add fine-tuned model tokenizer trained on multilingual conversations based on decapoda-research/llama-13b-hf model

delete base model tokenizer

committing original decapoda-research/llama-13b-hf model