Tim Moon
Tim Moon
I think initializing FP8 weights with a constructor kwarg makes a lot of sense. In effect, the `fp8_model_init` context is an indirect way of passing a boolean arg to the...
Are you building the 1.9 release or the main branch? This looks like an error that was fixed with https://github.com/NVIDIA/TransformerEngine/pull/949. If that doesn't fix it, perhaps it's something with the...
I've gone ahead and cherry-picked #949 into the 1.8 release.
/te-ci pytorch
/te-ci pytorch
/te-ci pytorch
/te-ci pytorch
/te-ci pytorch
/te-ci pytorch
/te-ci pytorch