JevAny Qwen3.8-27B LoRA
SimpleJev's LoRA adapters and pointer head that turn an open model into a typed-decision model: given a state, a question and the options, it returns a probability for every option without generating text. This entry is the Qwen3.8-27B release; the others are linked.
Adapter only (rank 8 plus a pointer head); users need Qwen3.8-27B themselves. The pointer readout takes very many options; the direct-token variant is capped at 255. Trained on 2.18M labelled decisions, building on Kev's code (Apache-2.0). Calibration figures are after temperature scaling on a separate partition. On SimpleJev's own cross-domain Transfer suite the 27B scores 86.04% (ECE 0.026); its own run on JevBench's 231 public items gives 90.04%, not a leaderboard result. Siblings on Qwen3.5-4B (pointer and direct-token), Gemma 4B and Muse-Glimmer-30B. Not related to Featherless AI's simple-jev server, nor to Jev or TypeSafe.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one probability
At a glance
| Parameters | 27B |
| Base model | Qwen/Qwen3.8-27B |
| Maker | SimpleJev |
| Released | 2026-09-29 |
| License | apache-2.0 |
| Reported accuracy | 86.0% |
Get the weights
pip install systemonemodels
systemone pull simplejev/jevany
The files are served from the maker's Hugging Face repository, SimpleJev/JevAny-Qwen3.8-27B-LoRA, and verified against the checksums recorded here.