code-critic-model/critic-sft-qwen-only

critic-sft-qwen-only

The Qwen-only SFT corpus from Steer, Don't Solve: Training Small Critic Models for Large Code Agents. It trains Qwen3-8B-Critic-SFT-Qwen-only and Qwen3-4B-Critic-SFT-Qwen-only, the Qwen-only arms of the corpus ablation in Table 3.

Each record is one critique point: a Qwen3-Next-80B-A3B-Instruct trajectory up to some step, followed by the critique that Claude Opus 4.6 wrote for it under the paper's high-level prompt. The main corpus, critic-sft-cwm-qwen, is this set plus 4,532 records from CWM-32B trajectories.

The set is smaller than the CWM one because Qwen3-Next finishes tasks in fewer steps, so fewer critique points are collected per trajectory. The tasks are the same 500 used for CWM, minus 17 on which the Qwen3-Next run produced no usable trajectory.

Format

One JSON object per line with a single field, messages, in ShareGPT layout: a system turn with the critic instructions, alternating user / assistant turns carrying the agent's trajectory (task and command output on the user side, agent reasoning and bash command on the assistant side), and a final assistant turn holding the teacher critique. Critiques the agent received earlier in the trajectory appear inside the user turn that follows them. The full description of the format, with an example critique, is on the critic-sft-cwm-qwen card.

Files

Using it

from datasets import load_dataset
ds = load_dataset("code-critic-model/critic-sft-qwen-only", split="train")

For LLaMA-Factory, point dataset_dir at a download of this repository and use dataset: prm_sft_train with the configs in finetuning/ of the critic-training repository.

Citation

@misc{gandhi2026steerdontsolvetraining,
  title={Steer, Don't Solve: Training Small Critic Models for Large Code Agents},
  author={Shubham Gandhi and Yiqing Xie and Atharva Naik and Ruichen Zhu and Carolyn Rose},
  year={2026},
  eprint={2606.21811},
  archivePrefix={arXiv},
  primaryClass={cs.SE},
  url={https://arxiv.org/abs/2606.21811}
}