namespace-Pt

Same issue here! `--bf16` yields the error while `--fp16` does not.

@duyc168 hi, crossentropy就是把概率相乘变成了log概率相加，本质上没有区别。没有试过直接用概率做蒸馏

Hi, NQ的一条训练样例： ```python { 'query_id': 10, 'query': 'why was there so much interest in cuba both before and after the civil war', 'answers': ['sugar markets'], 'pos': ["Spanish–American War American Civil...

1. msmarco本来就是一个qa数据集，answers是从[这里](https://huggingface.co/datasets/ms_marco/viewer/v2.1)对应过去的 2. 通过fine-tuned bge

@Hieunohair Did you solve it? I have the same problem.

hi，感谢指出，之前忘记删了，已更新。

Hi, please try specify `--dtype fp32` in the training script.

Hi, this is the direct result of contrastive learning. It only guarantees the positives have **higher** scores than negatives, while not assuring that their gaps are large enough. You can...

Hi，暂时不可以，我们公开的模型是之前给llama训得。如果要适配千问，需要为其额外训练

ColBERT is used for reranking, while LLM-Embedder is used for retrieval. They can be combined.