search

pytorch / ao

PyTorch native quantization and sparsity for training and inference
BSD 3-Clause "New" or "Revised" License
1.41k stars 140 forks source link

Assertion information on precompute_float8_dynamic_scale_for_fsdp is misleading #1108

Open goldhuang opened 2 hours ago

goldhuang commented 2 hours ago

https://github.com/pytorch/ao/blob/main/torchao/float8/fsdp_utils.py#L44-L48

Should it be raise NotImplementedError("Only supports dynamic scaling")?

msaroufim commented 1 hour ago

Yeah I think you're right, wanna just make a PR to fix this?