# Ruoxi Qiu: canopy-jev

> 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.

- Page: https://systemonemodels.ai/camellia86/canopy-jev
- API: https://api.systemonemodels.ai/v1/models/camellia86/canopy-jev
- Download: `pip install systemonemodels && systemone pull camellia86/canopy-jev`

## Facts

| | |
|---|---|
| Maker | Ruoxi Qiu (https://systemonemodels.ai/camellia86) |
| Decides | choice, noul, score |
| Architecture | canopy-jev |
| Base model | qwen/qwen3.8-27b |
| Parameters | 27B |
| Licence | apache-2.0 |
| Availability | Open weights |
| Released | 2026-10-02 |
| Latest version | 0.1.0 |

## Reported evaluation

Suite: JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer. Numbers are the publisher's own.

- Decision accuracy: 87.4%

## Model card

<!-- generated by scripts/seed_catalog.py; edit content/models/catalog.yaml -->

# Canopy-Jev-27B

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 rank-16 LoRA on independent copies of the last four text layers plus a ten-letter numerical prior (25 MB of artifacts). The base downloads separately at a pinned revision; it needs the maker's canopy_jev loader, Linux, CUDA and at least 64 GiB of GPU memory. Choice takes 1 to 10 candidates; noul and score are adaptations of that readout. Text only. The maker's own run on JevBench's public items (a third-party benchmark) gives 202/231; the release checkpoint was picked after inspecting several endpoints, so it is not blind, and training used repurposed MMLU and SuperGPQA material. Not calibrated for applications. Weights Apache-2.0, code MIT. Not affiliated with TypeSafe.

## What it decides

- **choice** — picks one option from a set
- **noul** — answers a yes/no question with one probability
- **score** — places the input on an ordered scale

## At a glance

| | |
|---|---|
| Parameters | 27B |
| Base model | `Qwen/Qwen3.8-27B` |
| Maker | Ruoxi Qiu |
| Released | 2026-10-02 |
| License | apache-2.0 |
| Reported accuracy | 87.4% |

## Get the weights

```bash
pip install systemonemodels
systemone pull camellia86/canopy-jev
```

The files are served from the maker's Hugging Face repository, [`Camellia86/Canopy-Jev-27B`](https://huggingface.co/Camellia86/Canopy-Jev-27B), and verified against the checksums recorded here.

## Read more

- [Model card](https://huggingface.co/Camellia86/Canopy-Jev-27B)
- [Code](https://github.com/Camellia86/Canopy-Jev)

---

*This page was opened by System One for Ruoxi Qiu, who can claim the organisation and take it over at any time.*

---

From System One Models — https://systemonemodels.ai/ · every System One model: https://systemonemodels.ai/system-one-models