• 动手学习RAG: 向量模型
• 动手学习RAG: moka-ai/m3e 模型微调deepspeed与对比学习
• 动手学习RAG: 大模型向量模型微调 intfloat/e5-mistral-7b-instruct
• 动手学习RAG：rerank模型微调实践 bge-reranker-v2-m3
• 动手学习RAG：迟交互模型colbert微调实践 bge-m3
• 动手学习RAG：大模型重排模型 bge-reranker-v2-gemma微调

本文我们研究的是 bge-reranker-v2-m3模型

• https://huggingface.co/BAAI/bge-reranker-v2-m3

1. 环境准备

pip install transformers
pip install open-retrievals

• 注意安装时是pip install open-retrievals，但调用时只需要import retrievals
• 欢迎关注最新的更新https://github.com/LongxingTan/open-retrievals

2. 使用 bge-reranker-v2-m3 重排模型

from retrievals import AutoModelForRanking, RerankCollator, RerankTrainDataset, RerankTrainer, ColBERT, RetrievalTrainDataset, ColBertCollator

model_name_or_path: str = 'BAAI/bge-reranker-v2-m3'
rerank_model = AutoModelForRanking.from_pretrained(model_name_or_path, use_fp16=True)
scores_list = rerank_model.compute_score([['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']])
print(scores_list)

scores_list = rerank_model.compute_score([['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']], normalize=True)
print(scores_list)

3. 微调 bge-reranker-v2-m3 重排模型

4. 评测