quantization of classifier with insilico perturbation

#550
by ZYSK-huggingface - opened

Hi ,
I noticed that in the InSilicoPerturber module, both pretrained and MTL models support automatic quantization within the code. I was wondering if similar quantization is also supported for the Classifier class?

Thanks for your question. The same method would work but you should confirm the layers that are loaded with respect to the layers of the fine-tuned classifier. For example, if loaded as Pretrained it may drop the last head layer so when you select the layer to use for the embeddings β€œ0” would be the current loaded last layer.

ctheodoris changed discussion status to closed

Sign up or log in to comment