[MM][CG] Enable encoder Cudagraph for Step3VL (#42224)

Signed-off-by: JisoLya <523420504@qq.com>
Signed-off-by: Soyaazz <523420504@qq.com>
This commit is contained in:
Soyaazz
2026-05-17 20:19:13 -07:00
committed by GitHub
parent 107210442d
commit 990f49bdcb
8 changed files with 384 additions and 22 deletions
@@ -41,6 +41,13 @@ def qwen_vl_chat_template(content: str) -> str:
return f"<|im_start|>user\n{content}<|im_end|>\n<|im_start|>assistant\n"
def step3_vl_chat_template(content: str) -> str:
return (
"<|begin▁of▁sentence|> You are a helpful assistant.<|BOT|>user\n "
f"<im_patch>{content} <|EOT|><|BOT|>assistant\n"
)
MODEL_CONFIGS: dict[str, VitCudagraphTestConfig] = {
"qwen2_5_vl": VitCudagraphTestConfig(
model="Qwen/Qwen2.5-VL-3B-Instruct",
@@ -90,6 +97,11 @@ MODEL_CONFIGS: dict[str, VitCudagraphTestConfig] = {
needs_video_metadata=False,
marks=[pytest.mark.core_model],
),
"step3_vl": VitCudagraphTestConfig(
model="stepfun-ai/Step3-VL-10B",
image_prompt=step3_vl_chat_template("What is in this image?"),
video_prompt=None,
),
}