late-interaction Open weights ST compatible text
lightonai/LateOn-Code-edge-pretrain
- Languages
- Englishgo-Code (Code)java-Code (Code)javascript-Code (Code)php-Code (Code)python-Code (Code)ruby-Code (Code)
- License
- apache-2.0
- Training code
- https://github.com/lightonai/pylate/blob/main/examples/train/lateon_code/pre-training.py
- Training data
- https://huggingface.co/datasets/lightonai/cornstack
- Adapted from
- mixedbread-ai/mxbai-edge-colbert-v0-17m
- Trained on
- AmazonQACornStackHotpotQALoTTEMSMARCOMultiLongDocRetrievalNQ
Parameters 17M
Active parameters 4M
Embedding dim 48
Max tokens 7,999
Memory 64 MB
Released 2026-02-12
Openness
5/6
- Open weights
- Open license
- Training code
- Training data
- Paper
- Model card