Ask-sola

如果希望将句子和词语之间做匹配应该如何操作合适呢？

1

我希望将一个句子直接进行分类，但是我手上没有一个标注好的分类数据集，所以只能求助于开放的模型，通过暴力匹配计算相似度来实现句子分类。那么如果直接将句子和词语编码后计算相似度，是否会有相似性的性能体现出来呢？

我使用的是python=3.8，发生了LSTM部分函数不支持，似乎需要pytorch1.3？（这好像已经是很难获取的版本了）

只训练了两个epoch的map就到了87.19%？？？

1

我用了百度云下载的主干特征提取网络，然后只训练了两个epoch，然后直接做getmap，得到了87.19%，这个情况正常吗  

Hello, I am trying to run this project, but when I was building RamCloud, I entered "make j8 DEBUG=no" and received the following error:  This seems to be related...

I can't see the printed logs during training

1

I used wandb to look at my training process and found that the printed process was overwritten, and although it did have training, there was a problem with the display....

Problem with fine-tuning bge-reranker-v2-gemma

2

I followed the code below to fine-tune the model: ```bash torchrun --nproc_per_node 2 \ -m FlagEmbedding.finetune.reranker.decoder_only.base \ --model_name_or_path BAAI/bge-reranker-v2-gemma \ --use_lora True \ --lora_rank 32 \ --lora_alpha 64 \ --use_flash_attn...

Hello, I read your paper and I noticed that the final loss iteration used a labeling method to calculate binary cross-entropy loss. If I understand correctly, this seems to be...

[BUG] apex has deleted amp

1

**Describe the bug** The code segment in DeepSpeed: from apex import ampat line 135 in engine.py under the runtime directory is no longer usable. https://github.com/NVIDIA/apex/issues/1912