#!/bin/bash
# Download both specialist GGUFs (public repos), then run llama-swap on 7860.
set -e

download() {
  local repo="$1" file="$2" dest="/models/$2"
  if [ -f "$dest" ]; then echo "✅ cached: $dest"; return; fi
  echo "⬇️  downloading $file from $repo ..."
  curl -fL --retry 3 --retry-delay 5 -o "$dest.part" \
    "https://huggingface.co/$repo/resolve/main/$file"
  mv "$dest.part" "$dest"
  echo "✅ ready: $dest"
}

download "LiquidAI/LFM2.5-1.2B-Thinking-GGUF" "LFM2.5-1.2B-Thinking-Q8_0.gguf"
download "LiquidAI/LFM2-1.2B-Tool-GGUF"       "LFM2-1.2B-Tool-Q8_0.gguf"

echo "Starting llama-swap on :7860 (models: lfm-thinking, lfm-tool)"
exec /usr/local/bin/llama-swap --config /app/config.yaml --listen 0.0.0.0:7860