[docs] simplify FastH3 model card

README.md

@@ -22,74 +22,47 @@ tags:
 
 # FastVideo-FastH3-4-step-Preview-v1-LoRA
 
-Private umbrella repository for the four rank-64 LoRA variants in the
-[FastH3 4-Step Preview v1
-family](https://huggingface.co/collections/FastVideo/fastvideo-fasth3).
-Each adapter reconstructs one corresponding distilled transformer from the
-MiniMax H3 base model.
+The compact LoRA releases for FastH3 Preview v1 from
+[FastVideo](https://github.com/hao-ai-lab/FastVideo). Each adapter reconstructs
+one four-forward FastH3 transformer from the MiniMax H3 base model.
 
-> **VSA requirement:** The three VSA adapters must be loaded and run with
-> FastVideo's Video Sparse Attention (VSA-H3) backend and kernel enabled. VSA
-> must be enabled before adapter loading so the compression-gate modules exist
-> for the adapter's `.set_weight` tensors. Dense attention is not a supported
-> substitute. The Dense / Data-Free adapter does not require VSA.
+[Blog](https://haoailab.com/blogs/fasth3-preview/) ·
+[Recommended full checkpoint](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree) ·
+[FastH3 collection](https://huggingface.co/collections/FastVideo/fastvideo-fasth3)
 
-## Variants
-
-| Variant | Adapter path | Full checkpoint | Attention | Training source | Step |
-|---|---|---|---|---|---:|
-| VSA / Data-Free | `vsa-datafree/adapter_model.safetensors` | [checkpoint](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree) | VSA, 90% sparse, tile 64 | Prompts only | 1300 |
-| VSA / Synthetic | `vsa-synthetic-step1300/adapter_model.safetensors` | [checkpoint](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1300) | VSA, 90% sparse, tile 64 | Synthetic Base-H3 videos | 1300 |
-| VSA / Synthetic | `vsa-synthetic-step1900/adapter_model.safetensors` | [checkpoint](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1900) | VSA, 90% sparse, tile 64 | Synthetic Base-H3 videos | 1900 |
-| Dense / Data-Free | `dense-datafree/adapter_model.safetensors` | [checkpoint](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree) | Dense | Prompts only | 1000 |
-
-`adapter_manifest.json` pins the source repository revision, LFS SHA-256,
-byte size, and full-checkpoint mapping for every file.
-
-The VSA / Data-Free adapter is the historical v10.5 step-1300 artifact. It
-predates later score-clock and FastGen-alignment corrections and is retained
-for exact v10.5 reproduction.
+> The three VSA adapters require FastVideo's VSA-H3 backend and kernel. Use
+> FastVideo's launchers rather than a generic PEFT loader; these adapters also
+> contain exact delta and VSA gate tensors.
 
-## Download one adapter
-
-Download only the variant you need:
+## Run the recommended adapter
 
 ```bash
-hf download FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA \
-  vsa-synthetic-step1900/adapter_model.safetensors \
-  --local-dir ./FastH3-Preview-v1-LoRA
+git clone https://github.com/hao-ai-lab/FastVideo.git
+cd FastVideo
+uv venv --python 3.12
+source .venv/bin/activate
+UV_TORCH_BACKEND=cu130 uv pip install -e ".[fasth3]"
 ```
 
-Then pass the resulting local `.safetensors` file as FastVideo's `lora_path`.
-The adapter must be supplied while constructing the pipeline, before weights
-are sharded.
-
-Do not pass the bare combined repository ID as `lora_path`: it contains four
-adapter files. Select and download one explicit variant first.
-
-Use the trained four-jump ladder `[999, 749, 500, 250]`, represented by five
-scheduler points, with guidance scale 1.0. VSA variants require FastVideo's H3
-VSA backend; the dense variant must run without VSA.
-
-## Adapter format
-
-These are FastVideo extended LoRA files, not pure PEFT adapters. In addition to
-rank-64 `lora_A` and `lora_B` factors, they can contain exact `.diff` and
-`.diff_b` tensors. VSA variants also contain `.set_weight` tensors for the
-compression gates that do not exist in the base MiniMax H3 transformer. A
-loader that ignores these keys will not reconstruct the distilled model.
-
-## Source repositories
+```bash
+bash examples/inference/basic/run_fasth3_lora_preview_vsa_datafree.sh \
+  --prompt "your prompt" \
+  --no-warmup \
+  --repeats 1
+```
 
-The original private repositories remain available for provenance and
-backward compatibility:
+Set `FASTH3_LORA_STRENGTH` to change the adapter strength from its default of
+`1.0`.
 
-- `FastVideo/FastH3-4-step-v1-LoRA`
-- `FastVideo/FastH3-4-step-v1.1-LoRA`
-- `FastVideo/FastH3-4-step-v1.2-LoRA`
-- `FastVideo/FastH3-Dense-4-step-v1-LoRA`
+## Variants
 
-## License
+| Adapter | Full checkpoint | FastVideo launcher |
+|---|---|---|
+| [VSA / Data-Free](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA/tree/main/vsa-datafree) | [VSA / Data-Free](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree) | `run_fasth3_lora_preview_vsa_datafree.sh` |
+| [VSA / Synthetic, step 1300](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA/tree/main/vsa-synthetic-step1300) | [VSA / Synthetic, step 1300](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1300) | `run_fasth3_lora_preview_vsa_synthetic_step1300.sh` |
+| [VSA / Synthetic, step 1900](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA/tree/main/vsa-synthetic-step1900) | [VSA / Synthetic, step 1900](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1900) | `run_fasth3_lora_preview_vsa_synthetic_step1900.sh` |
+| [Dense / Data-Free](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA/tree/main/dense-datafree) | [Dense / Data-Free](https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree) | `run_fasth3_lora_preview_dense_datafree.sh` |
 
-These adapters inherit the MiniMax H3 Community License. Review its territory
-and acceptable-use terms before use or redistribution.
+The VSA launchers select VSA-H3 automatically; the dense launcher disables it.
+This preview is for text-to-audio-video generation and inherits the
+[MiniMax H3 Community License](LICENSE).