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.
A 364M Hebrew decision encoder on NeoDictaBERT: give it a state, a question and 2 to 7 free-text options and it returns a probability for each option in one forward pass, about 70 ms on a laptop CPU. Hebrew only.
Decides
choice, noul, score
choice, classify, route
Architecture
canopy-jev
nativ
Fine-tuned from
qwen/qwen3.8-27b
dicta-il/neodictabert
License
apache-2.0
cc-by-4.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.4%
96.2%
Calibration error
—
0.010
Valid action rate
—
—
Median latency
—
—
p95 latency
—
—
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 nativ-he-decision?
canopy-jev is from Ruoxi Qiu and nativ-he-decision from Yoav Pinto. Both have open weights you can download and run. Both answer choice questions. Only canopy-jev answers noul and score. Only nativ-he-decision answers classify and route. nativ-he-decision is the smaller model, at 364M parameters to 27B. canopy-jev is licensed apache-2.0; nativ-he-decision, cc-by-4.0.
Which is more accurate, canopy-jev or nativ-he-decision?
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, nativ-he-decision 96.2% on nativ-bench user-intent task, 4 options (MASSIVE Hebrew, 2,973 items; the maker's own suite) — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, canopy-jev or nativ-he-decision?
canopy-jev: Free (open weights). nativ-he-decision: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run canopy-jev or nativ-he-decision locally?
Yes, both: systemone pull camellia86/canopy-jev and systemone pull yoav-pinto/nativ-he-decision download the weights.
Evaluation suite
JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer
nativ-bench user-intent task, 4 options (MASSIVE Hebrew, 2,973 items; the maker's own suite)
