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

Improve the output precision of embedding models (#19092)

Browse Source
This commit is contained in:
wang.yuqi
2025-06-04 19:48:57 +08:00
committed by GitHub
parent 8711bc5e68
commit 35cf32df30
8 changed files with 69 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/models/language/pooling/test_gte.py
View File

@@ -11,27 +11,21 @@ MODELS = [
########## BertModel
EmbedModelInfo("thenlper/gte-large",
architecture="BertModel",
dtype="float32",
enable_test=True),
EmbedModelInfo("thenlper/gte-base",
architecture="BertModel",
dtype="float32",
enable_test=False),
EmbedModelInfo("thenlper/gte-small",
architecture="BertModel",
dtype="float32",
enable_test=False),
EmbedModelInfo("thenlper/gte-large-zh",
architecture="BertModel",
dtype="float32",
enable_test=False),
EmbedModelInfo("thenlper/gte-base-zh",
architecture="BertModel",
dtype="float32",
enable_test=False),
EmbedModelInfo("thenlper/gte-small-zh",
architecture="BertModel",
dtype="float32",
enable_test=False),
########### NewModel
EmbedModelInfo("Alibaba-NLP/gte-multilingual-base",
@@ -46,7 +40,6 @@ MODELS = [
########### Qwen2ForCausalLM
EmbedModelInfo("Alibaba-NLP/gte-Qwen2-1.5B-instruct",
architecture="Qwen2ForCausalLM",
dtype="float32",
enable_test=True),
########## ModernBertModel
EmbedModelInfo("Alibaba-NLP/gte-modernbert-base",