Native MLX FP16 conversion of Laya's multilingual checkpoint (mmBERT) for Apple silicon, with a 1,024-token context. The base for multilingual fine-tunes in Laya Studio.
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
choice, score, noul
choice, score, noul, classify
Architecture
laya
open-jev
Fine-tuned from
convaiinnovations/laya-multilingual
microsoft/deberta-v3-large
License
apache-2.0
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
—
28 ms
p95 latency
—
—
Evaluation suite
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-multilingual-mlx and open-jev-deberta-v3-large?
laya-multilingual-mlx is from aac6fef and open-jev-deberta-v3-large from Kotoba Labs. Both have open weights you can download and run. Both answer choice, score and noul questions. Only open-jev-deberta-v3-large answers classify. laya-multilingual-mlx reads up to 1K tokens of state, against 512 tokens for open-jev-deberta-v3-large. laya-multilingual-mlx is the smaller model, at 322M parameters to 434M.
Which is more accurate, laya-multilingual-mlx 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-multilingual-mlx does not, so there is no comparison to make without your own test.
Which is cheaper, laya-multilingual-mlx or open-jev-deberta-v3-large?
laya-multilingual-mlx: Free (open weights). open-jev-deberta-v3-large: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run laya-multilingual-mlx or open-jev-deberta-v3-large locally?
Yes, both: systemone pull aac6fef/laya-multilingual-mlx and systemone pull kotoba-labs/open-jev-deberta-v3-large download the weights.
—
Kotoba held-out test, seen question types (banking77, SST-5, BoolQ)
