Native MLX FP16 conversion of Laya for Apple silicon. The same ModernBERT-large encoder, decision transformer and heads, running in MLX with no PyTorch; the checkpoint Laya Studio fine-tunes from.
One of the earliest Jev-shaped scorers. A sequence-classification head on Qwen3.5-4B-Base scores each (state, question, option) triple and softmaxes per question. Non-commercial licence.
Decides
choice, score, noul
choice, score, noul, classify
Architecture
laya
scorer
Fine-tuned from
convai-innovations/laya
qwen/qwen3.5-4b-base
License
apache-2.0
cc-by-nc-4.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
—
70.7%
Calibration error
—
0.044
Valid action rate
—
—
Median latency
—
112 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-mlx and system-one-qwen3-5-4b-scorer?
laya-mlx is from aac6fef and system-one-qwen3-5-4b-scorer from pngwn. Both have open weights you can download and run. Both answer choice, score and noul questions. Only system-one-qwen3-5-4b-scorer answers classify. laya-mlx is the smaller model, at 421M parameters to 4.0B. laya-mlx is licensed apache-2.0; system-one-qwen3-5-4b-scorer, cc-by-nc-4.0.
Which is more accurate, laya-mlx or system-one-qwen3-5-4b-scorer?
Only system-one-qwen3-5-4b-scorer publishes an accuracy figure (70.7% on author-reported); laya-mlx does not, so there is no comparison to make without your own test.
Which is cheaper, laya-mlx or system-one-qwen3-5-4b-scorer?
laya-mlx: Free (open weights). system-one-qwen3-5-4b-scorer: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run laya-mlx or system-one-qwen3-5-4b-scorer locally?
Yes, both: systemone pull aac6fef/laya-mlx and systemone pull pngwn/system-one-qwen3-5-4b-scorer download the weights.
