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

87
tests/models/language/pooling/test_gte.py
View File

@@ -4,57 +4,58 @@ from typing import Any
import pytest
from ...utils import check_transformers_version
from .embed_utils import EmbedModelInfo, correctness_test_embed_models
from ...utils import (CLSPoolingEmbedModelInfo, EmbedModelInfo,
LASTPoolingEmbedModelInfo, check_transformers_version)
from .embed_utils import correctness_test_embed_models
from .mteb_utils import mteb_test_embed_models
MODELS = [
########## BertModel
EmbedModelInfo("thenlper/gte-large",
architecture="BertModel",
enable_test=True),
EmbedModelInfo("thenlper/gte-base",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("thenlper/gte-small",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("thenlper/gte-large-zh",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("thenlper/gte-base-zh",
architecture="BertModel",
enable_test=False),
EmbedModelInfo("thenlper/gte-small-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("thenlper/gte-large",
architecture="BertModel",
enable_test=True),
CLSPoolingEmbedModelInfo("thenlper/gte-base",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("thenlper/gte-small",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("thenlper/gte-large-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("thenlper/gte-base-zh",
architecture="BertModel",
enable_test=False),
CLSPoolingEmbedModelInfo("thenlper/gte-small-zh",
architecture="BertModel",
enable_test=False),
########### NewModel
EmbedModelInfo("Alibaba-NLP/gte-multilingual-base",
architecture="GteNewModel",
enable_test=True),
EmbedModelInfo("Alibaba-NLP/gte-base-en-v1.5",
architecture="GteNewModel",
enable_test=True),
EmbedModelInfo("Alibaba-NLP/gte-large-en-v1.5",
architecture="GteNewModel",
enable_test=True),
CLSPoolingEmbedModelInfo("Alibaba-NLP/gte-multilingual-base",
architecture="GteNewModel",
enable_test=True),
CLSPoolingEmbedModelInfo("Alibaba-NLP/gte-base-en-v1.5",
architecture="GteNewModel",
enable_test=True),
CLSPoolingEmbedModelInfo("Alibaba-NLP/gte-large-en-v1.5",
architecture="GteNewModel",
enable_test=True),
########### Qwen2ForCausalLM
EmbedModelInfo("Alibaba-NLP/gte-Qwen2-1.5B-instruct",
architecture="Qwen2ForCausalLM",
enable_test=True),
LASTPoolingEmbedModelInfo("Alibaba-NLP/gte-Qwen2-1.5B-instruct",
architecture="Qwen2ForCausalLM",
enable_test=True),
########## ModernBertModel
EmbedModelInfo("Alibaba-NLP/gte-modernbert-base",
architecture="ModernBertModel",
enable_test=True),
CLSPoolingEmbedModelInfo("Alibaba-NLP/gte-modernbert-base",
architecture="ModernBertModel",
enable_test=True),
########## Qwen3ForCausalLM
EmbedModelInfo("Qwen/Qwen3-Embedding-0.6B",
architecture="Qwen3ForCausalLM",
dtype="float32",
enable_test=True),
EmbedModelInfo("Qwen/Qwen3-Embedding-4B",
architecture="Qwen3ForCausalLM",
dtype="float32",
enable_test=False),
LASTPoolingEmbedModelInfo("Qwen/Qwen3-Embedding-0.6B",
architecture="Qwen3ForCausalLM",
dtype="float32",
enable_test=True),
LASTPoolingEmbedModelInfo("Qwen/Qwen3-Embedding-4B",
architecture="Qwen3ForCausalLM",
dtype="float32",
enable_test=False),
]