服务器无法连接huggingface下载模型
报错代码
import torch
from FlagEmbedding.visual.modeling import Visualized_BGE
def image_retrival():
############ Use Visualized BGE doing composed image retrieval
model = Visualized_BGE(model_name_bge = "BAAI/bge-m3", model_weight="/data/suanfa/meng/mybge-m3/Visualized_m3.pth")
model.eval()
with torch.no_grad():
query_emb = model.encode(image="../image/whitedogwithyellowleaf.jpg/", text="Make the background dark, as if the camera has taken the photo at night")
candi_emb_1 = model.encode(image="../image/4dogs_on_lawn.jpg")
candi_emb_2 = model.encode(image="../image/whitedog.jpg")
sim_1 = query_emb @ candi_emb_1.T
sim_2 = query_emb @ candi_emb_2.T
print(sim_1, sim_2)
if __name__=='__main__':
image_retrival()
报错
OSError: We couldn't connect to 'https://huggingface.co' to load this file, couldn't find it in the cached files and it looks like BAAI/bge-m3 is not the path to a directory containing a file named config.json.
Checkout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-model'.
解决方法1：offline-mode形式
1.打开hugging face 官网：https://huggingface.co/docs/transformers/installation#offline-mode 并点击Models
2.搜索我们需要的模型
输入使用的模型，并点击模型
3.下载文件
点击file and versions
通过后面的下载按钮下载圈出的五个文件到本地
4.上传文件到服务器
(py38_bge) [root@x.x.x.x /data/suanfa/meng/mybge-m3/bge-m3]$ ll
total 21664
-rw-r--r-- 1 root root 687 Oct 9 20:01 config.json
-rw-r--r-- 1 root root 5069051 Oct 10 19:50 sentencepiece.bpe.model
-rw-r--r-- 1 root root 125 Oct 10 19:53 special_tokens_map.json
-rw-r--r-- 1 root root 444 Oct 9 20:01 tokenizer_config.json
-rw-r--r-- 1 root root 17098108 Oct 9 20:01 tokenizer.json
5.修改py文件
model = Visualized_BGE(model_name_bge = "/data/suanfa/meng/mybge-m3/bge-m3", model_weight="/data/suanfa/meng/mybge-m3/Visualized_m3.pth"
6.运行报错
model.encode行报错
IndexError: index out of range in self
7.问题排查
和解决方法2下载下来的文件进行了对比，发现两个special_tokens_map.json文件内容不一致，于是替换成解决方法2下载下来的该文件替换方法1的该文件，问题成功解决，模型输出：
tensor([[0.5751]]) tensor([[0.5856]])
解决方法2：通过镜像下载模型
1.访问网站：https://hf-mirror.com/
里面有详细步骤：
pip install -U huggingface_hub
export HF_ENDPOINT=https://hf-mirror.com
