laya
Datasets
All datasets matching “laya”laya-bio
Laya-Bio: short-sequence candidate-scoring benchmark and reproducibility data
This repository packages the data and saved results used by Laya-Bio: Candidate Scoring and Reliability on Short Biological Sequences (Liang Wang, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology). The main study uses two closed-set tasks, four model conditions and three training seeds, with no additional neural continual pretraining.
Companion… See the full description on the dataset page: https://huggingface.co/datasets/dnagpt/laya-bio.laya-jev-benchmark
Laya vs Jev
TypeSafe released Jev on 15 September 2026, a closed Model that returns typed
Decisions instead of Text. Three Days later an open Reproduction appeared,
Laya (convaiinnovations/laya, Apache 2.0, 421M).
Laya's Model Card claims 83.8% against Jev's 67.8% and calls it a "+16.0%
Advantage". Those two Numbers are from two different Benchmarks, so the
Comparison says nothing.
I ran both on Benchmarks where Jev has published Numbers. One RTX 5090.
Everything below is… See the full description on the dataset page: https://huggingface.co/datasets/Luni/laya-jev-benchmark.open-jev-laya-benchlaya-bio-historical-corpora
Laya-Bio: historical corpora and exact BPE training samples
Author: Liang Wang, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology.
This is an archival snapshot of seven historical raw corpus files plus the two exact sampled files used to fit the inherited DNA/protein BPE tokenizers. The archive preserves source bytes, including original line endings and record order. Names such as 32g are historical names; measured byte counts below… See the full description on the dataset page: https://huggingface.co/datasets/dnagpt/laya-bio-historical-corpora.laya-formatting-fragility
Laya Formatting Fragility — a perturbation suite for typed decision models
Small decision models ("decide, don't chat" — Laya, AgentJev and friends)
return typed answers with probabilities in milliseconds. Their answers can
be sensitive to formatting: option key names, option order, and state
phrasing change results even when the situation and the gold answer are
identical. Vendor benchmarks don't measure this. This dataset does.
Every row pair differs in exactly one formatting… See the full description on the dataset page: https://huggingface.co/datasets/pranaysuyash/laya-formatting-fragility.laya-session-guard-risk-ranked-metrics
Laya risk-ranked ASPI run: RESEARCH-ONLY, REJECTED, metrics only, no weights published
RESEARCH-ONLY. Negative result. Rejected as a replacement for any deployed guard.
Not a safety guard and not for enforcement.
This is a metrics release. No checkpoint weights are published here, and none are
published anywhere: the run was rejected before its weights were ever downloaded, and they
remain in a private Kaggle output. Nothing in this repository can be loaded as a model.
The run… See the full description on the dataset page: https://huggingface.co/datasets/hxrikp/laya-session-guard-risk-ranked-metrics.