madewithlaya

Catalog / Documents & data

0104GitHub

zero-shot-ie-bench: GLiNER 2.5, GLiFormer, GLiClass, Laya, Von and Jev in one web UI

Seven zero-shot extraction and classification families demoed, benchmarked and cross-compared, with sizes, licences and costs in one table.

Open source ↗ github.comcostfree for the local rowstime-
umstek/zero-shot-ie-benchREADME ↗
# zero-shot-ie-bench

Seven zero-shot information-extraction and classification families —
extractor encoders, a purpose-built classifier, and typed-decision engines
(local and cloud) — demoed, benchmarked, and cross-compared in one repo
with a web UI.

| System | Kind | Size | License / cost |
|---|---|---|---|
| [GLiNER 2.5](https://github.com/fastino-ai/GLiNER2) (`fastino/gliner2.5-*`) | local extractor encoder (boundary arch) | 74M / 194M / 287M | Apache 2.0, free |
| [GLiFormer](https://github.com/Knowledgator/GLiFormer) (`knowledgator/gliformer-*`) | local extractor encoder (layout-aware DeBERTa) | ~190M / 575.6M | Apache 2.0, free |
| [GLiClass](https://github.com/knowledgator/gliclass) (`knowledgator/gliclass-*-v3.0`) | local zero-shot classifier (all labels, one pass) | 33M / 151M / 187M / 439M | Apache 2.0, free |
| [Laya](https://huggingface.co/convaiinnovations/laya) (`laya`) | local typed-decision engine (choice/score/noul) | 421M (322M multilingual) | Apache 2.0, free |
| [von-1.0](https://huggingface.co/wfzyx/von-1.0) (`von-sdk`) | local typed-decision engine (System One protocol) | 396M | Apache 2.0, free |
| [open-alternative-jev](https://github.com/ikermoel/open-alternative-jev) (`so1`) | local decision harness over any ChatML LLM (logprobs) | BYO LLM (tested Qwen2.5-0.5B) | MIT, free |
| [Jev](https://www.typesafe.ai/) (`jev-latest` via System One API) | cloud typed-decision engine (choice/score/noul) | closed | paid API |

The GLiNER lineage forked: Urchade Zaratiana (original GLiNER author,
ex-Knowledgator) is on the GLiNER2 paper with Fastino; Knowledgator kept the
classic `gliner` package, built GLiFormer on it and also maintains GLiClass.
Laya's card positions it explicitly as the open local counterpart of cloud
Jev; von speaks the same System On

Also filed under Documents & data

  1. 0103

    Zero-shot benchmark: Laya's advertised win over Jev does not reproduce on four datasets

    Frozen protocol, raw predictions, calibration and latency for Jev 1.13, PrismNLI-0.4B and Laya. Laya ties Jev on dair-ai/emotion and trails by 10–33 points elsewhere.

    @elcronos · Documents & data

  2. 0102

    jevbench: Jev vs GPT-5-mini, Claude Sonnet 5, fine-tuned DistilBERT, BART NLI and Laya

    Six classifiers over SST-2, AG News and more, reporting accuracy, macro-F1, ECE, p50/p95 latency, throughput and cost side by side. Laya runs locally on MPS or CPU.

    @dhruvmehra · Documents & data · $0 for the local rows

  3. 0098

    Agentic GraphRAG over S&P 10-Ks with Laya as a shadow decision engine beside Jev

    Anthony Alcaraz's talk demo routes typed decisions through Jev and shows Laya's answers beside them. On the TSMC run Laya rejected every true claim; that's the point.

    @AnthonyAlcaraz · Documents & data · free (Laya shadow); Bedrock for reasoning

  4. 0087

    chunklaya: why Laya stops reading after ~200 tokens, and a chunking harness that fixes it

    Measures the decision head's attention drop-off on long inputs across 400 documents, then chunks, screens and aggregates, or locates the one passage that matters.

    @myxamediyar · Documents & data · free, no extra cost over a larger context window