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

[Bugfix] Fix tensor parallel issue in Qwen3 reranker weight loading (#20682)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
yurhett
2025-07-12 11:52:43 +08:00
committed by GitHub
parent b1235c3e10
commit 11c0198615
3 changed files with 38 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/language/pooling/mteb_utils.py
View File

@@ -268,7 +268,8 @@ def mteb_test_rerank_models(hf_runner,
model_info: RerankModelInfo,
vllm_extra_kwargs=None,
hf_model_callback=None,
vllm_mteb_encoder=VllmMtebEncoder):
vllm_mteb_encoder=VllmMtebEncoder,
atol=MTEB_RERANK_TOL):
if not model_info.enable_test:
# A model family has many models with the same architecture,
# and we don't need to test each one.
@@ -301,4 +302,4 @@ def mteb_test_rerank_models(hf_runner,
print("SentenceTransformers:", st_dtype, st_main_score)
print("Difference:", st_main_score - vllm_main_score)
assert st_main_score == pytest.approx(vllm_main_score, abs=MTEB_RERANK_TOL)
assert st_main_score == pytest.approx(vllm_main_score, abs=atol)