FROM ubuntu:24.04

# ── System dependencies ───────────────────────────────────────────────────────
ENV DEBIAN_FRONTEND=noninteractive
RUN apt-get update && apt-get install -y --no-install-recommends \
        build-essential \
        cmake \
        curl \
        wget \
        ca-certificates \
        python3 \
        python3-pip \
        python3-dev \
        python3-venv \
        git \
    && rm -rf /var/lib/apt/lists/*

# ── Python env ────────────────────────────────────────────────────────────────
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

# PyTorch CPU (conversion only — no GPU needed)
RUN pip install --upgrade pip && \
    pip install --no-cache-dir \
        torch==2.5.1+cpu \
        --index-url https://download.pytorch.org/whl/cpu

# HF ecosystem + conversion deps
RUN pip install --no-cache-dir \
        transformers \
        huggingface_hub[cli] \
        safetensors \
        sentencepiece \
        tiktoken \
        gguf \
        numpy \
        tqdm

# ── Clone llama.cpp (fast — no compile here, that happens at container start) ─
# We need the source for two things:
#   1. convert_hf_to_gguf.py (Python, runs immediately)
#   2. llama-quantize source (compiled at runtime to avoid build timeout)
RUN git clone --depth 1 https://github.com/ggml-org/llama.cpp.git /opt/llama.cpp
RUN pip install --no-cache-dir -r /opt/llama.cpp/requirements.txt || true

# ── Runtime secrets ───────────────────────────────────────────────────────────
ENV HF_TOKEN=""
ENV HF_DATASET_REPO=""
ENV LLAMA_CPP_DIR="/opt/llama.cpp"

# ── Working directories ───────────────────────────────────────────────────────
RUN mkdir -p /workspace/model-fp8 \
             /workspace/model-bf16 \
             /workspace/model-gguf \
             /workspace/output

WORKDIR /workspace

COPY scripts/ /workspace/scripts/
RUN chmod +x /workspace/scripts/*.sh

ENTRYPOINT ["/bin/bash", "/workspace/scripts/entrypoint.sh"]