Phi-1.5 implementation

[lurosenb]

Phi 1.5 Summary

Phi-1.5 is a 1.3 billion parameter Transformer that outperforms some larger models. It was trained using sources similar to phi-1 (i.e. https://arxiv.org/pdf/2306.11644.pdf) plus synthetic NLP texts, and was not fine-tuned using RL. It's open-source, and ideally suited for QA, chat, and code prompts. Because it is not safeguarded, it occasionally generates irrelevant text.

Tasks

1. Add well supported example notebook of using phi-1.5 from the hugging face implementation i.e https://huggingface.co/microsoft/phi-1_5
2. Integrate into experimental pipeline for larger LLM effort