Laya fine-tune that scans untrusted text (extracted files, knowledge-base documents, agent skills, tool descriptions) for prompt injection and data exfiltration in English and German, as a noul question answered in one forward pass. 322M parameters, PyTorch and ONNX.
An independent Jev-shaped reproduction on DeBERTa-v3-large. One state and any number of choice, score and yes/no questions go in, and a calibrated distribution per question comes out of one forward pass. Public gold labels only.
Decides
noul, classify
choice, score, noul, classify
Architecture
laya
open-jev
Fine-tuned from
convaiinnovations/laya-multilingual
microsoft/deberta-v3-large
License
Unspecified; training data includes CC BY-NC-SA 4.0 material (non-commercial), not relicensed
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
—
85.4%
Calibration error
—
0.022
Valid action rate
—
—
Median latency
95.4 ms
28 ms
p95 latency
149 ms
Figures are from each model’s manifest; accuracy and latency are what the publishers report, on their own suites and hardware. Add a third model.
Questions
What is the difference between laya-cybersec and open-jev-deberta-v3-large?
laya-cybersec is from TextCortex and open-jev-deberta-v3-large from Kotoba Labs. Both have open weights you can download and run. Both answer noul and classify questions. Only open-jev-deberta-v3-large answers choice and score. laya-cybersec reads up to 1K tokens of state, against 512 tokens for open-jev-deberta-v3-large. laya-cybersec is the smaller model, at 322M parameters to 434M. laya-cybersec is licensed other; open-jev-deberta-v3-large, apache-2.0.
Which is more accurate, laya-cybersec or open-jev-deberta-v3-large?
Only open-jev-deberta-v3-large publishes an accuracy figure (85.4% on Kotoba held-out test, seen question types (banking77, SST-5, BoolQ)); laya-cybersec does not, so there is no comparison to make without your own test.
Which is cheaper, laya-cybersec or open-jev-deberta-v3-large?
laya-cybersec: Free (open weights). open-jev-deberta-v3-large: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Which is faster, laya-cybersec or open-jev-deberta-v3-large?
By their publishers’ figures, open-jev-deberta-v3-large answers in about 28 ms at the median and laya-cybersec in about 95.4 ms — measured on different hardware, so treat it as a rough guide.
Can I run laya-cybersec or open-jev-deberta-v3-large locally?
Yes, both: systemone pull textcortex/laya-cybersec and systemone pull kotoba-labs/open-jev-deberta-v3-large download the weights.
—
Evaluation suite
TextCortex saved run: 279 single-window inputs, PyTorch batch one
Kotoba held-out test, seen question types (banking77, SST-5, BoolQ)
