# HIT-TMG (Lychee Team): jevembed

> An embedding model turned decision model. Qwen3-Embedding-4B with a merged LoRA, driven by the open JevEmbed toolkit, which scores Choice, Score and Noul questions from embedding similarity and returns a distribution without generating text.

- Page: https://systemonemodels.tech/hit-tmg/jevembed
- API: https://api.systemonemodels.tech/v1/models/hit-tmg/jevembed
- Download: `pip install systemonemodels && systemone pull hit-tmg/jevembed`

## Facts

| | |
|---|---|
| Maker | HIT-TMG (Lychee Team) (https://systemonemodels.tech/hit-tmg) |
| Decides | choice, score, noul, classify |
| Architecture | jevembed |
| Base model | qwen/qwen3-embedding-4b |
| Parameters | 4.0B |
| Context | 1K tokens |
| Licence | apache-2.0 |
| Availability | Open weights |
| Released | 2026-09-28 |
| Latest version | 2026.09 |

## Reported evaluation

Suite: JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging. Numbers are the publisher's own.

- Decision accuracy: 85.9%

## Model card

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

# JevEmbed-Qwen3-Embedding-4B

An embedding model turned decision model. Qwen3-Embedding-4B with a merged LoRA, driven by the open JevEmbed toolkit, which scores Choice, Score and Noul questions from embedding similarity and returns a distribution without generating text.

Trained for one epoch on the 1.6M-question JevEmbed-Data training split (LoRA rank 64, 16 GPUs). On that dataset's 66,482-question test split the final checkpoint reaches 85.9% overall hard-label accuracy (Choice 90.3%, Score level 73.4%, Noul 95.9%), against 36.3% for the untuned base. Inputs are truncated at 1,024 tokens. No JevBench result is published. Siblings on the same recipe: JevEmbed-Qwen3-Embedding-0.6B and JevEmbed-KaLM-Embedding-V2.5. The toolkit (Apache-2.0) also covers data synthesis, fine-tuning, a benchmark module and a playground. Training-data source licences vary; see the dataset's processing report.

## 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 calibrated probability
- **classify** — assigns a category from a fixed taxonomy

## At a glance

| | |
|---|---|
| Parameters | 4B |
| Base model | `qwen/qwen3-embedding-4b` |
| Maker | HIT-TMG (Lychee Team) |
| Released | 2026-09-28 |
| License | apache-2.0 |
| Reported accuracy | 85.9% |

## Get the weights

```bash
pip install systemonemodels
systemone pull hit-tmg/jevembed
```

The files are served from the maker's Hugging Face repository, [`HIT-TMG/JevEmbed-Qwen3-Embedding-4B`](https://huggingface.co/HIT-TMG/JevEmbed-Qwen3-Embedding-4B), and verified against the checksums recorded here.

## Read more

- [Model card](https://huggingface.co/HIT-TMG/JevEmbed-Qwen3-Embedding-4B)
- [JevEmbed toolkit](https://github.com/HITsz-TMG/JevEmbed)
- [JevEmbed-Data](https://huggingface.co/datasets/HIT-TMG/JevEmbed-Data)
- [JevEmbed-Qwen3-Embedding-0.6B](https://huggingface.co/HIT-TMG/JevEmbed-Qwen3-Embedding-0.6B)

---

*This page was opened by System One for HIT-TMG (Lychee Team), who can claim the organisation and take it over at any time.*

---

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