Alp Dener
Results
42
comments of
Alp Dener
@anderson101866 This should be fixed now in TE/main as of PR #1087. Could you check and close the issue if resolved?
Hi @vgoklani -- TE modules can be initialized under the `with te.fp8_model_init():` context to allocate their primary weights in FP8 (as `te.Float8Tensor`s) instead of allocating at a higher precision and...