Update README.md

根据4.7日更新代码，解决冲突

新增ernie系列模型作为embedding model,默认依旧使用GanymedeNil/text2vec-large-chinese
显存不足时可尝试使用ernie-tiny作为embedding模型
调用方式为embeddings = HuggingFaceEmbeddings(model_name=embedding_model_dict["ernie-tiny"], )

fix: 修复chatglm模型被复制，显存占用过多

Update requirements.txt

model作为类成员会在类实例化时进行一次复制
这导致每询问一个问题显存占用就会翻倍
通过将model改成全局变量修复这个问题

修复requirements.txt文件

修复了运行pip install -r requirements.txt提示ERROR: Invalid requirement: '"detectron2@git+https://github.com/facebookresearch/detectron2.git@v0.6#egg=detectron2"' (from line 4 of requirements.txt)
Hint: = is not a valid operator. Did you mean == ?的错误

修复运行pip install -r requirements.txt提示ERROR: Invalid requirement: '"detectron2@git+https://github.com/facebookresearch/detectron2.git@v0.6#egg=detectron2"' (from line 4 of requirements.txt)
Hint: = is not a valid operator. Did you mean == ?的错误