nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[docs] v2: GPU bring-up results — 20 models verified on H100

Browse Source 
V2_PORTING_STATUS.md now records the GPU bring-up outcome: 20 models generate
real video/audio on H100 (the 7 prior + 13 newly-ported), with the remaining
split into fastvideo/env-blocked (SLA/VSA kernels, transformers incompat, fastvideo
registry/flash_attn gaps) and HF-access-blocked (gated cosmos2/flux2/sd35) — none
a v2 recipe bug.
This commit is contained in:
SolitaryThinker
2026-07-04 17:15:34 +00:00
committed by Peiyuan Zhang
parent 0ac5367b54
commit 7ef2d9083e
1 changed files with 18 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/inference/basic/V2_PORTING_STATUS.md
+18 -2
View File
@@ -35,10 +35,26 @@ samplers/loops live in-package. Registration is one row in `v2/registry.py:_BUCK
Plus the **Wan2.1 i2v cluster** (Fun-1.3B-InP GPU-verified; I2V-14B-480P/720P + Wan2.2-I2V-A14B MoE reuse
the i2v card) — CLIP image-encoder + first-frame `[mask|cond]` → 36ch DiT.
## Newly ported — CPU-verified end-to-end; GPU load/run is BRINGUP
## GPU bring-up results (real weights on H100 NVL, single-GPU, TORCH_SDPA)
**20 models generate real video/audio on GPU** — the 7 above + **13 of the newly-ported** archs, each run
end-to-end through the real `VideoGenerator` (resolve → stamp → CUDA load → generate). The rest are blocked
by a **fastvideo-shared-code / missing-kernel / HF-access** wall, NOT a v2 recipe bug (the v2 recipes are
faithful — e.g. cosmos25's DiT+VAE produced finite output; only its Qwen2.5-VL encoder hit a library
incompat). All ports also resolve + run end-to-end on the CPU toy backend (`test_bucket_c_ports.py`).
| GPU status | Models |
|---|---|
| ✅ **Verified** (real GPU output) | stable_audio (audio), matrixgame2, matrixgame3, gen3c, wan_fun_control, lucy_edit, hunyuangamecraft, hunyuan_video, hunyuan_video15, longcat (13.58B), sfwan22 (2×14B MoE, expert offload), lingbotworld (2×14B, offload), fastwan (TI2V-5B-FullAttn DMD) |
| 🚫 fastvideo/env-blocked | **cosmos25** (DiT+VAE ran; Qwen2.5-VL encoder → transformers 5.12.1 incompat in fastvideo); **kandinsky5** (fastvideo registry registers a bare `PipelineConfig`); **hyworld** (fastvideo DiT hardcodes `flash_attn`, not built); **turbowan** 1.3B/i2v + **fastwan** VSA-variants (SLA/VSA sparse-attn params + Triton kernels need nvcc) |
| 🚫 access-blocked (HF-gated) | cosmos2, flux2, sd35 (no HF token in this env) |
To unblock the env-blocked: build `fastvideo-kernel` (SLA/VSA Triton, needs nvcc); pin a fastvideo-compatible
`transformers` for the Qwen2.5-VL encoder; add a Kandinsky5 `PipelineConfig` + an SDPA fallback in the
hyworld DiT (all fastvideo-side / environment, not v2 recipe work).
## Newly ported (recipe details)
Each resolves through the registry AND runs end-to-end on the CPU toy backend via the public `Engine`
path (the `v2/tests/test_bucket_c_ports.py` regression guard), emitting the correct modality artifact.
GPU bring-up pending per-model (cosmos2/flux2/sd35 weights are HF-gated; the FastVideo/* orgs are public).
**15 net-new architectures** (each a new `TorchComponent` adapter + recipe):
- **cosmos2** (Cosmos-Predict2-2B-Video2World) — EDM-Karras denoiser; new `CosmosDenoiseLoop` +