#!/usr/bin/env bash
set -eu

MODEL_PATH="${MODEL_PATH:-/models/Qwen3-1.7B-Instruct-Q4_K_M.gguf}"
LLM_MODEL_NAME="${LLM_MODEL_NAME:-qwen3-1.7b-instruct-q4_k_m}"
LLM_CONTEXT_SIZE="${LLM_CONTEXT_SIZE:-4096}"
LLM_THREADS="${LLM_THREADS:-4}"
LLM_GPU_LAYERS="${LLM_GPU_LAYERS:-0}"
LLM_PORT="${LLM_PORT:-8080}"
LLAMA_SERVER_PORT="${LLAMA_SERVER_PORT:-8081}"

if [ ! -f "$MODEL_PATH" ]; then
  echo "llm startup failed: model file not found at $MODEL_PATH" >&2
  echo "Mount a GGUF model into ./models/qwen3 and set MODEL_PATH if the filename differs." >&2
  exit 1
fi

if [ ! -r "$MODEL_PATH" ]; then
  echo "llm startup failed: model file is not readable at $MODEL_PATH" >&2
  exit 1
fi

echo "Starting llm shim model=$LLM_MODEL_NAME model_path=$MODEL_PATH public_port=$LLM_PORT upstream_port=$LLAMA_SERVER_PORT ctx=$LLM_CONTEXT_SIZE threads=$LLM_THREADS gpu_layers=$LLM_GPU_LAYERS"

exec python -m uvicorn main:app --host 0.0.0.0 --port "$LLM_PORT"