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

[Model] Pooling models default to using chunked prefill & prefix caching if supported. (#20930)

Browse Source 
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-08-12 00:41:37 +08:00
committed by GitHub
parent 16fb668b61
commit 84cf78acee
31 changed files with 452 additions and 261 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
tests/models/language/pooling/test_qwen3_reranker.py
View File

@@ -8,15 +8,16 @@ import torch
from tests.conftest import HfRunner
from tests.utils import multi_gpu_test
from .mteb_utils import RerankModelInfo, mteb_test_rerank_models
from ...utils import LASTPoolingRerankModelInfo, RerankModelInfo
from .mteb_utils import mteb_test_rerank_models
RERANK_MODELS = [
RerankModelInfo("Qwen/Qwen3-Reranker-0.6B",
architecture="Qwen3ForSequenceClassification",
enable_test=True),
RerankModelInfo("Qwen/Qwen3-Reranker-4B",
architecture="Qwen3ForSequenceClassification",
enable_test=False)
LASTPoolingRerankModelInfo("Qwen/Qwen3-Reranker-0.6B",
architecture="Qwen3ForSequenceClassification",
enable_test=True),
LASTPoolingRerankModelInfo("Qwen/Qwen3-Reranker-4B",
architecture="Qwen3ForSequenceClassification",
enable_test=False)
]