FlagEmbedding
FlagEmbedding copied to clipboard
FlagOpen
Retrieval and Retrieval-augmented LLMs
What are the hardware configuration requirements for running the bge-m3 model? example: GPU model &VRAM& Memory
预训练问题
File "/data/FlagEmbedding/FlagEmbedding/bge_m3.py", line 6, in import datasets ModuleNotFoundError: No module named 'datasets'
https://github.com/FlagOpen/FlagEmbedding/blob/bd38bd350054d0dba39ea8d602afac1fab141b35/FlagEmbedding/reranker/data.py#L42 代码中padding=False item = self.tokenizer.encode_plus( qry_encoding, doc_encoding, truncation=True, max_length=self.args.max_len, padding=False, ) 但是参数这里又说的是会pad。所以实际训练的时候，是padding了吗？ max_len: int = field( default=512, metadata={ "help": "The maximum total input sequence length after tokenization for input text....
If it was not released publicly yet, when it will be avaiable? Thanks a lot.
Some weights of XLMRobertaForSequenceClassification were not initialized from the model checkpoint at /bge/FlagEmbedding/examples/reranker/sft_model/0221-4/merged and are newly initialized: ['classifier.out_proj.bias', 'classifier.dense.bias', 'classifier.dense.weight', 'classifier.out_proj.weight'] You should probably TRAIN this model on a down-stream...
您好我想咨询下，BGE-m3支持大文档的内容检索吗？ 我想做个智能客服，知识库是以文档的形式存储，拿用户的query在文档中检索出某一句作为答案回复，支持这样吗？ 我理解文档中提到的检索，还是两个句子的相似度检索，并不能直接从一个文档中提取出某句话来回答用户的问题；
关于解决 #464 的修改。 避免困难样本挖掘时，当召回负样本数量少于预设负采样数量，会随机采样到正样本、或重复采样负样本的问题。 修改为，默认从 `corpus` 中剔除正例和已召回的负例，再进行随机采样；若剔除后 `corpus` 为空，说明需要重复采样负样本才能满足负采样数量要求，则只剔除正样本、重复采样负样本即可。
您好 我在微调bge 1.5的过程中，想在训练的过程中同时加入评估集，然后根据评估集合的loss去保存最优模型，于是我新增了一下一个参数 eval_dataset=eval_dataset， evaluation_strategy "epoch"， save_strategy "epoch"， save_total_limit 3 load_best_model_at_end True 但是结果遇到了报错，在评估的过程中找不到对应的loss Traceback (most recent call last): File "C:\Users\Lichengyang\Desktop\FlagEmbedding-master\FlagEmbedding-master\FlagEmbedding\baai_general_embedding\finetune\run.py", line 116, in main() File "C:\Users\Lichengyang\Desktop\FlagEmbedding-master\FlagEmbedding-master\FlagEmbedding\baai_general_embedding\finetune\run.py", line 102, in main...
