embedding模型 Instructor-embedding安装/推理报错解决方案

背景

在按照https://github.com/xlang-ai/instructor-embedding中的指引安装embedding模型instructor时，遇到了多种报错。比如：

INSTRUCTOR._load_sbert_model() got an unexpected keyword argument 'local_files_only'

 或

instructor embedding cannot import name 'cached_download' from 'huggingface_hub' 

以及

No such file or directory: 'hkunlp/instructor-large/modules.json'

原因是现在的transformers、huggingface库更新速度太快，基本上月更，所以需要修补各种版本上的问题。

解决方案

直接用sentence-transformer框架进行推理，读取权重文件。

pip install sentence-transformers -q

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("hkunlp/instructor-large", trust_remote_code=True)

sentence = "3D ActionSLAM: wearable person tracking in multi-floor environments"
instruction = "Represent the Science title:"
embeddings = model.encode([[instruction,sentence]])
print(embeddings)

注：默认embeddings未经过normalize