entrypoint.sh

#!/bin/bash
set -e

echo "=================================="
echo "Starting HF Space with llama.cpp"
echo "=================================="

if [ ! -d /data ]; then
  export HF_HOME=/workspace/data/.cache/huggingface
  export HUGGINGFACE_HUB_CACHE=/workspace/data/.cache/huggingface
  export MODEL_CACHE_DIR=/workspace/data/models
else
  export HF_HOME=/data/.cache/huggingface
  export HUGGINGFACE_HUB_CACHE=/data/.cache/huggingface
  export MODEL_CACHE_DIR=/data/models
fi

mkdir -p "${HF_HOME}" "${HUGGINGFACE_HUB_CACHE}" "${MODEL_CACHE_DIR}"

echo "MODEL_REPO_ID=${MODEL_REPO_ID}"
echo "MODEL_FILENAME=${MODEL_FILENAME}"
echo "PORT=${PORT}"
echo "HF_HOME=${HF_HOME}"
echo "MODEL_CACHE_DIR=${MODEL_CACHE_DIR}"

# Download GGUF if missing
if [ ! -f "${MODEL_CACHE_DIR}/${MODEL_FILENAME}" ]; then
  echo "Downloading GGUF from Hugging Face..."
  python3 - <<'PY'
import os
from huggingface_hub import hf_hub_download

repo = os.environ["MODEL_REPO_ID"]
filename = os.environ["MODEL_FILENAME"]
local_dir = os.environ["MODEL_CACHE_DIR"]

path = hf_hub_download(
    repo_id=repo,
    filename=filename,
    local_dir=local_dir,
)
print(f"Downloaded model to: {path}")
PY
else
  echo "Model already present: ${MODEL_CACHE_DIR}/${MODEL_FILENAME}"
fi

echo "Starting llama.cpp server..."
export LD_LIBRARY_PATH="/app:/app/build/bin:/app/build/lib:/usr/local/lib:${LD_LIBRARY_PATH}"
echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}"

# Optional debug
find /app -name "libmtmd.so*" -o -name "libllama.so*" -o -name "libggml*.so*" || true
ldd /app/llama-server || true
ldconfig || true

/app/llama-server \
  -m "${MODEL_CACHE_DIR}/${MODEL_FILENAME}" \
  --host 127.0.0.1 \
  --port 8000 \
  > /tmp/llama.log 2>&1 &

LLAMA_PID=$!
echo "llama.cpp PID: ${LLAMA_PID}"

echo "Waiting for llama.cpp server..."
for i in $(seq 1 60); do
  if curl -sf http://127.0.0.1:8000/v1/models >/dev/null; then
    echo "llama.cpp is ready"
    break
  fi
  echo "waiting... ($i/60)"
  sleep 2
done

if ! curl -sf http://127.0.0.1:8000/v1/models >/dev/null; then
  echo "llama.cpp failed to start"
  echo "==== llama.cpp logs ===="
  cat /tmp/llama.log || true
  exit 1
fi

cd /workspace
exec python3 app_gradio.py