WizardCoder Inference accuracy dropped a lot compared to fastchat or vllm

[sanigochien]

Is it possible that it is bug about llama.cpp?
Please forgive me if it is not a proper issue reporting.

[KerfuffleV2]

#3986 - Is this relevant for you? (I'm assuming WizardCoder is based on Deepseek Coder models).

[GANJAC]

Hi @sanigochien @KerfuffleV2 the same with this model:

https://huggingface.co/TheBloke/deepseek-coder-6.7B-instruct-GGUF

With any quantization, is the same #3986 ?

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.