Text2Vec加载BAAI_bge-large-zh-v1.5进行相似度计算得到的结果有误

[sslalxh]

提问时请尽可能提供如下信息：

基本信息

• 你使用的操作系统: macOS 14.2.1
• 你使用的Python版本: 3.8.16
• 你使用的Pytorch版本: torch 2.1.0
• 你使用的bert4torch版本: 0.4.9.post2
• 你加载的预训练模型:bge-large-zh-v1.5

核心代码

sentences_1 = ["样例数据-1", "样例数据-2"]
    sentences_2 = ["样例数据-3", "样例数据-4"]
    print(root_model_path)
    
    print('=========================================sentence transformer====================================')
    from sentence_transformers import SentenceTransformer

    model = SentenceTransformer(root_model_path)
    embeddings_1 = model.encode(sentences_1, normalize_embeddings=True)
    embeddings_2 = model.encode(sentences_2, normalize_embeddings=True)
    similarity = embeddings_1 @ embeddings_2.T
    print(similarity)

    print('=========================================bert4torch====================================')
    from bert4torch.pipelines import Text2Vec

    text2vec = Text2Vec(checkpoint_path=root_model_path, device='mps')
    embeddings_1 = text2vec.encode(sentences_1, normalize_embeddings=True)
    embeddings_2 = text2vec.encode(sentences_2, normalize_embeddings=True)
    similarity = embeddings_1 @ embeddings_2.T
    print(similarity)

命令行打印的信息

model/BAAI_bge-large-zh-v1.5
=========================================sentence transformer====================================
[[0.85533315 0.8520633 ]
 [0.87456286 0.8557935 ]]
=========================================bert4torch====================================
Loading checkpoint shards: 100%|██████████| 1/1 [00:01<00:00,  1.16s/it]
[[0.9882161  0.98629344]
 [0.9826301  0.9869268 ]]

想问问为什么用Text2Vec加载模型encode并计算相似度会与SentenceTransformer的结果不一致，尝试了其他sentence，通过Text2Vec的计算结果都是大于0.9，求教

[Tongjilibo]

我刚试了下，指定文件路径，两边结果是一致的啊，bert4torch_config.json文件地址是https://huggingface.co/Tongjilibo/bert4torch_config/tree/main/bge-large-zh-v1.5 但是也有问题，就是指定model_name的时候有点bug，我需要修复一下

[Tongjilibo]

pip install git+https://github.com/Tongjilibo/torch4keras.git
pip install git+https://github.com/Tongjilibo/bert4torch.git@dev

我修改了一下逻辑，可以用以上命令pip install最新版，下面为调用代码

import os
os.environ['HF_ENDPOINT'] = "https://hf-mirror.com"
# root_model_path = 'E:\pretrain_ckpt\embedding\BAAI@bge-large-en-v1.5'
root_model_path = 'BAAI/bge-large-zh-v1.5'
# root_model_path = '/data/pretrain_ckpt/embedding/BAAI--bge-large-zh-v1.5'

sentences_1 = ["样例数据-1", "样例数据-2"]
sentences_2 = ["样例数据-3", "样例数据-4"]

print('=========================================sentence transformer====================================')
from sentence_transformers import SentenceTransformer
model = SentenceTransformer(root_model_path)
embeddings_1 = model.encode(sentences_1, normalize_embeddings=True)
embeddings_2 = model.encode(sentences_2, normalize_embeddings=True)
similarity = embeddings_1 @ embeddings_2.T
print(similarity)


print('=========================================bert4torch====================================')
from bert4torch.pipelines import Text2Vec
text2vec = Text2Vec(checkpoint_path=root_model_path, device='cuda')
embeddings_1 = text2vec.encode(sentences_1, normalize_embeddings=True)
embeddings_2 = text2vec.encode(sentences_2, normalize_embeddings=True)
similarity = embeddings_1 @ embeddings_2.T
print(similarity)

# 输出结果
=========================================sentence transformer====================================
[[0.85533345 0.8520633 ]
 [0.8745628  0.8557939 ]]
=========================================bert4torch====================================
Loading checkpoint shards: 100%|█████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.57it/s]
[[0.85533345 0.8520633 ]
 [0.8745628  0.8557937 ]]