supriyar

icon company @facebook

Results 18 comments of supriyar

QAT model drops accuracy after converting with torch.ao.quantization.convert

cc @andrewor14

Question: How to use Float8InferenceLinear with FSDP1/2?

cc @drisspg @jainapurva

Torchao import time

thanks for reporting this @felipemello1! looks like some culprits are `torchao.float8.float8_utils`: ~437ms individual time and `torchao.quantization.autoquant`: ~220ms individual time and ~999ms cumulative Also noticed that the float8 related modules have...

Is it possible to export a QAT model in AWQ Format?

cc @andrewor14

NVFP4 MoE Training Status

cc @danielvegamyhre @vkuzo

Torchao opt resuming from ckpt requires `weights_only=False`

@gau-nernst any thoughts on what might be the issue?

Errors when convert quantized model to onnx

cc @andrewor14

Loading 8bit optimizer state from checkpoint causes dtype mismatch

cc @gau-nernst is this something you can take a look at?