Puffin 7B, everythingLM, ShareGPT small trained for LoRA. Everything is using sharegpt format