Lu Dai

About padding_side='left'

6

Hi, I noticed that you set padding_side='left' in finetune.py. However, in llama the default padding side seems to be 'right'. Would this inconsistency causes certain problems such as performance drop?

DDP NCCL Error when finetuning on longer sequences

4

Hi, I tried to finetune alpaca-lora on longer instructions dataset, but **only** change the dataset would cause ddp error listed below: It happens in the middle of training, and when...

Hello, thanks for the great work! I'm curious about the intuition and implementation of the geodesic loss in Vpose training. Would you please explain the mathematics behind it and its...

能修改前端的python版本

1

尊敬的开发者们， 我现在需要在python版本上面给前端加一些功能。但是 qanything-python 分支的前端是预编译好的代码 但1.3.0 和之前的包含front_end版本似乎只有docker？请问有没有包含front_end的python版本呢

hi authors, Since the code is only for longbench and infinitebench, whether the method can generalize to other long context problems, instead of retrieving keys?