Downloads · 30 days
114
10% of all-time downloads
code-critic-model/Qwen3-4B-Critic-SFT-Qwen-only
Qwen3-4B-Critic-SFT-Qwen-only is a text generation model from code-critic-model. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
A 4B critic from Steer, Don't Solve: Training Small Critic Models for Large Code Agents, trained on critiques of Qwen3-Next-80B-A3B trajectories only. It is one arm of the training-corpus ablation (Table 3). The 4B cr…
Downloads · 30 days
114
10% of all-time downloads
All-time downloads
1.2K
Public
Parameters
4.4B
17.7 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors17.6 GB · 100%
From the Hugging Face model README
A 4B critic from Steer, Don't Solve: Training Small Critic Models for Large Code Agents, trained on critiques of Qwen3-Next-80B-A3B trajectories only. It is one arm of the training-corpus ablation (Table 3). The 4B critic trained on the mixed CWM plus Qwen3-Next corpus is Qwen3-4B-Critic-SFT.
The critic reads a coding agent's trajectory every k steps and returns a structured critique: detected error categories, evidence, a recovery action, task status, and one line of overall guidance. It does not write the patch.
All released models and datasets are listed on the organization page. Code and configs are in the critic-training repository.
| Paper location | Row label |
|---|---|
| Table 3, corpus ablation | 4B, Qwen-only |
Original run name: qwen3-4b-instruct-2507-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only-opus-distill-32k-multiturn.
code-critic-model/critic-sft-qwen-only, 1,915 examples.
Identical to Qwen3-4B-Critic-SFT apart from the data. Full-parameter SFT with LLaMA-Factory, config finetuning/qwen3_4b_critic_full_sft_l40s_train_multiturn_resumable.yaml.
| Setting | Value |
|---|---|
| Base model | Qwen/Qwen3-4B-Instruct-2507 |
| 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 | bf16 |
Resolve rate on SWE-bench Verified, from Table 3 of the paper.
| Coding agent | No critic | + this critic | + Qwen3-4B-Critic-SFT (CWM + Qwen data) |
|---|---|---|---|
| Qwen3-Next-80B-A3B | 20.0 | 24.6 | 24.2 |
| Qwen3-32B | 8.8 | 10.8 | 11.4 |
Same serving and launch procedure as Qwen3-4B-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.
@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}
}