Philip May
Philip May
I also want to train phi-2 models. Is this bug still not fixed? @l3utterfly @winglian
Is it possible that a DPO dataset can (technicaly) not be used for evaluation and instead a SFT dataset should be used? And that is the reason why a `val_set_size:...
Well. I actualy found a solution for this question above. For DPO a `val_set_size` > 0 does not work at all. Do I set `val_set_size: 0`. If you want to...
@filippo82 can you please review this? Many thanks.
> Can we figure out another way of integrating this? I think it's valuable to attempt to load other file formats. I think of a step by step approach: 1....
> The example deepspeed configs are in the deepspeed_configs folder. @noobmaster29 But they do not offer an CPU offloading config.
IMO this is a very important PR. When a model is tested the correct chat template should be applied. Otherwise it is not a fair comparison.
This is a really interesting plugin. Unfortunately, since this question has not been answered for so long, I wonder if it is still actively maintained.
Hi, what about this issue? Is this software still maintained by someone? Just asking because I like to use software that is maintained and fixes bugs when reported. Thanks Philip
Nice one. Thanks for reporting @JonathanSchaber