---
library_name: gliner2
license: other
license_name: unverified-review-required
base_model: fastino/gliner2-multi-v1
language:
  - en
  - de
  - es
  - fr
  - ru
  - ar
  - zh
  - hi
  - sw
  - yo
datasets:
  - unimelb-nlp/wikiann
tags:
  - gliner2
  - information-extraction
  - named-entity-recognition
  - relation-extraction
  - event-extraction
  - text-classification
metrics:
  - f1
  - precision
  - recall
pipeline_tag: token-classification
---

# gliner2-multi-v1_wikiann

A [GLiNER2](https://github.com/fastino-ai/GLiNER2) multi-task information-extraction model (entities, relations, events, and classification) fine-tuned from `fastino/gliner2-multi-v1`.

## ⚠️ License at a glance

- **Effective license:** Unverified — review required
- **Commercial use:** Unverified
- **All dataset licenses verified:** No

See [License](#license) for the full determination and per-dataset terms.

## Model details

- **Base model:** [`fastino/gliner2-multi-v1`](https://huggingface.co/fastino/gliner2-multi-v1)
- **Library:** `gliner2`
- **Tasks:** entity, relation, event, and classification extraction
- **Experiment:** `gliner2-multi-v1_wikiann`

## Training data

**1** dataset used for this run.

| Dataset | Task(s) | Language | License | Source |
|---|---|---|---|---|
| WikiANN (PAN-X) | NER (176 languages, PER/ORG/LOC) | en, de, es, fr, ru, ar, zh, hi, sw, yo | see card | [link](https://huggingface.co/datasets/unimelb-nlp/wikiann) |

**Dataset notes**

- **WikiANN (PAN-X)** — Silver-standard Wikipedia NER over 176 languages (Pan et al. 2017; Rahimi et al. 2019 balanced splits); PER/ORG/LOC mapped to person/organization/location. Streamed at train time via data.hf_streaming, never written to disk; the trained language subset is set per run and shown above.

## Training procedure

| Setting | Value |
|---|---|
| Trained on | 2026-08-02 |
| Duration | — |
| Epochs | 1 |
| Batch size | 16 (× 2 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_entity_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 | 0.941 → 0.970 | 0.870 → 0.896 | 0.904 → 0.932 | 11112 |

### Best checkpoint (validation)

Micro precision / recall / F1, strict → relaxed.

| Category | Precision | Recall | F1 | Support |
|---|--:|--:|--:|--:|
| entity | 0.937 → 0.967 | 0.864 → 0.892 | 0.899 → 0.928 | 11165 |

## 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-multi-v1 — see model card

**Unverified — verify the upstream terms before redistribution**
- WikiANN (PAN-X) (see card)
- gliner2-multi-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](#training-data)).

---
_Model card generated automatically at the end of training (2026-08-02)._