code-critic-model

Qwen3-8B-Critic-SFT-CWM-only

README

License: apache-2.0

Where it appears in the paper

Table with columns: Paper location, Row label
Paper location	Row label
Table 3, corpus ablation	`8B, CWM-only`

Original run name: qwen3-8b-full-sft-prm-r2egym-swebench-instructions-k5-opus-distill-32k-lr5e6-multiturn.

Training data

code-critic-model/critic-sft-cwm-only, 4,532 examples.

Tasks: 500 R2E-Gym instances from matplotlib, moto, and sympy, disjoint from SWE-bench Verified.
Agent that produced the trajectories: CWM-32B, 500 trajectories.
Teacher: Claude Opus 4.6, queried every 5 agent steps with the high-level prompt.

This corpus is the CWM half of critic-sft-cwm-qwen.

Training setup

Identical to Qwen3-8B-Critic-SFT apart from the data. Full-parameter SFT with LLaMA-Factory, config finetuning/qwen3_8b_critic_full_sft_l40s_train_multiturn_resumable.yaml.

Table with columns: Setting, Value
Setting	Value
Base model	Qwen/Qwen3-8B
Chat template	`qwen3_nothink`
Sequence length	32,768 tokens
Loss	final critique turn only (`mask_history: true`)
Hardware	8 x L40S, effective batch 8
Optimizer	AdamW, lr 5e-6, cosine, warmup ratio 0.1
Epochs	3
Precision

Results

Resolve rate on SWE-bench Verified, from Table 3 of the paper.

Table with columns: Coding agent, No critic, + this critic, + Qwen3-8B-Critic-SFT (CWM + Qwen data)
Coding agent	No critic	+ this critic	+ Qwen3-8B-Critic-SFT (CWM + Qwen data)
Qwen3-Next-80B-A3B	20.0	23.6	25.2
Qwen3-32B	8.8	11.4	13.8

The critic trained only on CWM trajectories still helps both Qwen agents, which is the transfer result the ablation is after. Adding Qwen3-Next trajectories to the corpus helps more.

How to use

Same serving and launch procedure as Qwen3-8B-Critic-SFT: serve with vLLM in bf16 and pass the served name to scripts/run_critic_max150.sh with --prm. The served name must have an entry in mini-swe-agent/configs/litellm_model_registry.json; add one for this model if you use a new name.

Citation

bibtex
@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}
}

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

code-critic-model

Model Tree

Base

Qwen/Qwen3-8B

Fine-tuned

this model

Input Modalities

Text

Output Modalities