---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
base_model: Qwen/Qwen3-4B
tags:
  - multi-agent
  - routing
---

# EvoRouter

Model weights for [EvoRouter](https://github.com/CenYangtze/EvoRouter). | [Paper (arXiv)](https://arxiv.org/abs/2610.35690)

Set up the evaluation environment, API endpoint, and datasets using the [repository instructions](https://github.com/CenYangtze/EvoRouter#installation).

Download the model into the evaluation directory:

```bash
python -m evorouter download --model-repo CyanOvO/EvoRouter
python -m evorouter evaluate --methods evorouter
```

To use a local copy:

```bash
python -m evorouter evaluate --model-path /path/to/EvoRouter
```

Run BBH logical deduction and LiveCodeBench with `--suite ood`, or run both evaluation suites with `--suite all`.

The model files include the tokenizer, chat template, and generation configuration. Keep the files together when copying the model directory.

[Apache 2.0](LICENSE).