gliner2-candle

Rust + Candle port of GLiNER2 entity extraction (not the original GLiNER).

Quick Start

1. Download a model (example):

hf download fastino/gliner2-large-v1 --local-dir ./model

2. Run inference:

cargo run --release -- \
  --model-dir ./model \
  --text "Apple was founded by Steve Jobs in Cupertino." \
  --entities "person,organization,location"

Notes

• This project targets the fastino/gliner2-* model family.
• CPU performance improves significantly with --release.
• GPU/accelerated backends are supported by Candle, but not configured here by default.

CLI

• --model-dir Path to a directory containing config.json, encoder_config/config.json, and model.safetensors.
• --text Input text.
• --entities Comma-separated list of entity types.
• --threshold Optional score threshold (default in code).
• --list-weights N Lists the first N weight keys (debug).

Source Layout

• src/model.rs Model load, forward, predict
• src/processor.rs Tokenization and input building
• src/span_rep.rs Span representation
• src/count_lstm.rs Count-aware structure embedding
• src/inference.rs Post-processing and span extraction
• src/main.rs CLI

License

MIT