GLiNER multitask v1.0
KnowledgatorOpen weights
The only checkpoint in the original GLiNER library that claims text classification. Classification is done by pointing the extractor at label strings, which is why it is slow and weak on decisions.
- typed-decisions
- —
- Metatext suite
- 0.378
- ECE untuned
- 0.223
- p50 per case
- 1.8 s
not run
1/7 over baseline
four CPU cores
Specifications
- Checkpoint
- knowledgator/gliner-multitask-v1.0
- Publisher
- Knowledgator
- Mechanism
- Span extractor pointed at labels
- Backbone
- DeBERTa-v3 (original gliner library)
- Parameters
- ~200M
- License
- Apache 2.0
- Runs as
- open weights, self-hosted
- Library
- gliner
- Primitives
- Choice, Noul, Extraction
- Pricing
- none — your compute
Metatext decisionsrank 21 of 21
Full leaderboard →| Configuration | Mode | Decisions | Beats baseline | ECE |
|---|---|---|---|---|
| class names as labelsheadline | serial | 0.378 | 1/7 | 0.223 |
| extractor pointing at a label (model-card recipe) | serial | 0.351 | 1/7 | 0.508 |
Per-question accuracy — class names as labels, serial mode, 120 judgments
| Task | Question | Type | Accuracy | Baseline | Beats? |
|---|---|---|---|---|---|
| agent_traces | outcome | choice | 0.071 | 0.357 | ❌ |
| agent_traces | needs_human | noul | 0.286 | 0.714 | ❌ |
| customer_service | department | choice | 0.650 | 0.400 | ✅ |
| customer_service | refund_requested | noul | 0.700 | 0.700 | ❌ |
| customer_service | frustration | score | 0.300 | 0.500 | ❌ |
| security_incidents | severity | choice | 0.250 | 0.312 | ❌ |
| security_incidents | actively_exploited | noul | 0.562 | 0.688 | ❌ |
Frequently asked questions
How does GLiNER multitask v1.0 compare to Jev?
On the Metatext suite GLiNER multitask v1.0 scores 0.378 against Jev's 0.863.
Is GLiNER multitask v1.0 open source?
Yes. knowledgator/gliner-multitask-v1.0 is released under Apache 2.0 and runs on your own hardware with the gliner library.
Can GLiNER multitask v1.0 extract entities?
Yes. GLiNER multitask v1.0 has a span head.