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.
Contrastive Language Models score a state against a set of candidate actions with a contrastive objective. Two projection heads on frozen Qwen3-8B embeddings, trained with InfoNCE; clm-serve maps Choice, Score and Noul onto candidate ranking.
Decides
choice, noul, score
choice, score, noul, rank, classify, route
Architecture
canopy-jev
clm
Fine-tuned from
qwen/qwen3.8-27b
qwen/qwen3-8b
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.4%
—
Calibration error
—
—
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 clm?
canopy-jev is from Ruoxi Qiu and clm from Contrastive-LM. Both have open weights you can download and run. Both answer choice, noul and score questions. Only clm answers rank, classify and route. clm is the smaller model, at 8.0B parameters to 27B.
Which is more accurate, canopy-jev or clm?
Only canopy-jev publishes an accuracy figure (87.4% on JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer); clm does not, so there is no comparison to make without your own test.
Which is cheaper, canopy-jev or clm?
canopy-jev: Free (open weights). clm: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run canopy-jev or clm locally?
Yes, both: systemone pull camellia86/canopy-jev and systemone pull contrastive-lm/clm download the weights.
—
Evaluation suite
JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer
