#!/usr/bin/env python3
"""Download all-MiniLM-L6-v2 ONNX model for semantic search.
This script downloads the pre-converted ONNX model from HuggingFace.
The Xenova model is optimized for ONNX Runtime and requires no PyTorch.
Usage:
python scripts/download_embedding_model.py
"""
import logging
from pathlib import Path
try:
from huggingface_hub import snapshot_download
from transformers import AutoTokenizer
except ImportError as e:
print(f"❌ Missing required dependency: {e}")
print("\nInstall with:")
print(" uv pip install transformers huggingface_hub")
print("\nOr install all dependencies:")
print(" uv sync --group dev")
raise SystemExit(1)
logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
logger = logging.getLogger(__name__)
def download_onnx_model() -> None:
"""Download Xenova/all-MiniLM-L6-v2 ONNX model from HuggingFace."""
# Xenova provides pre-converted ONNX models
model_name = "Xenova/all-MiniLM-L6-v2"
logger.info(f"Downloading {model_name} from HuggingFace...")
logger.info("This is a pre-converted ONNX model - no PyTorch required!")
logger.info("")
try:
# Download model files including ONNX
logger.info("Downloading model files (tokenizer + ONNX model)...")
cache_dir = snapshot_download(
repo_id=model_name,
allow_patterns=[
"*.json",
"*.txt",
"*.onnx",
"model.*",
"tokenizer*",
"vocab*",
"config*",
],
)
logger.info(f"✅ Model files downloaded to: {cache_dir}")
# Verify ONNX model exists
onnx_files = list(Path(cache_dir).glob("*.onnx"))
if onnx_files:
logger.info(f"✅ Found {len(onnx_files)} ONNX file(s):")
for onnx_file in onnx_files:
logger.info(f" • {onnx_file.name}")
else:
logger.warning(
"⚠️ No ONNX files found - model may still work with tokenizers"
)
logger.info("")
logger.info("=" * 70)
logger.info("✅ ONNX Embedding Model Downloaded Successfully!")
logger.info("=" * 70)
logger.info("")
logger.info("Model Details:")
logger.info(f" • Model: {model_name}")
logger.info(" • Type: Pre-converted ONNX (no PyTorch needed)")
logger.info(f" • Cache: {cache_dir}")
logger.info(" • Dimensions: 384")
logger.info(" • Max tokens: 256 (recommended: 128)")
logger.info("")
logger.info("Next steps:")
logger.info(" 1. Update reflection adapter to use this model")
logger.info(" 2. Restart the MCP server")
logger.info(" 3. Use semantic search: reflect_on_past('your query')")
logger.info("")
except Exception as e:
logger.exception(f"❌ Failed to download model: {e}")
logger.info("\nTroubleshooting:")
logger.info(" • Check internet connection")
logger.info(" • Verify HuggingFace Hub is accessible")
logger.info(" • Try: pip install --upgrade transformers huggingface_hub")
raise SystemExit(1)
if __name__ == "__main__":
download_onnx_model()
