#!/usr/bin/env bash
set -euo pipefail

MODEL_REPO="${MELTMIND_GGUF_REPO:-AyyYOO/MiniCPM4-8B-Q4_K_M-GGUF}"
MODEL_FILE="${MELTMIND_GGUF_FILE:-minicpm4-8b-q4_k_m.gguf}"
MODEL_DIR="${MELTMIND_MODEL_DIR:-$PWD/.runtime/models}"
MODEL_PATH="$MODEL_DIR/$MODEL_FILE"
PROJECT_LLAMA_SERVER="$PWD/.runtime/llama.cpp/llama-b9637/llama-server"
LLAMA_SERVER="${LLAMA_SERVER_BIN:-$PROJECT_LLAMA_SERVER}"

if [[ ! -x "$LLAMA_SERVER" ]] && ! command -v "$LLAMA_SERVER" >/dev/null 2>&1; then
  echo "llama-server was not found. Install llama.cpp, then rerun this script."
  echo "Project: https://github.com/ggml-org/llama.cpp"
  exit 1
fi

mkdir -p "$MODEL_DIR"
if [[ ! -f "$MODEL_PATH" ]]; then
  echo "Downloading $MODEL_REPO/$MODEL_FILE into $MODEL_DIR"
  hf download "$MODEL_REPO" "$MODEL_FILE" --local-dir "$MODEL_DIR"
fi

exec "$LLAMA_SERVER" \
  --model "$MODEL_PATH" \
  --host "${MELTMIND_LLM_HOST:-127.0.0.1}" \
  --port "${MELTMIND_LLM_PORT:-8080}" \
  --ctx-size "${MELTMIND_CONTEXT_SIZE:-32768}" \
  --n-gpu-layers "${MELTMIND_GPU_LAYERS:-99}" \
  --threads "${MELTMIND_THREADS:-8}" \
  --parallel "${MELTMIND_PARALLEL:-1}" \
  --jinja