code-critic-model/qwen3-4b-sft-prm

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

code-critic-model/qwen3-4b-sft-prm is a 4 billion parameter Qwen3-based language model, fine-tuned as a critic model. It is specifically designed for evaluating and providing feedback on code, leveraging the critic-sft-cwm-qwen dataset. This model's primary differentiation lies in its specialization as a code critic, intended to steer rather than solve coding problems. It is optimized for code evaluation tasks within a 32768 token context window.

Loading preview...

Overview

code-critic-model/qwen3-4b-sft-prm is a 4 billion parameter language model built upon the Qwen3-4B-Instruct-2507 architecture. This model has been specifically fine-tuned to function as a critic model, rather than a generative or problem-solving model. Its development is rooted in the "Steer, Don't Solve" paradigm, indicating its purpose is to provide evaluative feedback and guidance, particularly in coding contexts.

Key Capabilities

  • Code Criticism: Specialized in evaluating code and providing constructive feedback.
  • Feedback Generation: Designed to generate critical assessments and suggestions for code improvements.
  • Qwen3 Foundation: Leverages the robust capabilities of the Qwen3 base model.
  • Context Length: Supports a substantial context window of 32768 tokens, suitable for analyzing larger code snippets or discussions.

Training Details

The model was fine-tuned on the critic-sft-cwm-qwen dataset. This dataset is crucial to its specialization, enabling it to understand and apply critical reasoning to code. The weights of this repository are byte-identical to Qwen3-4B-Critic-SFT, which is the primary reference for this critic model.

Good For

  • Automated Code Review: Assisting developers by providing automated feedback on code quality, potential issues, or adherence to best practices.
  • Educational Tools: Integrating into platforms that teach programming, offering students immediate critiques on their code.
  • Research in Code Evaluation: Serving as a base for further research into AI-driven code assessment and feedback mechanisms.

Finetunes
1 models