#!/bin/bash

# Report descrittivi: 0.6 ok; 0.55 più stabile
TEMP=${BASE_TEMP:-0.6}
exec /app/llama-server $BASE_MEDIUM_MODEL \
  -c $BASE_CONTEXT_SIZE -ngl $BASE_GPU_LAYERS -n $BASE_MAX_TOKENS \
  --temp $TEMP --top-p 0.9 --top-k 40 --repeat-penalty 1.1 \
  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
  --jinja \
  --timeout 600 --host 0.0.0.0 --port 8092 &
PID=$!

cleanup() {
    echo "Stopping llama-server..."
    kill $PID 2>/dev/null
    wait $PID 2>/dev/null
    exit 0
}

trap cleanup SIGTERM SIGINT

wait $PID