Environment Setup for Semantic Search
Instructions
glhf uses model2vec-rs with Potion-base-32M for embeddings. The model downloads automatically on first use.
Quick Setup
No manual setup required:
bash
cargo build --release ./target/release/glhf index
The embedding model (~130MB) will download to the HuggingFace cache on first run.
Model Details
| Property | Value |
|---|---|
| Model | minishlab/potion-base-32M |
| Dimensions | 512 |
| Size | ~130MB |
| Cache Location | ~/.cache/huggingface/ |
Verify Setup
bash
# Run embedding tests cargo test embed -- --ignored
Common Issues
| Error | Solution |
|---|---|
Failed to load model | Check internet connection, model will auto-download |
No space left | Clear HuggingFace cache: rm -rf ~/.cache/huggingface/ |
Slow first run | Normal - model downloads once, then cached |
Skip Embeddings
For text-only search (faster indexing, no model download):
bash
glhf index --skip-embeddings
This enables FTS5 search but disables semantic/hybrid modes.