chaofan
chaofan
reranker的分数区间范围很大，不同reranker分数所在的区间也不一致，它主要是根据分数的相对高低来判断相关性的 如果必须设置阈值的话，可以用reranker跑几个相关 / 不相关的例子分析一下
bge_reranker_large不支持2k的长度，可以用bge-reranker-v2-m3进行微调
主要看一下模型的表现是否正常，embedding模型微调后可以用embedding模型获取hard negatives，再对reranker进行微调
It is not compatible. You can use it by [LLARA-usage](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/LLARA#usage)
Sorry, our current training is not compatible with DeepSpeed ZeRO3. We recommend using a lower stage.
这个可以参考[flash attention](https://github.com/Dao-AILab/flash-attention) 的安装方案 CPU 机器上一般是没法进行模型微调的，只用推理的话是不需要安装[finetune]的
这个正常来讲使用统一的格式，对相对排名影响不是很大的 如果结果比较差，可以考虑使用[bge-reranker-v2-m3](https://huggingface.co/BAAI/bge-reranker-v2-m3)
ColBERT vector and sparse embedding are finetuned together. If you want to remove the `colbert vector`, you need to remove the code in the [finetune module](https://github.com/FlagOpen/FlagEmbedding/blob/f9f673e4ff159324d39c20a0c29686ca1e849963/FlagEmbedding/finetune/embedder/encoder_only/m3/modeling.py#L256).
Since sparse embeddings are not normalized, the sparse embedding similarity between identical embeddings cannot reach 1. It doesn't need normalization.
The `pos` and `neg` samples in the dataset should be stored in list format, so that the retrieved hard negatives will be complete sentences. Otherwise, if strings are passed in,...