Convert HF format model to gguf format and the model turns very big

[SeanZhang7]

I converted the merged HF model(Codellama-7B) into gguf format with examples/convert_legacy_llama.py and it went from 14 gb to 25 gb. Is this reasonable?

And when I try to use convert_hf_to_gguf.py, I got the error message: ValueError: Duplicated key name 'tokenizer.chat_template'