commit fine-tuned tokenizer model on conversations based on decapoda-research/llama-7b-hf model

delete base model tokenizer

commit my first llama-7b-hf model