Add phi3 support

[pprp]

@casper-hansen Thank you for your invitation.

This PR introduces the support for phi3 for autoawq.

Due to the fact that the phi3 hasn't been released to transformer package, I conducted experiments on the development branch, a.k.a 4.41.0.dev0

I conducted experiments on RTX4090 and evaluated the perpelexity of the quantized phi3 (microsoft/Phi-3-mini-128k-instruct).

[umangkeshri]

Hey @pprp, I was trying to use these code changes for Phi-3 quantization for my use case, and getting below warning is it expected?
Token indices sequence length is longer than the specified maximum sequence length for this model (8322 > 4096). Running this sequence through the model will result in indexing errors

[vackosar]

@pprp This pull request is waiting for a review?

[pprp]

@vackosar Yes, or you can review until the phi-3 is formally supported by transformers library.

[stan-hua]

Hey @pprp, great job with the implementation! Would it possible to also add support for Phi-3-small at this time?

Phi-3-Small isn't implemented in transformers yet, but it is available in the HF repo. For example, Phi-3-small-8k-instruct

https://huggingface.co/microsoft/Phi-3-small-128k-instruct/blob/main/modeling_phi3_small.py

[casper-hansen]

Sorry for taking so long. I was away for a month, but I'm back now.