GLiClass instruct large v1.0
KnowledgatorOpen weights
The newest GLiClass, trained with prompts and label descriptions.
- typed-decisions
- —
- Metatext suite
- 0.545
- ECE untuned
- 0.308
- p50 per case
- 1.0 s
not run
3/7 over baseline
four CPU cores
Specifications
- Checkpoint
- knowledgator/gliclass-instruct-large-v1.0
- Publisher
- Knowledgator
- Mechanism
- Label-relevance classifier
- Backbone
- DeBERTa-v3-large, instruction-tuned
- Parameters
- ~440M
- License
- Apache 2.0
- Runs as
- open weights, self-hosted
- Library
- gliclass
- Primitives
- Choice, Noul
- Pricing
- none — your compute
Metatext decisionsrank 11 of 21
Full leaderboard →| Configuration | Mode | Decisions | Beats baseline | ECE |
|---|---|---|---|---|
| criteria + single-hypothesis Noul (correct payload)headline | serial | 0.545 | 3/7 | 0.308 |
| default payload | serial | 0.370 | 2/7 | 0.553 |
| criteria folded into labels | serial | 0.361 | 2/7 | 0.536 |
Per-question accuracy — criteria + single-hypothesis Noul (correct payload), serial mode, 120 judgments
| Task | Question | Type | Accuracy | Baseline | Beats? |
|---|---|---|---|---|---|
| agent_traces | outcome | choice | 0.357 | 0.357 | ❌ |
| agent_traces | needs_human | noul | 0.714 | 0.714 | ❌ |
| customer_service | department | choice | 0.650 | 0.400 | ✅ |
| customer_service | refund_requested | noul | 0.750 | 0.700 | ✅ |
| customer_service | frustration | score | 0.300 | 0.500 | ❌ |
| security_incidents | severity | choice | 0.375 | 0.312 | ✅ |
| security_incidents | actively_exploited | noul | 0.688 | 0.688 | ❌ |
Frequently asked questions
How does GLiClass instruct large v1.0 compare to Jev?
On the Metatext suite GLiClass instruct large v1.0 scores 0.545 against Jev's 0.863.
Is GLiClass instruct large v1.0 open source?
Yes. knowledgator/gliclass-instruct-large-v1.0 is released under Apache 2.0 and runs on your own hardware with the gliclass library.
Can GLiClass instruct large v1.0 extract entities?
No. GLiClass instruct large v1.0 answers questions about text and never returns spans; use a GLiNER-family model for extraction.