yeeeqichen comments

Repositories
Issues
Comments

Results 3 comments of


                                            yeeeqichen

QLoRA Training does not give expected results

@karths8 @amdnsr I think the problem is rooted in the `find_all_linear_names` function, which is incompatible with `QLoRA` setting, the correct way to implement this function should be like following ```...

Using bos_token, but it is not set yet.

For the bos_token setting, you could refer to `qlora.py line 671`

ValueError: You can't train a model that has been loaded in 8-bit precision on a different device than the one you're training on.

> It seems that the code is default trained on the first gpu, but that gpu is occupied by other program. So when I run it, I set auto and...