Ruoxi Qiu's decision adapter for a frozen Qwen3.8-27B: a 6.3M-parameter LoRA on the last four text layers plus a numerical prior. One shared state feeds isolated question branches; each Choice, Noul or Score question returns probabilities with no generated tokens.
Small open typed-decision model with calibrated probabilities, first-class Nepali coverage.
Decides
choice, noul, score
choice, score, noul, classify, route
Architecture
canopy-jev
mira
Fine-tuned from
qwen/qwen3.8-27b
jhu-clsp/mmbert-small
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.4%
61.3%
Calibration error
—
0.083
Valid action rate
—
100.0%
Median latency
—
30 ms
p95 latency
—
43 ms
Evaluation suite
JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer
s1-decision-bench
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 canopy-jev and mira?
canopy-jev is from Ruoxi Qiu and mira from SAGEA. Both have open weights you can download and run. Both answer choice, noul and score questions. Only mira answers classify and route.
Which is more accurate, canopy-jev or mira?
They report on different suites — canopy-jev 87.4% on JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer, mira 61.3% on s1-decision-bench — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, canopy-jev or mira?
canopy-jev: Free (open weights). mira: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run canopy-jev or mira locally?
Yes, both: systemone pull camellia86/canopy-jev and systemone pull sagea/mira download the weights.
