rel_only → ner_only → emb_only).OnarexModel.save_pretrained (config.yaml + pytorch_model.bin).| Field | Value |
|---|---|
| Encoder | answerdotai/ModernBERT-large |
| Hidden size | 1024 |
| BiLSTM | True |
| Checkpoint phase | emb_only |
| Selection metric | triple_emb_cosine = 0.957874 (step 120000) |
config.yaml — model / training / eval hyperparameterspytorch_model.bin — full state_dictbest.json — optional metadata for the best validation checkpointREADME.md — this model cardonarex Python package is not published yet. With the research repo installed:1from onarex.model import OnarexModel
2
3model = OnarexModel.from_pretrained("cb-ai/onarex")
4# or from a local download of this folder
5# model = OnarexModel.from_pretrained("./onarex_model_emb_only_best")inference.py); they are not bundled as a standalone Hub pipeline here.pipeline without the Onarex codebaserel_only — REL trunk (+ encoder / BiLSTM when enabled)ner_only — NER trunk (encoder frozen)emb_only — emb trunk (encoder frozen); this upload is typically the final emb best ckptemb_only are ModernBERT-pooled description vectors (see training config data.* paths).1@misc{onarex2026,
2 title={Onarex: Ontology-conditioned NER, Relation Extraction, and Triple Embedding},
3 year={2026},
4 howpublished={Hugging Face model card: cb-ai/onarex},
5}