Pawan Jayakumar

icon indicating a website link https://vayuda.github.io icon indicating an email [email protected]

icon company UC San Diego icon location Masters student at UCSD studying ML systems and security

Results 3 comments of Pawan Jayakumar

[WIP] Activation Aware Weight Quantization (AWQ)

> @vayuda please let me know when you address the comments, I think this just needs a final update and we can merge Should finish this tmrw

Add weight tensor-wise scaling for INT8 quantized and mixed-precision training

So basically for [quantize_int8_rowwise](https://github.com/vayuda/ao/blob/main/torchao/prototype/quantized_training/int8.py#L17) we would pass in a quantization granularity that could either be set to row-wise or tensor-wise. In the case of tensor-wise, even though the scale is...

[Proposal] Support R1 Distills

Hello! Any update on this? I feel that even just having one of them figured out would be a great benefit before working out the kinks for the larger models....