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

USER root
RUN apt-get update && apt-get install -y curl

# Download Gemma-3n-E2B GGUF
RUN mkdir -p /models && \
    curl -L https://huggingface.co/unsloth/gemma-3n-E2B-it-GGUF/resolve/main/gemma-3n-E2B-it-Q4_0.gguf -o /models/model.gguf && \
    chown -R 1000:1000 /models

USER 1000

# High-Performance Server Configuration
ENV LLAMA_ARG_MODEL=/models/model.gguf
ENV LLAMA_ARG_HOST=0.0.0.0
ENV LLAMA_ARG_PORT=7860
ENV LLAMA_ARG_CTX_SIZE=8192

# --- PERFORMANCE TUNING ---
# 1. Threading: Match physical cores. (Standard 16GB PC usually has 8-12 cores)
ENV LLAMA_ARG_THREADS=8
# 2. Flash Attention: Critical for speed (30% + boost)
ENV LLAMA_ARG_FLASH_ATTN=true
# 3. No MMap: Sometimes faster on RAM-constrained systems to load fully into memory
ENV LLAMA_ARG_NO_MMAP=false 
# 4. Memory Locking: Prevents the OS from swapping the model to disk
ENV LLAMA_ARG_MLOCK=true 
# 5. Batching: Smaller batches for faster single-user response
ENV LLAMA_ARG_BATCH_SIZE=512
ENV LLAMA_ARG_UBATCH_SIZE=128

HEALTHCHECK --interval=30s --timeout=15s --start-period=10s --retries=3 \
    CMD curl -f http://localhost:7860/health || exit 1

ENTRYPOINT ["/app/llama-server"]