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

117
tests/models/language/pooling/test_baai.py
View File

@@ -2,73 +2,78 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
from ...utils import EmbedModelInfo, RerankModelInfo
from ...utils import (CLSPoolingEmbedModelInfo, CLSPoolingRerankModelInfo,
EmbedModelInfo, LASTPoolingEmbedModelInfo,
RerankModelInfo)
from .embed_utils import correctness_test_embed_models
from .mteb_utils import mteb_test_embed_models, mteb_test_rerank_models
MODELS = [
########## BertModel
EmbedModelInfo("BAAI/bge-base-en",
architecture="BertModel",
enable_test=True),
EmbedModelInfo("BAAI/bge-base-zh",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-small-en",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-small-zh",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-large-en",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-large-zh",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-large-zh-noinstruct",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-base-en-v1.5",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-base-zh-v1.5",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-small-en-v1.5",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-small-zh-v1.5",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-large-en-v1.5",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("BAAI/bge-large-zh-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-base-en",
architecture="BertModel",
enable_test=True),
CLSPoolingEmbedModelInfo("BAAI/bge-base-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-small-en",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-small-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-large-en",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-large-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-large-zh-noinstruct",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-base-en-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-base-zh-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-small-en-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-small-zh-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-large-en-v1.5",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("BAAI/bge-large-zh-v1.5",
architecture="BertModel",
enable_test=False),
########## XLMRobertaModel
EmbedModelInfo("BAAI/bge-m3",
architecture="XLMRobertaModel",
enable_test=True),
CLSPoolingEmbedModelInfo("BAAI/bge-m3",
architecture="XLMRobertaModel",
enable_test=True),
########## Qwen2Model
EmbedModelInfo("BAAI/bge-code-v1",
architecture="Qwen2Model",
dtype="float32",
enable_test=True),
LASTPoolingEmbedModelInfo("BAAI/bge-code-v1",
architecture="Qwen2Model",
dtype="float32",
enable_test=True),
]
RERANK_MODELS = [
########## XLMRobertaForSequenceClassification
RerankModelInfo("BAAI/bge-reranker-base",
architecture="XLMRobertaForSequenceClassification",
enable_test=True),
RerankModelInfo("BAAI/bge-reranker-large",
architecture="XLMRobertaForSequenceClassification",
enable_test=False),
RerankModelInfo("BAAI/bge-reranker-v2-m3",
architecture="XLMRobertaForSequenceClassification",
enable_test=False)
CLSPoolingRerankModelInfo(
"BAAI/bge-reranker-base",
architecture="XLMRobertaForSequenceClassification",
enable_test=True),
CLSPoolingRerankModelInfo(
"BAAI/bge-reranker-large",
architecture="XLMRobertaForSequenceClassification",
enable_test=False),
CLSPoolingRerankModelInfo(
"BAAI/bge-reranker-v2-m3",
architecture="XLMRobertaForSequenceClassification",
enable_test=False)
]