microsoft

/

Phi-3-small-128k-instruct

like 173

Follow

Microsoft 6.45k

I am not able to resolve this error while trying to finetune this model.
I have loaded the model as bf16 using torch_dtype=torch.bfloat16 in the from_pretrained function. I have also added an explicit cast using model = model.to(torch.bfloat16).
The same exact code works flawlessly for the 'mini' version of the model, but not for this.
Any guidance would be greatly appreciated.
Thanks.