What is the difference between two llama-2 models?

[kjh2159]

Most of all, thank you for your team's awesome project!

I want to ask you one question.
I saw the mllmTeam's huggingface repository and there are a lot of model files.
By the way, for example, in the case of llama-2-7b-mllm, there are two versions of llama-2-7b-chat, named llama-2-7b-chat-q4_0_4_4.mllm and llama-2-7b-chat-q4_k.mllm.
(I saw the repository here: https://huggingface.co/mllmTeam/llama-2-7b-mllm/tree/main)
What are the differences between them and what do k and 0_4_4 stand for?

Thank you for your answer.

[chenghuaWang]

for Q4_0 and Q4_K pls check hugging face document: https://huggingface.co/docs/hub/en/gguf#quantization-types.
Q4_0_4_4 is a quantization type proposed by llama.cpp: ggml-org/llama.cpp#5780 (review). In simple terms, Q4_0_4_4 means tying together four Q4_0s, which is to say, storing together blocks of 4x32 elements.

[kjh2159]

@chenghuaWang
Thank you for your answer. I've probed the explanation of the given links. Your insight was impressive. I appreciate you again.