scripts/setup_models.sh

#!/usr/bin/env bash
# factory-rag model setup: Ollama pulls + docling artifacts + HF models.
set -euo pipefail

# --- Ollama models ---
ollama pull qwen3.6:27b
ollama pull qwen3-vl:8b

# --- docling artifacts (positional args, NO --only flag -- verified) ---
docling-tools models download layout tableformer rapidocr

# --- embedder + reranker via snapshot_download ---
python - <<'PY'
from huggingface_hub import snapshot_download
for repo in [
    "Snowflake/snowflake-arctic-embed-l-v2.0",
    "BAAI/bge-reranker-v2-m3",
]:
    print("prefetching", repo)
    snapshot_download(repo)
PY

# --- offline-run guidance ---
echo ""
echo "For offline runs set:" 
echo "  export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1"
echo "  export DOCLING_ARTIFACTS_PATH=~/.cache/docling/models  (default)"