[Bugfix][MM][CG] Enable dual-path ViT CUDA graph for Step3-VL (#46034)

Signed-off-by: shen-shanshan <467638484@qq.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Shanshan Shen
2026-06-28 00:06:43 -07:00
committed by GitHub
co-authored by Isotr0py
parent c7ca0bccae
commit 35e6c86caa
2 changed files with 87 additions and 125 deletions
+1 -1
View File
@@ -136,7 +136,7 @@ Models opt-in to encoder CUDA Graphs by implementing the [SupportsEncoderCudaGra
| `Qwen3VLForConditionalGeneration` | `Qwen3-VL` | ✅︎ | ✅︎ | ❌︎ |
| `Qwen3_5ForConditionalGeneration` | `Qwen3.5`, `Qwen3.6` | ✅︎ | ✅︎ | ❌︎ |
| `Qwen3_5MoeForConditionalGeneration` | `Qwen3.5-MoE`, `Qwen3.6-MoE` | ✅︎ | ✅︎ | ❌︎ |
| `Step3VLForConditionalGeneration` | `Step3-VL` | ✅︎ | ❌︎ | ❌︎ |
| `Step3VLForConditionalGeneration` | `Step3-VL` | ✅︎ | ❌︎ | ✅︎ |
!!! note
Encoder CUDA Graphs have currently been tested with `--mm-encoder-attn-backend=FLASH_ATTN` and `--mm-encoder-attn-backend=FLASHINFER` on Blackwell GPUs.