PyRUA-Lean Coder v0.1-4B

Paper | Code | Docs | Project page

PyRUA-Lean Coder v0.1-4B is Qwen3.5-4B trained with GRPO in the PyRUA-Lean framework, in which a robot agent acts by writing Python cells against a robot object robo instead of making one tool call per step. It was trained on 45 LIBERO-PRO task instances only.

Results

Success rate (%) in the PyRUA-Lean framework, one episode per task instance:

RoboTwin 2.0 and both RoboCasa365 sets are benchmarks the model was not trained on at all.

Usage

The model is served by any OpenAI-compatible server and driven by PyRUA-Lean's openai runtime. With SGLang:

python -m sglang.launch_server --model-path DAGroup-PKU/PyRUA-Lean-Coder-v0.1-4B --port 30000 \
    --context-length 65536 --tool-call-parser qwen3_coder --reasoning-parser qwen3 \
    --constrained-json-disable-any-whitespace

Then one LIBERO-PRO episode, on a task the model never saw in training (RPent's LIBERO-PRO stack installed as the Quick Start describes):

export RPENT_ROOT=/path/to/rpent OPENAI_API_KEY=none
pyrualean play --robot libero --arm cells --suite libero_goal_swap --task 3 --seed 0 \
    --runtime openai --base-url http://127.0.0.1:30000/v1 --api-key-env OPENAI_API_KEY \
    --model DAGroup-PKU/PyRUA-Lean-Coder-v0.1-4B \
    --openai-max-tokens 8192 --openai-tool-choice required --openai-temperature 1.0 \
    --openai-extra-body '{"top_p": 0.95, "top_k": 20}' \
    --images on-demand --feedback pure --guides --max-turns 40 --budget-s 5400 \
    --out runs/coder-demo --cuda-device 0
pyrualean summarize runs/coder-demo

These are the settings of the paper's evaluation. The model is trained for the PyRUA-Lean prompt and tools; outside them it behaves like a general 4B chat model.

Training

GRPO from Qwen3.5-4B in the PyRUA-Lean framework, on 45 LIBERO-PRO task instances; Section 5 of the paper describes it.

Citation

@article{si2026fewer,
  title={Fewer tokens, better action: GPT-6 Astra robot agents with 14\% higher success rate but 65\% fewer tokens},
  author={Si, Ruiyang and Bi, Jianxin and Yang, Shunyu and Ni, Rui and Huang, Wenbo and Wang, Qiang and Jiang, Shulong and Zhang, Yintianrun and Ruan, Feixiang and Zhang, Kewei and Wang, Duomin and Li, Xiuyu and Feng, Haiwen and Dong, Zhen and Zhou, Daquan},
  journal={arXiv preprint arXiv:2610.01939},
  year={2026}
}

The model inherits the Apache 2.0 license of Qwen3.5-4B.