Fastino Releases GLiNER2.5-Decide: A 340M Open-Weight Decision Model That Runs on CPU
Fastino Labs released GLiNER2.5-Decide, a 340M Apache 2.0 open-weight decision model for agent routing, triage, and guardrails that runs on CPU.
GLiNER2.5-Decide is a non-generative 340M-parameter classifier built on a DeBERTa-v3-large encoder, fine-tuned from gliner2-large-v1, that returns structured answers with probabilities, confidence, and feasibility metadata via two-stage constrained joint decoding. It scored 60.1% exact-match accuracy on Fastino's internal 17-dataset Fast Decisions suite, beating 4B-class decoders, with p50 latency of 167.3 ms on CPU and 38.3 ms on an NVIDIA V100. Family variants include GLiNER2.5-Decide-1B (59.6%) and GLiNER2.5-multi-Decide (287M, 56.7%); weights ship under Apache 2.0 and support air-gapped deployment.
- 340M open-weight model, Apache 2.0, runs on CPU or air-gapped environments
- Constrained joint decoding enforces cross-question rules for guardrail decisions
- 60.1% average on internal 17-dataset suite, leading 9 of 17 datasets
- 167.3 ms p50 CPU latency at 64 tokens; 38.3 ms on V100
- Organizations
- Fastino LabsGLiNER2.5-Decide
From the original article
4 min read at marktechpost.com
Fastino Labs has released GLiNER2. 5-Decide, a 340M-parameter open-weight decision model. It takes text and a schema of typed questions and returns structured answers.
By Sana Hassan · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.