_interface(args.backend_url, default_params)
+    
     print(f"Starting Gradio frontend at http://{args.host}:{args.port}")
+    print(f"Backend URL: {args.backend_url}")
     print(f"T2V Models: {args.t2v_model_paths}")
     
     from fastapi import FastAPI, Request, HTTPException
@@ -562,10 +637,10 @@ def main():
     
     app = FastAPI()
     
-    @app.get("/logo.png")
+    @app.get("/logo.svg")
     def get_logo():
         return FileResponse(
-            "assets/full.svg",
+            "assets/logos/logo.svg", 
             media_type="image/svg+xml",
             headers={
                 "Cache-Control": "public, max-age=3600",
@@ -575,7 +650,7 @@ def main():
     
     @app.get("/favicon.ico")
     def get_favicon():
-        favicon_path = "assets/icon-simple.svg"
+        favicon_path = "assets/logos/icon_simple.svg"
         
         if os.path.exists(favicon_path):
             return FileResponse(
@@ -608,7 +683,7 @@ def main():
             <meta property="og:url" content="{base_url}/">
             <meta property="og:title" content="FastWan">
             <meta property="og:description" content="Make video generation go blurrrrrrr">
-            <meta property="og:image" content="{base_url}/logo.png">
+            <meta property="og:image" content="{base_url}/logo.svg">
             <meta property="og:image:width" content="1200">
             <meta property="og:image:height" content="630">
             <meta property="og:site_name" content="FastWan">
@@ -617,7 +692,7 @@ def main():
             <meta property="twitter:url" content="{base_url}/">
             <meta property="twitter:title" content="FastWan">
             <meta property="twitter:description" content="Make video generation go blurrrrrrr">
-            <meta property="twitter:image" content="{base_url}/logo.png">
+            <meta property="twitter:image" content="{base_url}/logo.svg">
             <link rel="icon" type="image/png" sizes="32x32" href="/favicon.ico">
             <link rel="icon" type="image/png" sizes="16x16" href="/favicon.ico">
             <link rel="apple-touch-icon" href="/favicon.ico">
@@ -652,5 +727,4 @@ def main():
 
 
 if __name__ == "__main__":
-    
     main() 

examples/inference/gradio/local/README.md

@@ -1,56 +0,0 @@
-# FastVideo Gradio Local 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.
-
-## Overview
-
-The demo uses the FastVideo framework to generate videos based on text prompts. It provides a simple web interface built with Gradio that allows users to:
-
-- Enter text prompts to generate videos
-- Customize video parameters (dimensions, number of frames, etc.)
-- Use negative prompts to guide the generation process
-- Set or randomize seeds for reproducibility
-
----
-
-## Usage
-
-Run the demo with:
-
-```bash
-python examples/inference/gradio/local/gradio_local_demo.py
-```
-
-This will start a web server at `http://0.0.0.0:7860` where you can access the interface.
-
----
-
-## Model Initialization
-
-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, *without ever needing to reload the model*.
-
-## Video Generation
-
-The core functionality is in the `generate_video` function, which:
-1. Processes user inputs
-2. Uses the FastVideo VideoGenerator from earlier to run inference (`generator.generate_video()`)
-
-## Gradio Interface
-
-The interface is built with several components:
-- A text input for the prompt
-- A video display for the result
-- Inference options in a collapsible accordion:
-  - Height and width sliders
-  - Number of frames slider
-  - Guidance scale slider
-  - Negative prompt options
-  - Seed controls
-
-### Inference Options
-
-- **Height/Width**: Control the resolution of the generated video
-- **Number of Frames**: Set how many frames to generate
-- **Guidance Scale**: Control how closely the generation follows the prompt
-- **Negative Prompt**: Specify what you don't want to see in the video
-- **Seed**: Control randomness for reproducible results

examples/inference/gradio/local/prompts_final.txt

@@ -1,11 +0,0 @@
-A dynamic shot of a sleek black motorcycle accelerating down an empty highway at sunset. The bike's engine roars as it gains speed, smoke trailing from the tires. The rider, wearing a black leather jacket and helmet, leans forward with determination, gripping the handlebars tightly. The camera follows the motorcycle from a distance, capturing the dust kicked up behind it, then zooms in to show the intense focus on the rider's face. The background showcases the endless road stretching into the horizon with vibrant orange and pink hues of the setting sun. Medium shot transitioning to close-up.
-A Jedi Master Yoda, recognizable by his green skin, large ears, and wise wrinkles, is performing on a small stage, strumming a guitar with great concentration. Yoda wears a casual robe and sits on a stool, his eyes closed as he plays, fully immersed in the music. The stage is dimly lit with spotlights highlighting Yoda, creating a mystical atmosphere. The background shows a live audience watching intently. Medium close-up shot focusing on Yoda's expressive face and hands moving gracefully over the guitar strings.
-A cute, fluffy panda bear is preparing a meal in a cozy, modern kitchen. The panda is standing at a wooden countertop, wearing a white chef’s hat and apron. It skillfully stirs a pot on the stove with one hand while holding a spatula in the other. The kitchen is well-lit, with appliances and cabinets in pastel colors, creating a warm and inviting atmosphere. The panda moves gracefully, with a focused and determined expression, as steam rises from the pot. Medium shot focusing on the panda’s actions at the stove.
-In a futuristic Tokyo rooftop during a heavy rainstorm, a robotic DJ stands behind a turntable, spinning vinyl records in a cyberpunk night setting. The robot has metallic, sleek body parts with glowing blue LED lights, and it moves gracefully with the beat. Raindrops create a shimmering effect as they hit the ground and the DJ. The surrounding environment features neon signs, towering skyscrapers, and a dark, misty atmosphere. The camera starts with a wide shot of the city skyline before zooming in on the DJ performing. Sci-fi, fantasy.
-A realistic animated scene featuring a polar bear playing a guitar. The polar bear is standing upright, wearing a cozy fur vest and fingerless gloves. It holds the guitar with both hands, strumming the strings with one hand while plucking them with the other, showcasing natural, fluid motions. The polar bear's expressive face shows concentration and joy as it plays. The background is a snowy Arctic landscape with icebergs and a clear blue sky. The scene captures the bear from a mid-shot angle, focusing on its interaction with the guitar.
-The scene opens to a breathtaking view of a tranquil ocean horizon at dusk, displaying a vibrant tapestry of oranges, pinks, and purples as the sun sets. In the foreground, tall, swaying palm trees frame the scene, their silhouettes stark against the colorful sky. The ocean itself shimmers with reflections of the sunset, creating a peaceful, almost ethereal atmosphere. A small boat can be seen in the distance, centered on the horizon, adding a sense of scale and solitude to the scene. The waves gently lap the shore, creating faint patterns on the sandy beach, which stretches across the foreground. Above, the sky is dotted with scattered clouds that catch the last light of the day, enhancing the drama and beauty of the scene. The overall mood is serene and contemplative, capturing a perfect moment of nature’s grandeur.
-A large, modern semi-truck accelerating down an empty highway, gaining speed with each second. The truck's powerful engine roars as it moves forward, smoke billowing from the tires. The camera starts from a wide shot, capturing the truck in the distance, then smoothly zooms in to follow the vehicle as it speeds up. The truck's headlights illuminate the road ahead, casting a bright glow. The truck driver can be seen through the windshield, focused and determined. The background shows the vast openness of the highway stretching into the horizon under a clear blue sky. Medium to close-up shots of the truck as it accelerates.
-Soft blue light pulses from the blade’s rune-etched hilt, illuminating nearby moss-covered roots and ferns. The surrounding trees are tall and gnarled, their branches curling like claws overhead. Fog swirls gently at ground level, parting slightly as a figure in a cloak approaches from the distance. Medium shot slowly zooming toward the sword, emphasizing its mystical aura.
-The video opens with a tranquil scene in the heart of a dense forest, emphasizing two large, textured tree trunks in the foreground framing the view. Sunlight filters through the canopy above, casting intricate patterns of light and shadow on the trees and the ground. Between the tree trunks, a clear view of a calm, muddy river unfolds, its surface shimmering under the gentle sunlight. The riverbank is decorated with a variety of small bushes and vibrant foliage, subtly transitioning into the deep greens of tall, leafy plants. In the background, the dense forest looms, filled with dark, towering trees, their branches intertwining to form an intricate canopy. The scene is bathed in the soft glow of the sun, creating a serene and picturesque setting. Occasional sunbeams pierce through the foliage, adding a magical aura to the landscape. The vibrant reds and oranges of the smaller plants add contrast, bringing warmth to the earthy tones of the scenery. Overall, this harmonious blend of natural elements creates a peaceful and idyllic forest setting.
-A lone figure stands on a large, moss-covered rock, surrounded by the soft rush of a nearby stream. The figure is wearing white sneakers and shorts, with a plaid shirt that hangs loosely in the breeze. The lighting creates dramatic shadows, enhancing the textures of the rock and the subtle movement of the water below. In the background, a waterfall cascades into the stream, completing this tranquil and serene nature scene.
-In an industrial setting, a person leans casually against a railing, exuding a sense of confidence and composure. They are wearing a striking outfit, consisting of a vibrant, patterned jacket over a simple white crop top, creating a bold contrast. The atmosphere is infused with warm, ambient lighting that casts soft shadows on the concrete walls and metallic surfaces. Intricate wiring and pipes form an intricate backdrop, enhancing the urban aesthetic. Their relaxed posture and direct, engaging gaze suggest a sense of ease in this industrial environment. This scene encapsulates a blend of modern fashion and gritty, urban architecture, creating a visually compelling narrative.

examples/inference/gradio/ray_serve_backend.py

@@ -26,7 +26,6 @@ SEED_RANGE_MAX = 1_000_000
 SUPPORTED_MODELS = [
     "FastVideo/FastWan2.1-T2V-1.3B-Diffusers", 
     "FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
-    "FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers",
 ]
 
 MODEL_CONFIGS = {
@@ -43,13 +42,6 @@ MODEL_CONFIGS = {
         "dit_cpu_offload": True,
         "vae_cpu_offload": False,
         "VSA_sparsity": 0.9,
-    },
-    "I2V-A14B": {
-        "num_cpus": 15,
-        "text_encoder_cpu_offload": True,
-        "dit_cpu_offload": True,
-        "vae_cpu_offload": False,
-        "VSA_sparsity": 0.0,
     }
 }
 
@@ -66,7 +58,6 @@ class VideoGenerationRequest(BaseModel):
     randomize_seed: bool = False
     return_frames: bool = False
     model_path: Optional[str] = None
-    image_data: Optional[str] = None  # Base64 encoded image for I2V
 
 
 class VideoGenerationResponse(BaseModel):
@@ -100,38 +91,11 @@ def encode_video_to_base64(frames: List[np.ndarray], fps: int = DEFAULT_FPS) ->
         return ""
 
 
-def save_image_from_base64(image_data: str, output_dir: str) -> Optional[str]:
-    """Save base64 image data to a temporary file and return the path."""
-    if not image_data:
-        return None
-    
-    try:
-        # Remove data URL prefix if present
-        if image_data.startswith('data:image/'):
-            image_data = image_data.split(',')[1]
-        
-        image_bytes = base64.b64decode(image_data)
-        
-        # Save to temporary file
-        os.makedirs(output_dir, exist_ok=True)
-        temp_image_path = os.path.join(output_dir, f"temp_input_{int(time.time() * 1000)}.png")
-        
-        with open(temp_image_path, 'wb') as f:
-            f.write(image_bytes)
-        
-        return temp_image_path
-        
-    except Exception as e:
-        print(f"Warning: Failed to save image: {e}")
-        return None
-
-
 def setup_model_environment(model_path: str) -> None:
-    # if "fullattn" in model_path.lower():
-    #     os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
-    # else:
-    #     os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "VIDEO_SPARSE_ATTN"
-    os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
+    if "fullattn" in model_path.lower():
+        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
+    else:
+        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "VIDEO_SPARSE_ATTN"
     os.environ["FASTVIDEO_STAGE_LOGGING"] = "1"
 
 
@@ -193,41 +157,22 @@ class BaseModelDeployment:
             num_gpus=1,
             use_fsdp_inference=True,
             text_encoder_cpu_offload=config["text_encoder_cpu_offload"],
-            dmd_denoising_steps=[1000, 850, 700, 550, 350, 275, 200, 125], # TODO: hardocde for I2V
-            dit_precision="fp32",  # TODO: hardocde for I2V
             dit_cpu_offload=config["dit_cpu_offload"],
             vae_cpu_offload=config["vae_cpu_offload"],
             VSA_sparsity=config["VSA_sparsity"],
             enable_stage_verification=False,
         )
         self.default_params = SamplingParam.from_pretrained(self.model_path)
-        self.default_params.seed = 1000
-        self.default_params.num_frames = 73
-        self.default_params.width = 832
-        self.default_params.height = 480
 
     def generate_video(self, video_request: VideoGenerationRequest) -> VideoGenerationResponse:
         total_start_time = time.time()
         
         params = prepare_sampling_params(video_request, self.default_params)
 
-        # Save image if provided (for I2V)
-        image_path = None
-        if video_request.image_data:
-            image_path = save_image_from_base64(video_request.image_data, self.output_path)
-            if image_path is None:
-                return VideoGenerationResponse(
-                    video_data=None,
-                    seed=params.seed,
-                    success=False,
-                    error_message="Failed to save input image",
-                )
-
         inference_start_time = time.time()
         result = self.generator.generate_video(
             prompt=video_request.prompt,
             sampling_param=params,
-            image_path=image_path,
             save_video=False,
             return_frames=False,
         )
@@ -240,13 +185,6 @@ class BaseModelDeployment:
         encoding_time = time.time() - encoding_start_time
         
         total_time = time.time() - total_start_time
-        
-        # Clean up temporary image file
-        if image_path and os.path.exists(image_path):
-            try:
-                os.remove(image_path)
-            except Exception as e:
-                print(f"Warning: Failed to remove temporary image file {image_path}: {e}")
 
         return VideoGenerationResponse(
             video_data=video_data,
@@ -262,7 +200,7 @@ class BaseModelDeployment:
 
 
 @serve.deployment(
-    ray_actor_options={"num_cpus": 2, "num_gpus": 1, "runtime_env": {"conda": "demo-fv"}},
+    ray_actor_options={"num_cpus": 2, "num_gpus": 1, "runtime_env": {"conda": "fv"}},
 )
 class T2VModelDeployment(BaseModelDeployment):
     def __init__(self, t2v_model_path: str, output_path: str = "outputs"):
@@ -272,7 +210,7 @@ class T2VModelDeployment(BaseModelDeployment):
 
 
 @serve.deployment(
-    ray_actor_options={"num_cpus": 16, "num_gpus": 1, "runtime_env": {"conda": "demo-fv"}},
+    ray_actor_options={"num_cpus": 16, "num_gpus": 1, "runtime_env": {"conda": "fv"}},
 )
 class T2V14BModelDeployment(BaseModelDeployment):
     def __init__(self, t2v_14b_model_path: str, output_path: str = "outputs"):
@@ -283,32 +221,18 @@ class T2V14BModelDeployment(BaseModelDeployment):
         print("✅ T2V 14B model initialized successfully")
 
 
-@serve.deployment(
-    ray_actor_options={"num_cpus": 15, "num_gpus": 1, "runtime_env": {"conda": "demo-fv"}},
-)
-class I2VModelDeployment(BaseModelDeployment):
-    def __init__(self, i2v_model_path: str, output_path: str = "outputs"):
-        super().__init__(i2v_model_path, output_path)
-        # Override environment for I2V model
-        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "FLASH_ATTN"
-        self._initialize_generator(MODEL_CONFIGS["I2V-A14B"])
-        print("✅ I2V model initialized successfully")
-
-
 app = FastAPI()
 limiter = Limiter(key_func=get_remote_address)
 app.state.limiter = limiter
 app.add_exception_handler(RateLimitExceeded, _rate_limit_exceeded_handler)
 
 
-@serve.deployment(num_replicas=1, ray_actor_options={"num_cpus": 1})
+@serve.deployment(num_replicas=50, ray_actor_options={"num_cpus": 2})
 @serve.ingress(app)
 class FastVideoAPI:
 
-    def __init__(self, t2v_deployments: Dict[str, DeploymentHandle], i2v_deployments: Dict[str, DeploymentHandle] = None):
+    def __init__(self, t2v_deployments: Dict[str, DeploymentHandle]):
         self.t2v_deployments = t2v_deployments
-        self.i2v_deployments = i2v_deployments or {}
-        self.all_deployments = {**self.t2v_deployments, **self.i2v_deployments}
         
         # Initialize Prometheus metrics
         self.request_count = Counter('fastvideo_requests_total', 'Total FastVideo requests', ['model_type', 'status'])
@@ -333,10 +257,10 @@ class FastVideoAPI:
         model_name = self._get_model_name(video_request.model_path)
         
         try:
-            if video_request.model_path not in self.all_deployments:
+            if video_request.model_path not in self.t2v_deployments:
                 raise ValueError(f"Model {video_request.model_path} not found")
             
-            response_ref = self.all_deployments[video_request.model_path].generate_video.remote(video_request)
+            response_ref = self.t2v_deployments[video_request.model_path].generate_video.remote(video_request)
             response = await response_ref
             
             self._record_metrics(model_name, "success", time.time() - start_time, response)
@@ -367,21 +291,18 @@ class FastVideoAPI:
 
 
 def validate_configuration(model_paths: List[str], replicas: List[int]) -> None:
-    assert len(model_paths) > 0, "At least one model must be specified"
     assert len(model_paths) == len(replicas), "Number of models and replicas must match"
     assert sum(replicas) <= NUM_GPUS, f"Total replicas ({sum(replicas)}) must be <= {NUM_GPUS}"
     
     for model, replica_count in zip(model_paths, replicas):
-        assert model in SUPPORTED_MODELS, f"Model {model} not supported. Supported models: {SUPPORTED_MODELS}"
+        assert model in SUPPORTED_MODELS, f"Model {model} not supported"
         assert replica_count > 0, f"Replicas must be greater than 0"
 
 
 def start_ray_serve(
     *,
-    t2v_model_paths: str = "",
-    t2v_model_replicas: str = "",
-    i2v_model_paths: str = "",
-    i2v_model_replicas: str = "",
+    t2v_model_paths: str,
+    t2v_model_replicas: str,
     output_path: str = "outputs",
     host: str = "0.0.0.0",
     port: int = 8000,
@@ -389,39 +310,21 @@ def start_ray_serve(
     if not ray.is_initialized():
         ray.init()
 
-    # Parse T2V models
-    t2v_paths = [p.strip() for p in t2v_model_paths.split(",") if p.strip()]
-    t2v_reps = [int(r.strip()) for r in t2v_model_replicas.split(",") if r.strip()] if t2v_model_replicas else []
-    
-    # Parse I2V models
-    i2v_paths = [p.strip() for p in i2v_model_paths.split(",") if p.strip()]
-    i2v_reps = [int(r.strip()) for r in i2v_model_replicas.split(",") if r.strip()] if i2v_model_replicas else []
-    
-    # Validate configurations
-    all_paths = t2v_paths + i2v_paths
-    all_replicas = t2v_reps + i2v_reps
-    validate_configuration(all_paths, all_replicas)
+    model_paths = t2v_model_paths.split(",")
+    replicas = [int(r) for r in t2v_model_replicas.split(",")]
+    validate_configuration(model_paths, replicas)
 
-    # Create T2V deployments
     t2v_deps = {}
-    for model_path, replica_count in zip(t2v_paths, t2v_reps):
+    for model_path, replica_count in zip(model_paths, replicas):
         t2v_dep = T2VModelDeployment.options(num_replicas=replica_count).bind(model_path, output_path)
         t2v_deps[model_path] = t2v_dep
 
-    # Create I2V deployments
-    i2v_deps = {}
-    for model_path, replica_count in zip(i2v_paths, i2v_reps):
-        i2v_dep = I2VModelDeployment.options(num_replicas=replica_count).bind(model_path, output_path)
-        i2v_deps[model_path] = i2v_dep
-
-    api = FastVideoAPI.bind(t2v_deps, i2v_deps)
+    api = FastVideoAPI.bind(t2v_deps)
     serve.run(api, route_prefix="/", name="fast_video")
 
     print(f"Ray Serve backend started at http://{host}:{port}")
-    for model_path, replica_count in zip(t2v_paths, t2v_reps):
+    for model_path, replica_count in zip(model_paths, replicas):
         print(f"T2V Model: {model_path} | Replicas: {replica_count}")
-    for model_path, replica_count in zip(i2v_paths, i2v_reps):
-        print(f"I2V Model: {model_path} | Replicas: {replica_count}")
     print(f"Health check: http://{host}:{port}/health")
     print(f"Video generation endpoint: http://{host}:{port}/generate_video")
 
@@ -437,20 +340,12 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="FastVideo Ray Serve Backend")
     parser.add_argument("--t2v_model_paths",
                         type=str,
-                        default="",
+                        default="FastVideo/FastWan2.1-T2V-1.3B-Diffusers,FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
                         help="Comma separated list of paths to the T2V model(s)")
     parser.add_argument("--t2v_model_replicas",
                         type=str,
-                        default="",
+                        default="4,4",
                         help="Comma separated list of number of replicas for the T2V model(s)")
-    parser.add_argument("--i2v_model_paths",
-                        type=str,
-                        default="FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers",
-                        help="Comma separated list of paths to the I2V model(s)")
-    parser.add_argument("--i2v_model_replicas",
-                        type=str,
-                        default="1",
-                        help="Comma separated list of number of replicas for the I2V model(s)")
     parser.add_argument("--output_path",
                         type=str,
                         default="outputs",
@@ -466,21 +361,13 @@ if __name__ == "__main__":
     
     args = parser.parse_args()
 
-    # Parse and validate all models
-    t2v_paths = [p.strip() for p in args.t2v_model_paths.split(",") if p.strip()]
-    t2v_reps = [int(r.strip()) for r in args.t2v_model_replicas.split(",") if r.strip()] if args.t2v_model_replicas else []
-    i2v_paths = [p.strip() for p in args.i2v_model_paths.split(",") if p.strip()]
-    i2v_reps = [int(r.strip()) for r in args.i2v_model_replicas.split(",") if r.strip()] if args.i2v_model_replicas else []
-    
-    all_paths = t2v_paths + i2v_paths
-    all_replicas = t2v_reps + i2v_reps
-    validate_configuration(all_paths, all_replicas)
+    model_paths = args.t2v_model_paths.split(",")
+    replicas = [int(r) for r in args.t2v_model_replicas.split(",")]
+    validate_configuration(model_paths, replicas)
     
     start_ray_serve(
         t2v_model_paths=args.t2v_model_paths,
         t2v_model_replicas=args.t2v_model_replicas,
-        i2v_model_paths=args.i2v_model_paths,
-        i2v_model_replicas=args.i2v_model_replicas,
         output_path=args.output_path,
         host=args.host,
         port=args.port,
@@ -489,4 +376,4 @@ if __name__ == "__main__":
     setup_signal_handlers()
     print("✅ FastVideo backend is running. Press Ctrl-C to stop.")
     while True:
-        time.sleep(3600) 
+        time.sleep(3600) 

examples/inference/gradio/serving/gradio_frontend.py

@@ -1,774 +0,0 @@
-import argparse
-import os
-import requests
-import base64
-import time
-import json
-from pathlib import Path
-import tempfile
-from io import BytesIO
-
-import gradio as gr
-from PIL import Image
-
-from fastvideo.configs.sample.base import SamplingParam
-
-
-MODEL_PATH_MAPPING = {
-    "FastWan2.1-T2V-1.3B": "FastVideo/FastWan2.1-T2V-1.3B-Diffusers",
-    "FastWan2.2-TI2V-5B-FullAttn": "FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
-    "CausalWan2.2-I2V-A14B-Preview": "FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers",
-}
-
-
-class RayServeClient:
-    def __init__(self, backend_url: str):
-        self.backend_url = backend_url
-        self.session = requests.Session()
-    
-    def check_health(self) -> bool:
-        try:
-            response = self.session.get(f"{self.backend_url}/health", timeout=5)
-            return response.status_code == 200
-        except requests.exceptions.RequestException:
-            return False
-    
-    def generate_video(self, request_data: dict) -> dict:
-        start_time = time.time()
-        
-        try:
-            headers = {"Content-Type": "application/json"}
-            
-            response = self.session.post(
-                f"{self.backend_url}/generate_video",
-                json=request_data,
-                headers=headers,
-                timeout=900  # 15 minutes timeout for longer video generation
-            )
-            
-            round_trip_time = time.time() - start_time
-            
-            if response.status_code == 200:
-                result = response.json()
-                backend_total = result.get("total_time", 0)
-                network_time = round_trip_time - backend_total
-                result["network_time"] = network_time
-                return result
-            else:
-                return {"success": False, "error_message": f"HTTP {response.status_code}: {response.text}"}
-                
-        except requests.exceptions.RequestException as e:
-            return {"success": False, "error_message": f"Request failed: {str(e)}"}
-
-
-def save_video_from_base64(video_data: str, output_dir: str, prompt: str) -> str:
-    if not video_data:
-        return None
-    
-    try:
-        if video_data.startswith('data:video/'):
-            video_data = video_data.split(',')[1]
-        
-        video_bytes = base64.b64decode(video_data)
-        
-        safe_prompt = prompt[:50].replace(' ', '_').replace('/', '_').replace('\\', '_')
-        video_filename = f"{safe_prompt}.mp4"
-        video_path = os.path.join(output_dir, video_filename)
-        
-        os.makedirs(output_dir, exist_ok=True)
-        
-        with open(video_path, 'wb') as f:
-            f.write(video_bytes)
-        
-        return video_path
-        
-    except Exception as e:
-        print(f"Failed to save video: {e}")
-        return None
-
-
-def encode_image_to_base64(image_input) -> str:
-    """Encode an image file path or in-memory image to a base64 string."""
-    if image_input is None:
-        return None
-    
-    mime_types = {
-        '.jpg': 'image/jpeg',
-        '.jpeg': 'image/jpeg',
-        '.png': 'image/png',
-        '.gif': 'image/gif',
-        '.webp': 'image/webp',
-    }
-    
-    try:
-        if isinstance(image_input, str):
-            if not os.path.exists(image_input):
-                return None
-            
-            with open(image_input, 'rb') as f:
-                image_bytes = f.read()
-            
-            ext = os.path.splitext(image_input)[1].lower()
-            mime_type = mime_types.get(ext, 'image/jpeg')
-        elif isinstance(image_input, Image.Image):
-            buffer = BytesIO()
-            image_to_save = image_input.convert("RGB")
-            image_to_save.save(buffer, format="PNG")
-            image_bytes = buffer.getvalue()
-            mime_type = 'image/png'
-        else:
-            return None
-        
-        image_base64 = base64.b64encode(image_bytes).decode('utf-8')
-        return f"data:{mime_type};base64,{image_base64}"
-        
-    except Exception as e:
-        print(f"Failed to encode image: {e}")
-        return None
-
-
-# def create_timing_display(inference_time, encoding_time, network_time, total_time, stage_execution_times, num_frames):
-#     dit_denoising_time = f"{stage_execution_times[5]:.2f}s" if len(stage_execution_times) > 5 else "N/A"
-#     
-#     timing_html = f"""
-#     <div style="margin: 10px 0;">
-#         <h3 style="text-align: center; margin-bottom: 10px;">⏱️ Timing Breakdown</h3>
-#         <div style="display: grid; grid-template-columns: repeat(5, 1fr); gap: 10px; margin-bottom: 10px;">
-#             <div class="timing-card timing-card-highlight">
-#                 <div style="font-size: 20px;">🚀</div>
-#                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">DiT Denoising</div>
-#                 <div style="font-size: 16px; color: #ffa200; font-weight: bold;">{dit_denoising_time}</div>
-#             </div>
-#             <div class="timing-card">
-#                 <div style="font-size: 20px;">🧠</div>
-#                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">E2E (w. vae/text encoder)</div>
-#                 <div style="font-size: 16px; color: #2563eb;">{inference_time:.2f}s</div>
-#             </div>
-#             <div class="timing-card">
-#                 <div style="font-size: 20px;">🎬</div>
-#                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">Video Encoding</div>
-#                 <div style="font-size: 16px; color: #dc2626;">{encoding_time:.2f}s</div>
-#             </div>
-#             <div class="timing-card">
-#                 <div style="font-size: 20px;">🌐</div>
-#                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">Network Transfer</div>
-#                 <div style="font-size: 16px; color: #059669;">{network_time:.2f}s</div>
-#             </div>
-#             <div class="timing-card">
-#                 <div style="font-size: 20px;">📊</div>
-#                 <div style="font-weight: bold; margin: 3px 0; font-size: 14px;">Total Processing</div>
-#                 <div style="font-size: 18px; color: #0277bd;">{total_time:.2f}s</div>
-#             </div>
-#         </div>"""
-#     
-#     if inference_time > 0:
-#         fps = num_frames / inference_time
-#         timing_html += f"""
-#         <div class="performance-card" style="margin-top: 15px;">
-#             <span style="font-weight: bold;">Generation Speed: </span>
-#             <span style="font-size: 18px; color: #6366f1; font-weight: bold;">{fps:.1f} frames/second</span>
-#         </div>"""
-#     
-#     return timing_html + "</div>"
-
-
-def load_example_prompts():
-    def contains_chinese(text):
-        return any('\u4e00' <= char <= '\u9fff' for char in text)
-    
-    def load_from_file(filepath):
-        prompts, labels = [], []
-        try:
-            with open(filepath, "r", encoding='utf-8') as f:
-                for line in f:
-                    line = line.strip()
-                    if line and not contains_chinese(line):
-                        label = line[:100] + "..." if len(line) > 100 else line
-                        labels.append(label)
-                        prompts.append(line)
-        except Exception as e:
-            print(f"Warning: Could not read {filepath}: {e}")
-        return prompts, labels
-    
-    # Load prompts from prompts.txt
-    examples, example_labels = load_from_file("examples/inference/gradio/serving/prompts.txt")
-    
-    if not examples:
-        examples = ["A crowded rooftop bar buzzes with energy, the city skyline twinkling like a field of stars in the background."]
-        example_labels = ["Crowded rooftop bar at night"]
-    
-    # Load image mappings from JSON file
-    prompt_to_image = {}
-    # Try to find the JSON file relative to project root
-    possible_json_paths = [
-        Path("prompts/mixkit_i2v.jsonl"),
-        Path(__file__).parent.parent.parent.parent / "prompts" / "mixkit_i2v.jsonl",
-    ]
-    json_path = None
-    for path in possible_json_paths:
-        if path.exists():
-            json_path = path
-            break
-    
-    if json_path and json_path.exists():
-        try:
-            with open(json_path, "r", encoding='utf-8') as f:
-                data = json.load(f)
-                # Get the project root directory (parent of prompts directory)
-                project_root = json_path.parent.parent
-                for item in data:
-                    prompt_text = item.get("prompt", "").strip()
-                    image_path = item.get("image_path", "")
-                    if prompt_text and image_path:
-                        # Resolve image path relative to project root
-                        full_image_path = project_root / image_path
-                        if full_image_path.exists():
-                            prompt_to_image[prompt_text] = str(full_image_path.absolute())
-        except Exception as e:
-            print(f"Warning: Could not load image mappings from {json_path}: {e}")
-    
-    # Create image paths list matching the prompts
-    example_images = []
-    for prompt in examples:
-        # Try exact match first
-        image_path = prompt_to_image.get(prompt)
-        if not image_path:
-            # Try fuzzy match (case-insensitive, whitespace normalized)
-            normalized_prompt = " ".join(prompt.split())
-            for json_prompt, img_path in prompt_to_image.items():
-                normalized_json = " ".join(json_prompt.split())
-                if normalized_prompt.lower() == normalized_json.lower():
-                    image_path = img_path
-                    break
-        example_images.append(image_path if image_path and os.path.exists(image_path) else None)
-    
-    return examples, example_labels, example_images
-
-
-def create_gradio_interface(backend_url: str, default_params: dict[str, SamplingParam]):
-    
-    client = RayServeClient(backend_url)
-    
-    def is_i2v_model(model_name: str) -> bool:
-        """Check if the model is an I2V model."""
-        return "I2V" in model_name
-    
-    def generate_video(
-        prompt, negative_prompt, use_negative_prompt, guidance_scale,
-        num_frames, height, width, model_selection, input_image, progress
-    ):
-        # Use default seed value (randomize_seed disabled)
-        seed = 1000
-        randomize_seed = False
-        if not client.check_health():
-            return None, f"Backend is not available. Please check if Ray Serve is running at {backend_url}", ""
-        
-        # Check if I2V model requires an image
-        if is_i2v_model(model_selection) and not input_image:
-            return None, "I2V models require an input image. Please upload an image.", ""
-        
-        # Validate dimensions
-        max_pixels = 720 * 1280
-        if height * width > max_pixels:
-            return None, f"Video dimensions too large. Maximum: 720x1280 pixels", ""
-        
-        if progress:
-            progress(0.1, desc="Checking backend health...")
-        
-        # Encode image if provided
-        image_data = None
-        if input_image:
-            if progress:
-                progress(0.2, desc="Encoding input image...")
-            image_data = encode_image_to_base64(input_image)
-            if not image_data:
-                return None, "Failed to encode input image", ""
-        
-        request_data = {
-            "prompt": prompt,
-            "negative_prompt": negative_prompt,
-            "use_negative_prompt": use_negative_prompt,
-            "seed": seed,
-            "guidance_scale": guidance_scale,
-            "num_frames": num_frames,
-            "height": height,
-            "width": width,
-            "randomize_seed": randomize_seed,
-            "return_frames": False,
-            "image_data": image_data,
-            "model_path": MODEL_PATH_MAPPING.get(model_selection, "FastVideo/FastWan2.1-T2V-1.3B-Diffusers")
-        }
-        
-        if progress:
-            progress(0.4, desc="Generating video...")
-        
-        response = client.generate_video(request_data)
-        
-        if progress:
-            progress(0.8, desc="Processing response...")
-        
-        if response.get("success", False):
-            video_data = response.get("video_data", "")
-            used_seed = response.get("seed", seed)
-            # inference_time = response.get("inference_time", 0.0)
-            # encoding_time = response.get("encoding_time", 0.0)
-            # total_time = response.get("total_time", 0.0)
-            # network_time = response.get("network_time", 0.0)
-            # stage_execution_times = response.get("stage_execution_times", [])
-            
-            # timing_details = create_timing_display(
-            #     inference_time, encoding_time, network_time, total_time, 
-            #     stage_execution_times, num_frames
-            # )
-            
-            if video_data:
-                if progress:
-                    progress(0.9, desc="Saving video...")
-                
-                video_path = save_video_from_base64(video_data, "outputs", prompt)
-                
-                if progress:
-                    progress(1.0, desc="Generation complete!")
-                
-                if video_path and os.path.exists(video_path):
-                    return video_path, used_seed, ""
-                else:
-                    return None, "Failed to save video", ""
-            else:
-                return None, "No video data received from backend", ""
-        else:
-            error_msg = response.get("error_message", "Unknown error occurred")
-            return None, f"Generation failed: {error_msg}", ""
-    
-    examples, example_labels, example_images = load_example_prompts()
-    
-    theme = gr.themes.Base().set(
-        button_primary_background_fill="#2563eb",
-        button_primary_background_fill_hover="#1d4ed8",
-        button_primary_text_color="white",
-        slider_color="#2563eb",
-        checkbox_background_color_selected="#2563eb",
-    )
-    
-    def get_default_values(model_name):
-        # model_path = MODEL_PATH_MAPPING.get(model_name)
-        # if model_path and model_path in default_params:
-        #     params = default_params[model_path]
-        #     return {
-        #         'height': params.height,
-        #         'width': params.width,
-        #         'num_frames': params.num_frames,
-        #         'guidance_scale': params.guidance_scale,
-        #     }
-        
-        return {
-            'height': 480,
-            'width': 832,
-            'num_frames': 73,
-        }
-    
-    # Get available models based on what's loaded
-    available_models = []
-    for model_name, model_path in MODEL_PATH_MAPPING.items():
-        if model_path in default_params:
-            available_models.append(model_name)
-    
-    # Select first available model as default
-    default_model = available_models[0] if available_models else "FastWan2.1-T2V-1.3B"
-    initial_values = get_default_values(default_model)
-    initial_show_image = is_i2v_model(default_model)
-    
-    with gr.Blocks(title="CausalWan", theme=theme) as demo:
-        gr.Image("assets/logos/logo.svg", show_label=False, container=False, height=80)
-        gr.HTML("""
-        <div style="text-align: center; margin-bottom: 10px;">
-            <p style="font-size: 18px;"> Make Video Generation Go Blurrrrrrr </p>
-            <p style="font-size: 18px;"> <a href="https://github.com/hao-ai-lab/FastVideo/tree/main" target="_blank">Code</a> | <a href="https://hao-ai-lab.github.io/blogs/fastvideo_causalwan_preview/" target="_blank">Blog</a> | <a href="https://hao-ai-lab.github.io/FastVideo/" target="_blank">Docs</a>  </p>
-        </div>
-        """)
-        
-        with gr.Accordion("🎥 What Is FastVideo?", open=False):
-            gr.HTML("""
-            <div style="padding: 20px; line-height: 1.6;">
-                <p style="font-size: 16px; margin-bottom: 15px;">
-                    FastVideo is an inference and post-training framework for diffusion models. It features an end-to-end unified pipeline for accelerating diffusion models, starting from data preprocessing to model training, finetuning, distillation, and inference. FastVideo is designed to be modular and extensible, allowing users to easily add new optimizations and techniques. Whether it is training-free optimizations or post-training optimizations, FastVideo has you covered.
-                </p>
-            </div>
-            """)
-        
-        with gr.Row():
-            model_selection = gr.Dropdown(
-                choices=available_models,
-                value=default_model,
-                label="Select Model",
-                interactive=True
-            )
-
-        with gr.Row():
-            example_dropdown = gr.Dropdown(
-                choices=example_labels,
-                label="Example Prompts",
-                value=None,
-                interactive=True,
-                allow_custom_value=False
-            )
-        
-        with gr.Row():
-            with gr.Column(scale=6):
-                prompt = gr.Text(
-                    label="Prompt",
-                    show_label=False,
-                    max_lines=3,
-                    placeholder="Describe your scene...",
-                    container=False,
-                    lines=3,
-                    autofocus=True,
-                )
-            with gr.Column(scale=1, min_width=120, elem_classes="center-button"):
-                run_button = gr.Button("Run", variant="primary", size="lg")
-        
-        with gr.Row():
-            with gr.Column():
-                error_output = gr.Text(label="Error", visible=False)
-                # timing_display = gr.Markdown(label="Timing Breakdown", visible=False)
-
-        with gr.Row(equal_height=False):
-            with gr.Column(scale=1):
-                with gr.Tabs():
-                    with gr.Tab("Input Image", visible=initial_show_image) as image_tab:
-                        gr.Markdown("**Please make sure you upload a 480x832 image**")
-                        input_image = gr.Image(
-                            label="",
-                            type="pil",
-                            height=400,
-                        )
-                    
-                    with gr.Tab("Advanced Options"):
-                        with gr.Group():
-                            with gr.Row():
-                                height = gr.Number(
-                                    label="Height",
-                                    value=initial_values['height'],
-                                    interactive=False,
-                                    container=True
-                                )
-                                width = gr.Number(
-                                    label="Width",
-                                    value=initial_values['width'],
-                                    interactive=False,
-                                    container=True
-                                )
-                            
-                            with gr.Row():
-                                num_frames = gr.Number(
-                                    label="Number of Frames",
-                                    value=initial_values['num_frames'],
-                                    interactive=False,
-                                    container=True
-                                )
-                                guidance_scale = gr.Number(
-                                    label="Guidance Scale",
-                                    value=1.0,
-                                    interactive=False,
-                                    container=True
-                                )
-                            
-                            with gr.Row():
-                                use_negative_prompt = gr.Checkbox(
-                                    label="Use negative prompt", value=False)
-                                negative_prompt = gr.Text(
-                                    label="Negative prompt",
-                                    max_lines=3,
-                                    lines=3,
-                                    placeholder="Enter a negative prompt",
-                                    visible=False,
-                                )
-
-                            # randomize_seed = gr.Checkbox(label="Randomize seed", value=False)
-                            seed_output = gr.Number(label="Used Seed", value=1000)
-        
-            with gr.Column(scale=1):
-                result = gr.Video(
-                    label="Generated Video", 
-                    show_label=True,
-                    height=500,
-                    container=True,
-                    autoplay=True,
-                )
-        
-        gr.HTML("""
-        <style>
-        .center-button {
-            display: flex !important;
-            justify-content: center !important;
-            height: 100% !important;
-            padding-top: 1.4em !important;
-        }
-        
-        .gradio-container {
-            max-width: 1400px !important;
-            margin: 0 auto !important;
-        }
-        
-        .gr-number input[readonly] {
-            background-color: var(--background-fill-secondary) !important;
-            border: 1px solid var(--border-color-primary) !important;
-            color: var(--body-text-color-subdued) !important;
-            cursor: default !important;
-            text-align: center !important;
-            font-weight: 500 !important;
-        }
-        </style>
-        """)
-        
-        def on_example_select(example_label):
-            if example_label and example_label in example_labels:
-                index = example_labels.index(example_label)
-                selected_prompt = examples[index]
-                selected_image_path = example_images[index] if index < len(example_images) else None
-                
-                if selected_image_path and os.path.exists(selected_image_path):
-                    try:
-                        with Image.open(selected_image_path) as img:
-                            selected_image = img.convert("RGB")
-                    except Exception:
-                        selected_image = None
-                else:
-                    selected_image = None
-                
-                return selected_prompt, selected_image
-            return "", None
-        
-        example_dropdown.change(
-            fn=on_example_select,
-            inputs=example_dropdown,
-            outputs=[prompt, input_image],
-        )
-        
-        gr.HTML("""
-        <div style="text-align: center; margin-top: 10px; margin-bottom: 15px;">
-            <p style="font-size: 16px; margin: 0;">The compute for this demo is generously provided by <a href="https://www.gmicloud.ai/" target="_blank">GMI Cloud</a>.  Note that this demo is meant as a preview of our distilled I2V model. Outside of few-step distillation, we have not yet fully optimized it for speed. Stay tuned for updates!</p>
-        </div>
-        """)
-        
-        use_negative_prompt.change(
-            fn=lambda x: gr.update(visible=x),
-            inputs=use_negative_prompt,
-            outputs=negative_prompt,
-        )
-        
-        def on_model_selection_change(selected_model):
-            if not selected_model:
-                selected_model = "FastWan2.1-T2V-1.3B"
-            
-            model_path = MODEL_PATH_MAPPING.get(selected_model)
-            show_image_input = is_i2v_model(selected_model)
-            
-            if model_path and model_path in default_params:
-                params = default_params[model_path]
-                return (
-                    gr.update(value=params.height),
-                    gr.update(value=params.width),
-                    gr.update(value=params.num_frames),
-                    gr.update(value=params.guidance_scale),
-                    gr.update(visible=show_image_input),
-                )
-            
-            return (
-                gr.update(value=448),
-                gr.update(value=832),
-                gr.update(value=20),
-                gr.update(value=3.0),
-                gr.update(visible=show_image_input),
-            )
-        
-        model_selection.change(
-            fn=on_model_selection_change,
-            inputs=model_selection,
-            outputs=[height, width, num_frames, guidance_scale, image_tab],
-        )
-        
-        def handle_generation(*args, progress=None, request: gr.Request = None):
-            model_selection, prompt, negative_prompt, use_negative_prompt, guidance_scale, num_frames, height, width, input_image = args
-            
-            result_path, seed_or_error, _ = generate_video(
-                prompt, negative_prompt, use_negative_prompt, guidance_scale, 
-                num_frames, height, width, model_selection, input_image, progress
-            )
-            
-            if result_path and os.path.exists(result_path):
-                return (
-                    result_path, 
-                    seed_or_error, 
-                    gr.update(visible=False),
-                )
-            else:
-                return (
-                    None, 
-                    seed_or_error, 
-                    gr.update(visible=True, value=seed_or_error),
-                )
-        
-        run_button.click(
-            fn=handle_generation,
-            inputs=[
-                model_selection,
-                prompt,
-                negative_prompt,
-                use_negative_prompt,
-                guidance_scale,
-                num_frames,
-                height,
-                width,
-                # randomize_seed,
-                input_image,
-            ],
-            outputs=[result, seed_output, error_output],  # timing_display removed
-            concurrency_limit=20,
-        )
-    
-    return demo
-
-
-def main():
-    parser = argparse.ArgumentParser(description="FastVideo Gradio Frontend")
-    parser.add_argument("--backend_url", type=str, default="http://localhost:8000",
-                        help="URL of the Ray Serve backend")
-    parser.add_argument("--t2v_model_paths", type=str,
-                        default="",
-                        help="Comma separated list of paths to the T2V model(s)")
-    parser.add_argument("--i2v_model_paths", type=str,
-                        default="FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers",
-                        help="Comma separated list of paths to the I2V model(s)")
-    parser.add_argument("--host", type=str, default="0.0.0.0",
-                        help="Host to bind to")
-    parser.add_argument("--port", type=int, default=7860,
-                        help="Port to bind to")
-    
-    args = parser.parse_args()
-    
-    default_params = {}
-    
-    # Load T2V model params
-    t2v_paths = [p.strip() for p in args.t2v_model_paths.split(",") if p.strip()]
-    for model_path in t2v_paths:
-        default_params[model_path] = SamplingParam.from_pretrained(model_path)
-    
-    # Load I2V model params
-    i2v_paths = [p.strip() for p in args.i2v_model_paths.split(",") if p.strip()]
-    for model_path in i2v_paths:
-        default_params[model_path] = SamplingParam.from_pretrained(model_path)
-    
-    demo = create_gradio_interface(args.backend_url, default_params)
-    
-    print(f"Starting Gradio frontend at http://{args.host}:{args.port}")
-    print(f"Backend URL: {args.backend_url}")
-    print(f"T2V Models: {args.t2v_model_paths}")
-    if args.i2v_model_paths:
-        print(f"I2V Models: {args.i2v_model_paths}")
-    
-    from fastapi import FastAPI, Request, HTTPException
-    from fastapi.responses import HTMLResponse, FileResponse
-    import uvicorn
-    
-    app = FastAPI()
-    
-    @app.get("/logo.svg")
-    def get_logo():
-        return FileResponse(
-            "assets/logos/logo.svg", 
-            media_type="image/svg+xml",
-            headers={
-                "Cache-Control": "public, max-age=3600",
-                "Access-Control-Allow-Origin": "*"
-            }
-        )
-    
-    @app.get("/favicon.ico")
-    def get_favicon():
-        favicon_path = "assets/logos/icon_simple.svg"
-        
-        if os.path.exists(favicon_path):
-            return FileResponse(
-                favicon_path, 
-                media_type="image/svg+xml",
-                headers={
-                    "Cache-Control": "public, max-age=3600",
-                    "Access-Control-Allow-Origin": "*"
-                }
-            )
-        else:
-            raise HTTPException(status_code=404, detail="Favicon not found")
-    
-    @app.get("/", response_class=HTMLResponse)
-    def index(request: Request):
-        base_url = str(request.base_url).rstrip('/')
-        return f"""
-        <!DOCTYPE html>
-        <html lang="en">
-        <head>
-            <meta charset="UTF-8" />
-            <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-            
-            <title>CausalWan</title>
-            <meta name="title" content="CausalWan">
-            <meta name="description" content="Make video generation go blurrrrrrr">
-            <meta name="keywords" content="FastVideo, video generation, AI, machine learning, CausalWan">
-            
-            <meta property="og:type" content="website">
-            <meta property="og:url" content="{base_url}/">
-            <meta property="og:title" content="CausalWan">
-            <meta property="og:description" content="Make video generation go blurrrrrrr">
-            <meta property="og:image" content="{base_url}/logo.svg">
-            <meta property="og:image:width" content="1200">
-            <meta property="og:image:height" content="630">
-            <meta property="og:site_name" content="CausalWan">
-            
-            <meta property="twitter:card" content="summary_large_image">
-            <meta property="twitter:url" content="{base_url}/">
-            <meta property="twitter:title" content="CausalWan">
-            <meta property="twitter:description" content="Make video generation go blurrrrrrr">
-            <meta property="twitter:image" content="{base_url}/logo.svg">
-            <link rel="icon" type="image/png" sizes="32x32" href="/favicon.ico">
-            <link rel="icon" type="image/png" sizes="16x16" href="/favicon.ico">
-            <link rel="apple-touch-icon" href="/favicon.ico">
-            <style>
-                body, html {{
-                    margin: 0;
-                    padding: 0;
-                    height: 100%;
-                    overflow: hidden;
-                }}
-                iframe {{
-                    width: 100%;
-                    height: 100vh;
-                    border: none;
-                }}
-            </style>
-        </head>
-        <body>
-            <iframe src="/gradio" width="100%" height="100%" style="border: none;"></iframe>
-        </body>
-        </html>
-        """
-    
-    app = gr.mount_gradio_app(
-        app, 
-        demo, 
-        path="/gradio",
-        root_path="/gradio",
-        allowed_paths=[
-            os.path.abspath("outputs"), 
-            os.path.abspath("fastvideo-logos"),
-            os.path.abspath("prompts"),
-            os.path.abspath("images"),
-            os.path.abspath(tempfile.gettempdir()),
-            os.path.abspath(os.path.join(tempfile.gettempdir(), "gradio")),
-        ]
-    )
-    
-    uvicorn.run(app, host=args.host, port=args.port)
-
-
-if __name__ == "__main__":
-    main() 

examples/inference/gradio/serving/prompts.txt

@@ -1,15 +0,0 @@
-Some friends dancing and having fun together in circles, at a party surrounded by colored lights at a party, in a fancy old place, in a view from below them.
-A man wearing grey shorts jumps rope in a gym, weights and gym equipment in the background.
-Flying over a peninsula covered in bushy trees, while discovering the sea around it, painted a beautiful turquoise blue, on a sunny day.
-Skillful cyclist doing a wheelie on a bike while riding through a forest, on a dirt road, surrounded by many trees, in the morning.
-In the video, a lone rider guides a majestic horse across an expansive, open field as the sun sets in the background. The rider, dressed in a classic blue shirt and wide-brimmed hat, sits confidently in the saddle, silhouetted against the warm glow of the evening sky. The horse moves gracefully, its mane and tail flowing with each step, creating a sense of harmony between horse and rider. Surrounding the pair, towering trees form a natural border, their leaves gently rustling in the breeze. The shadows lengthen on the ground, accentuating the serene and timeless feel of the scene. The distant hills and wooden fences frame the horizon, adding depth to the tranquil landscape. A few horses graze peacefully in the background, blending into the pastoral setting. The overall ambiance evokes a sense of calmness and quietude, capturing a perfect moment in the golden light of dusk.
-A silver SUV drives along a winding, snow-covered mountain road, with dense pine trees blanketed in snow lining both sides. The scene is serene, with the vehicle moving smoothly, possibly on a winter journey or vacation. As the SUV disappears around the bend, another, darker SUV follows, creating a sense of motion and perspective on the snow-dusted asphalt. The towering, snow-laden rock formation to the right contrasts with the dark green of the pines, highlighting the peacefulness of the wintry landscape.
-A man and a woman playing in a field with grass, during a bright afternoon, while cars pass by in the distance.
-A saxophonist wearing a blazer dances while playing a song in a park.
-Romantic couple embracing and looking at each other in the middle of a forest, during a break on a road trip through nature.
-Young woman cleaning her house decorated with plants and decorations, while dancing happily to music in her headphones.
-Man dressed in 80's style dances very happily in his kitchen while listening to music on his radio and drinking wine.
-Pair of jazz musicians performing a song with their saxophone and trombone on an abandoned train.
-A young woman with short hair wearing pink sunglasses chews gum and makes a bubble gum with the city in the background.
-Natural aerial landscape with a relief covered with abundant trees and vegetation and a thick layer of mist.
-Loving couple sitting on a log on the shore of a lake outside, sharing an affectionate hug.

examples/inference/gradio/serving/start.sh

@@ -1,5 +0,0 @@
-python examples/inference/gradio/serving/start_ray_serve_app.py \
---t2v_model_paths "" \
---t2v_model_replicas "" \
---i2v_model_paths "FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers" \
---i2v_model_replicas "1"

examples/inference/gradio/start.sh

@@ -0,0 +1,3 @@
+python examples/inference/gradio/start_ray_serve_app.py \
+--t2v_model_paths "FastVideo/FastWan2.1-T2V-1.3B-Diffusers,FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers" \
+--t2v_model_replicas "4,4"