undefined with deepseek-lite-ggml and llama.cpp

[micsthepick]

Describe the bug any call to the chat completion api in llama.cpp returns a string of "undefined" with the DeepSeek-Coder-V2-Lite-Instruct-Q6_K.gguf model

To Reproduce Download the model, run with the following command: ./llama-server -ngl 99 --ctx-size 16384 -m models/DeepSeek-Coder-V2-Lite-Instruct/DeepSeek-Coder-V2-Lite-Instruct-Q6_K.gguf , add a chat provider for llamacpp at localhost:8080/v1/chat/completions.

Expected behavior undefined is NOT repeated and the generated text makes at least some level of sense

Screenshots

Logging logs.txt

API Provider as above, llama.cpp default port on localhost:8080/v1/chat/completions

Chat or Auto Complete? chat

Model Name DeepSeek-Coder-V2-Lite-Instruct-Q6_K.gguf

Desktop (please complete the following information):

• OS: Windows 11 Edu
• Browser N/A

Additional context

[Raizo62]

Hi

And if you set as Provider the value "ollama" (instead of "llamacpp", even if you use llama.cpp) ?