[Feature Request] Support PyTorch GPT-2 Models

[nomyTx]

It would be very useful if there was a script to convert a PyTorch GPT-2 model to ggml.

[appvoid]

It would be my world if this is achieved since it seems to be simple with cerebras models. I'm not so good at the core of this library but the minimal approach llms it's one core philosophy of mine. Having said that I do have trained this model recently, but I'm not capable of make it work with ggml yet.

Having gpt2 quantized models "packaged" into a binary ready to use makes it pretty easy and actually usable for realtime scenarios on cpu. So please, someone who knows how to do this, help!

Edit:
Oh also, not forget to mention that gpt2 models perform better than cerebra's ones which I see has a proper convertion!