fix HPU could not handle float16 in attention.py.

#68
by sywangyi - opened
No description provided.

@yao-matrix

sywangyi changed pull request status to open

mpt model need to be changed in Attention for the following error. https://huggingface.co./mosaicml/mpt-7b/blob/main/attention.py#L49
causal_mask = attn_weight.new_ones(s, s, dtype=torch.float16)
RuntimeError: float16/half is not supported on Gaudi.

@daking please help review, thanks

Cannot merge
This branch has merge conflicts in the following files:
  • attention.py

Sign up or log in to comment