wlsaidhi commited on
Commit
211fba6
·
verified ·
1 Parent(s): d9ff03c

[docs]: align FastH3 Preview v1 naming

Browse files

Update the model card and package self-references after the server-side repository rename.

README.md CHANGED
@@ -16,16 +16,18 @@ tags:
16
  - minimax-h3
17
  - fastvideo
18
  - fasth3
 
19
  ---
20
 
21
- # FastVideo-FastH3-Dense-4-step-v1
22
 
23
- Private FastVideo evaluation snapshot of the `dmd2_sp4_fsdp32_v12_datafree_mixed_dense_fa4` MiniMax-H3
24
- student at training step 1000. It generates synchronized video and audio
25
- with four transformer forwards.
 
26
 
27
- This repository is intentionally private. It is an internal checkpoint for
28
- evaluation and UI integration, not a public release recommendation.
29
 
30
  ## Checkpoint identity
31
 
@@ -61,29 +63,32 @@ export FASTVIDEO_DMD_DENOISING_STEPS=999,749,500,250
61
  export FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
62
 
63
  python examples/inference/basic/basic_minimax_h3_t2v.py \
64
- --model-path FastVideo/FastVideo-FastH3-Dense-4-step-v1 \
65
  --prompt '<MiniMax-H3 T2AV prompt>' \
66
  --steps 5 \
67
  --num-gpus 4
68
  ```
69
 
70
  The five scheduler grid points execute the four trained jumps. Use guidance
71
- scale 1.0 and dense FlashAttention (the FA4 route on the pinned GB200 stack). Compilation is an optional runtime
72
- optimization; first validate output parity on the pinned FastVideo commit.
 
73
 
74
  ## Training summary
75
 
76
- - Method: data-free DMD2 with carried backward-simulation ODE rollouts. Training consumes text conditioning and student-generated latents rather than target video latents.
 
 
77
  - Student grid: `[999, 749, 500, 250]`.
78
- - Student attention: dense FlashAttention (the FA4 route on the pinned GB200 stack).
 
79
  - Global batch size: 64.
80
  - Student and critic learning rate: `2e-06` and
81
  `2e-06`.
82
  - Precision: fp32 training state and bf16 inference export.
83
- - Training topology: 32 GB200 GPUs, SP=4,
84
  HSDP replicate=1 / shard=32.
85
- - Strict continuous score clock: `True`;
86
- shift `2.4`, warp max `0.999`.
87
 
88
  ## Limitations
89
 
 
16
  - minimax-h3
17
  - fastvideo
18
  - fasth3
19
+ - preview
20
  ---
21
 
22
+ # FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree
23
 
24
+ FastH3 4-Step Preview v1 checkpoint from FastVideo. This variant is the
25
+ `dmd2_sp4_fsdp32_v12_datafree_mixed_dense_fa4` data-free DMD2 MiniMax-H3
26
+ student at training step 1000. It generates synchronized video and audio with
27
+ four transformer forwards and uses dense FlashAttention.
28
 
29
+ See the complete [FastH3 4-Step Preview v1
30
+ collection](https://huggingface.co/collections/FastVideo/fastvideo-fasth3).
31
 
32
  ## Checkpoint identity
33
 
 
63
  export FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
64
 
65
  python examples/inference/basic/basic_minimax_h3_t2v.py \
66
+ --model-path FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree \
67
  --prompt '<MiniMax-H3 T2AV prompt>' \
68
  --steps 5 \
69
  --num-gpus 4
70
  ```
71
 
72
  The five scheduler grid points execute the four trained jumps. Use guidance
73
+ scale 1.0 and dense FlashAttention (the FA4 route on the pinned B200 stack).
74
+ Compilation is an optional runtime optimization; first validate output parity
75
+ on the pinned FastVideo commit.
76
 
77
  ## Training summary
78
 
79
+ - Method: data-free DMD2 with carried backward-simulation ODE rollouts.
80
+ Training consumes text conditioning and student-generated latents rather
81
+ than target video latents.
82
  - Student grid: `[999, 749, 500, 250]`.
83
+ - Student attention: dense FlashAttention (the FA4 route on the pinned B200
84
+ stack).
85
  - Global batch size: 64.
86
  - Student and critic learning rate: `2e-06` and
87
  `2e-06`.
88
  - Precision: fp32 training state and bf16 inference export.
89
+ - Training topology: 32 B200 GPUs, SP=4,
90
  HSDP replicate=1 / shard=32.
91
+ - Continuous score-clock shift: `2.4`; warp max: `0.999`.
 
92
 
93
  ## Limitations
94
 
fastvideo_inference.json CHANGED
@@ -11,7 +11,7 @@
11
  ],
12
  "fastvideo_commit": "aa6e6c1c1e15f879011a47996c7d6bc8381eaf89",
13
  "guidance_scale": 1.0,
14
- "model_id": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
15
  "num_gpus": 4,
16
  "num_inference_steps": 5,
17
  "schema_version": "fasth3-inference-contract-v1",
 
11
  ],
12
  "fastvideo_commit": "aa6e6c1c1e15f879011a47996c7d6bc8381eaf89",
13
  "guidance_scale": 1.0,
14
+ "model_id": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
15
  "num_gpus": 4,
16
  "num_inference_steps": 5,
17
  "schema_version": "fasth3-inference-contract-v1",
modular_model_index.json CHANGED
@@ -6,7 +6,7 @@
6
  "diffusers",
7
  "MiniMaxH3Scheduler",
8
  {
9
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
10
  "revision": null,
11
  "subfolder": "audio_scheduler",
12
  "type_hint": [
@@ -20,7 +20,7 @@
20
  "diffusers",
21
  "AutoencoderKLMiniMaxH3Audio",
22
  {
23
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
24
  "revision": null,
25
  "subfolder": "audio_vae",
26
  "type_hint": [
@@ -34,7 +34,7 @@
34
  "transformers",
35
  "Qwen3VLProcessor",
36
  {
37
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
38
  "revision": null,
39
  "subfolder": "processor",
40
  "type_hint": [
@@ -48,7 +48,7 @@
48
  "diffusers",
49
  "MiniMaxH3Scheduler",
50
  {
51
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
52
  "revision": null,
53
  "subfolder": "scheduler",
54
  "type_hint": [
@@ -62,7 +62,7 @@
62
  "transformers",
63
  "Qwen3VLForConditionalGeneration",
64
  {
65
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
66
  "revision": null,
67
  "subfolder": "text_encoder",
68
  "type_hint": [
@@ -76,7 +76,7 @@
76
  "transformers",
77
  "Qwen2TokenizerFast",
78
  {
79
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
80
  "revision": null,
81
  "subfolder": "tokenizer",
82
  "type_hint": [
@@ -90,7 +90,7 @@
90
  "diffusers",
91
  "MiniMaxH3Transformer3DModel",
92
  {
93
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
94
  "revision": null,
95
  "subfolder": "transformer",
96
  "type_hint": [
@@ -118,7 +118,7 @@
118
  "diffusers",
119
  "AutoencoderKLMiniMaxH3",
120
  {
121
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
122
  "revision": null,
123
  "subfolder": "vae",
124
  "type_hint": [
 
6
  "diffusers",
7
  "MiniMaxH3Scheduler",
8
  {
9
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
10
  "revision": null,
11
  "subfolder": "audio_scheduler",
12
  "type_hint": [
 
20
  "diffusers",
21
  "AutoencoderKLMiniMaxH3Audio",
22
  {
23
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
24
  "revision": null,
25
  "subfolder": "audio_vae",
26
  "type_hint": [
 
34
  "transformers",
35
  "Qwen3VLProcessor",
36
  {
37
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
38
  "revision": null,
39
  "subfolder": "processor",
40
  "type_hint": [
 
48
  "diffusers",
49
  "MiniMaxH3Scheduler",
50
  {
51
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
52
  "revision": null,
53
  "subfolder": "scheduler",
54
  "type_hint": [
 
62
  "transformers",
63
  "Qwen3VLForConditionalGeneration",
64
  {
65
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
66
  "revision": null,
67
  "subfolder": "text_encoder",
68
  "type_hint": [
 
76
  "transformers",
77
  "Qwen2TokenizerFast",
78
  {
79
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
80
  "revision": null,
81
  "subfolder": "tokenizer",
82
  "type_hint": [
 
90
  "diffusers",
91
  "MiniMaxH3Transformer3DModel",
92
  {
93
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
94
  "revision": null,
95
  "subfolder": "transformer",
96
  "type_hint": [
 
118
  "diffusers",
119
  "AutoencoderKLMiniMaxH3",
120
  {
121
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
122
  "revision": null,
123
  "subfolder": "vae",
124
  "type_hint": [
provenance.json CHANGED
@@ -13,7 +13,7 @@
13
  ],
14
  "fastvideo_commit": "aa6e6c1c1e15f879011a47996c7d6bc8381eaf89",
15
  "lineage_parent_run_id": "20260825T010407Z-v12-datafree-dense-fa4",
16
- "model_id": "FastVideo/FastVideo-FastH3-Dense-4-step-v1",
17
  "schema_version": "fasth3-checkpoint-provenance-v1",
18
  "source_config_sha256": "50519042643df51fd115dce3294ee38a703a9cf496c28560dd07bb265d60b7dd",
19
  "student_dtype": "bfloat16",
 
13
  ],
14
  "fastvideo_commit": "aa6e6c1c1e15f879011a47996c7d6bc8381eaf89",
15
  "lineage_parent_run_id": "20260825T010407Z-v12-datafree-dense-fa4",
16
+ "model_id": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree",
17
  "schema_version": "fasth3-checkpoint-provenance-v1",
18
  "source_config_sha256": "50519042643df51fd115dce3294ee38a703a9cf496c28560dd07bb265d60b7dd",
19
  "student_dtype": "bfloat16",