search

naseemap47 / YOLO-NAS

Train and Inference your custom YOLO-NAS model by Single Command Line
Apache License 2.0
98 stars 13 forks source link

Quantization aware training support #38

Closed jeck00119 closed 10 months ago

jeck00119 commented 1 year ago

Hi! It will be helpful if you could add CLI support for quantization aware training and maybe an option to save the model .pt because i want to use ultralytics for inference. Thank you!

naseemap47 commented 1 year ago

Hi @jeck00119 , I will be working on quantization aware training ASAP. I don't have time in this week, fully packed. But I will look into it. model .pt have meta data (class names, model type, size, etc..) but .pth don't have. But super-gradients didn't give feature for that. I also asked them for this. Currently not available. If you find any please let me know. But to over come this problem (class names), I manually added dummy loading for this.

If you have any questions and any improvements needed please let me know. Thank you.

hdiamond-foundationstack commented 1 year ago

@naseemap47 Any update on quantization aware training?

naseemap47 commented 1 year ago

@hdiamond-foundationstack If didn't get the time to complete. I will get it done ASAP.

naseemap47 commented 1 year ago

Hi @jeck00119 @hdiamond-foundationstack , I completed the development #49 I am facing issue while running QAT. NameError: name 'SelectiveQuantizer' is not defined https://github.com/Deci-AI/super-gradients/issues/1205#issuecomment-1606573396 They will fix this issue in the upcoming release. (https://github.com/Deci-AI/super-gradients/issues/1205#issuecomment-1612862427) If you find any solution, please let me know. Thank you.