FROM ghcr.io/ggml-org/llama.cpp:server

ENV HOST=0.0.0.0
ENV PORT=7860
EXPOSE 7860

# llama serve -hf constructai/VibeThinker-3B-GGUF:UD-Q4_K_XL
CMD ["-hf", "constructai/VibeThinker-3B-GGUF:UD-Q4_K_XL", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "--alias", "vibethinker-3b-q4", \
    "--temp", "1.0", \
    "--top-p", "0.95", \
    "--top-k", "1", \
    "--cache-type-k", "q8_0", \
     "-c", "16384"]

# https://huggingface.co/WeiboAI/VibeThinker-3B temperature=1.0 top_p=0.95 top_k=-1
# 65536 scale the value downward to 32768 (32K) or 16384 (16K) if OOM