biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Support Step1 Model (#32511)

Browse Source 
Signed-off-by: xieli <xieli@stepfun.com>
This commit is contained in:
Li Xie
2026-01-18 18:20:46 +08:00
committed by GitHub
parent fe36bf5e80
commit c826c72a96
9 changed files with 472 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/test_initialization.py
View File

@@ -115,8 +115,11 @@ def can_initialize(
# FIXME: A hack to bypass FA3 assertion because our CI's L4 GPU
# has cc==8.9 which hasn't supported FA3 yet. Remove this hack when
# L4 supports FA3.
# Step1ForCausalLM requires TRITON_ATTN for use_alibi_sqrt support.
attention_config = (
{"backend": "TRITON_ATTN"} if model_arch == "GptOssForCausalLM" else None
{"backend": "TRITON_ATTN"}
if model_arch in ("GptOssForCausalLM", "Step1ForCausalLM")
else None
)
if model_arch == "WhisperForConditionalGeneration":
m.setenv("VLLM_WORKER_MULTIPROC_METHOD", "spawn")