Switch from AutoGPTQ to GPTQModel

[ntoxeg]

Currently when trying to export a quantized model LLaMA Factory requires auto_gptq to be installed.
As per the project's page:

🚨 AutoGPTQ development has stopped. Please switch to GPTQModel as drop-in replacement. 🚨

The problem I have because of this is that I can't build AutoGPTQ with the latest Pytorch in a CUDA 12.8 environment, as it gives a compilation error. On the other hand, I am able to build GPTQModel in that environment. Also, it seems LLaMA Factory tries to import auto_gptq regardless of the quantization method chosen in the config.