tankeui

Results 3 comments of tankeui

Full parameter fine-tuning bugs

Additionally, it works when I use LLaMA3-8B. (16-bit train: False)

multi-gpu full para train error

> It seems like a cuda device mismatch issue. > > > [rank1]: RuntimeError: CUDA error: invalid device ordinal > > I guess you've set CUDA_VISIBLE_DEVICES somewhere else accidently and...

DPO+ZeRO train error

Thanks! It works!