What version should I chose if I want to fine-tune and quantize it?
#1
by
JPSun56
- opened
As title,
should I chose QAT version or origianl one is better as further fine-tune without QAT.
Hi @JPSun56 ,
Welcome to Google Gemma family of open source models, if you would like to fine-tune the model further it's best recommend to use the original one. By using the original model you could able to get the more precise model weights for better performance. As gemma-3-1b doesn't require much memory or computation power you could run the model in normal Colab T4 GPU also.
Thanks.