DeepSpeed
DeepSpeed copied to clipboard
load linear layer weight with given dtype
bf16 inference fails due to data type mismatch as half is default value
Please review this
Hi @jeffra :) can you please help reviewing this change?
Hi @RezaYazdaniAminabadi @mrwyattii @jeffra @awan-10 @cmikeh2 @arashb please review this change.
Hi @RezaYazdaniAminabadi @mrwyattii @jeffra @awan-10 @cmikeh2 @arashb please review this change.
The changes look good to me 👍 I think we can merge it after fixing the comment from @tjruwase