Setting	Value
Trained on	2026-07-19
Duration	1h 3m
Throughput	9.9 samples/s
Epochs	10
Batch size	2 (× 8 grad-accum)
Encoder LR	1e-05
Task-head LR	0.0003
Weight decay	0.01
Scheduler	cosine_restarts (warmup 0.05)
Precision	bf16
Max grad norm	1.0
Best-checkpoint metric	eval_relation_strict_micro_f1
Seed	42

Evaluation

Decision threshold: 0.5 (config default).

Blind test (held-out test splits)

Micro precision / recall / F1, strict → relaxed.

Category	Precision	Recall	F1	Support
entity	0.868 → 0.915	0.853 → 0.900	0.860 → 0.907	10705
relation	0.633 → 0.771	0.186 → 0.227	0.287 → 0.350	17348

Best checkpoint (validation)

Micro precision / recall / F1, strict → relaxed.

Category	Precision	Recall	F1	Support
entity	0.856 → 0.908	0.842 → 0.893	0.849 → 0.900	12255
relation	0.650 → 0.792	0.195 → 0.237	0.300 → 0.365	18321

License

Effective license: Unverified — review required. This model is a derivative of its base model and every training dataset, so the most restrictive term across all of them governs the whole model.

Commercial use: Unverified
Share-alike obligation: No
All licenses verified: No
Base model: gliner2-large-v1 — see model card

Unverified — verify the upstream terms before redistribution

Re-DocRED (see card)
gliner2-large-v1 (see model card)

License strings are copied verbatim from each dataset's card/source and from tools/train/dataset_registry.yaml. "see card"/"see source"/"other" mean the upstream declares no clear license — treat as unverified. This summary is informational, not legal advice; confirm terms before redistribution or commercial use.

Citation

If you use this model, please cite GLiNER2 and the underlying datasets (linked in Training data).

Model card generated automatically at the end of training (2026-07-19).

whr778/gliner2-large-v1-redocred

gliner2-large-v1_redocred

⚠️ License at a glance

Model details

Training data

Training procedure

Evaluation

Blind test (held-out test splits)

Best checkpoint (validation)

License

Citation

whr778/
gliner2-large-v1-redocred