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

12
tests/models/language/pooling/test_cross_encoder.py
View File

@@ -2,13 +2,15 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
from .mteb_utils import RerankModelInfo, mteb_test_rerank_models
from ...utils import (CLSPoolingRerankModelInfo, LASTPoolingRerankModelInfo,
RerankModelInfo)
from .mteb_utils import mteb_test_rerank_models
RERANK_MODELS = [
RerankModelInfo("cross-encoder/ms-marco-TinyBERT-L-2-v2",
architecture="BertForSequenceClassification"),
RerankModelInfo("tomaarsen/Qwen3-Reranker-0.6B-seq-cls",
architecture="Qwen3ForSequenceClassification")
CLSPoolingRerankModelInfo("cross-encoder/ms-marco-TinyBERT-L-2-v2",
architecture="BertForSequenceClassification"),
LASTPoolingRerankModelInfo("tomaarsen/Qwen3-Reranker-0.6B-seq-cls",
architecture="Qwen3ForSequenceClassification")
]