• 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：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: 大模型向量模型微调 intfloat/e5-mistral-7b-instruct
• 动手学习RAG：大模型重排模型 bge-reranker-v2-gemma微调

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. 使用大模型做重排

from retrievals import LLMRankermodel_name = 'BAAI/bge-reranker-v2-gemma'
model = LLMRanker.from_pretrained(model_name,causal_lm=True,use_fp16=True,)score = model.compute_score(['query', 'passage'])
print(score)scores = 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)

3. 微调

MODEL_NAME='BAAI/bge-reranker-v2-gemma'
TRAIN_DATA="/root/kag101/src/open-retrievals/t2/t2_ranking.jsonl"
OUTPUT_DIR="/root/kag101/src/open-retrievals/t2/ft_out"torchrun --nproc_per_node 1 \-m retrievals.pipelines.rerank \--output_dir ${OUTPUT_DIR} \--overwrite_output_dir \--model_name_or_path $MODEL_NAME \--model_type llm \--causal_lm True \--use_lora True \--data_name_or_path $TRAIN_DATA \--task_prompt "Given a query A and a passage B, determine whether the passage contains an answer to the query by providing a prediction of either 'Yes' or 'No'." \--query_instruction "A: " \--document_instruction 'B: ' \--positive_key positive \--negative_key negative \--learning_rate 2e-4 \--num_train_epochs 3 \--per_device_train_batch_size 4 \--gradient_accumulation_steps 16 \--dataloader_drop_last True \--max_len 256 \--train_group_size 4 \--logging_steps 10 \--save_steps 20000 \--save_total_limit 1 \--bf16

4. 评测

在C-MTEB中进行评测。微调前保留10%的数据集作为测试集验证

微调前的指标：

微调后的指标：

{"dataset_revision": null,"mteb_dataset_name": "CustomReranking","mteb_version": "1.1.1","test": {"evaluation_time": 77.35,"map": 0.7057362287508586,"mrr": 0.8166538440773136}
}

微调后map从0.637上升至0.706，mrr从0.734上升至0.816