#!/bin/sh
set -eu

MODEL_PATH="${MODEL_PATH:-/models/NVIDIA-Nemotron-3-Nano-4B-BF16.Q4_K_M.gguf}"
PORT="${PORT:-7860}"
CTX_SIZE="${CTX_SIZE:-2048}"
N_GPU_LAYERS="${N_GPU_LAYERS:-0}"

SERVER_BIN="/app/llama-server"
if [ ! -x "$SERVER_BIN" ]; then
  SERVER_BIN="$(command -v llama-server || true)"
fi
if [ -z "$SERVER_BIN" ]; then
  echo "Could not find llama-server" >&2
  find / -name llama-server -type f 2>/dev/null | head >&2 || true
  exit 1
fi

if [ ! -f "$MODEL_PATH" ]; then
  echo "Model not found at $MODEL_PATH" >&2
  ls -lah /models || true
  exit 1
fi

echo "Starting llama.cpp server with $MODEL_PATH"
exec "$SERVER_BIN" \
  -m "$MODEL_PATH" \
  --host 0.0.0.0 \
  --port "$PORT" \
  -c "$CTX_SIZE" \
  -ngl "$N_GPU_LAYERS"