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

[New Model]: Support GteNewModelForSequenceClassification (#23524)

Browse Source 
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-08-28 15:36:42 +08:00
committed by GitHub
parent 186aced5ff
commit 11a7fafaa8
13 changed files with 157 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
tests/models/language/pooling/test_qwen3_reranker.py
View File

@@ -11,12 +11,20 @@ from tests.utils import multi_gpu_test
from ...utils import LASTPoolingRerankModelInfo, RerankModelInfo
from .mteb_utils import mteb_test_rerank_models
qwen3_reranker_hf_overrides = {
"architectures": ["Qwen3ForSequenceClassification"],
"classifier_from_token": ["no", "yes"],
"is_original_qwen3_reranker": True,
}
RERANK_MODELS = [
LASTPoolingRerankModelInfo("Qwen/Qwen3-Reranker-0.6B",
architecture="Qwen3ForSequenceClassification",
hf_overrides=qwen3_reranker_hf_overrides,
enable_test=True),
LASTPoolingRerankModelInfo("Qwen/Qwen3-Reranker-4B",
architecture="Qwen3ForSequenceClassification",
hf_overrides=qwen3_reranker_hf_overrides,
enable_test=False)
]
@@ -74,18 +82,7 @@ class Qwen3RerankerHfRunner(HfRunner):
@pytest.mark.parametrize("model_info", RERANK_MODELS)
def test_rerank_models_mteb(vllm_runner, model_info: RerankModelInfo) -> None:
assert model_info.architecture == "Qwen3ForSequenceClassification"
vllm_extra_kwargs: dict[str, Any] = {
"hf_overrides": {
"architectures": ["Qwen3ForSequenceClassification"],
"classifier_from_token": ["no", "yes"],
"is_original_qwen3_reranker": True,
}
}
mteb_test_rerank_models(Qwen3RerankerHfRunner, vllm_runner, model_info,
vllm_extra_kwargs)
mteb_test_rerank_models(Qwen3RerankerHfRunner, vllm_runner, model_info)
@pytest.mark.parametrize("model_info", RERANK_MODELS)
@@ -96,11 +93,6 @@ def test_rerank_models_mteb_tp(vllm_runner,
assert model_info.architecture == "Qwen3ForSequenceClassification"
vllm_extra_kwargs: dict[str, Any] = {
"hf_overrides": {
"architectures": ["Qwen3ForSequenceClassification"],
"classifier_from_token": ["no", "yes"],
"is_original_qwen3_reranker": True,
},
"tensor_parallel_size": 2,
}