New Features
* auto device selection, cpu, gpu, multi-gpu
* added 8bit + 4bit support for lora/qlora
* added bitsandbytes support
* add support for hardware-agnostic quantized models using GGML
* added RAG generation using Pinecone
* added chatGPT finetuning
Fixes
* chatgpt finetune model changed to gpt-3.5-turbo-0613
Others
* refactored completions
* corrected class names to use CamelCase
* removed duplicated example files and added missing requirements