cublya/Qwen3-Code-Reasoning-4B
cublya/Qwen3-Code-Reasoning-4B is a 4 billion parameter causal language model, LoRA-finetuned from Qwen3-4B-Thinking-2507 by GetSoloTech. It is specifically optimized for competitive programming and complex code reasoning tasks, trained on a high-quality dataset of programming problems with solutions achieving at least 50% test case pass rates. This model excels at generating well-reasoned programming solutions and understanding problem constraints, with a context length of 4096 tokens, configurable up to 262,144.
Loading preview...
Model Overview
cublya/Qwen3-Code-Reasoning-4B is a 4 billion parameter model, LoRA-finetuned by GetSoloTech from the Qwen3-4B-Thinking-2507 base model. It is specifically designed to enhance performance in competitive programming and code reasoning challenges. The model leverages the advanced 'thinking capabilities' inherited from its base and is trained using Unsloth with QLoRA for efficient adaptation.
Key Capabilities
- Enhanced Code Reasoning: Optimized for solving complex programming problems with detailed reasoning.
- High-Quality Solutions: Trained on a curated dataset of competitive programming problems (from TACO, APPS, CodeContests, Codeforces) where solutions achieved a pass rate of at least 50% on test cases.
- Structured Output: Generates well-reasoned and comprehensive programming solutions, including proper edge case handling.
- Efficient Training: Utilizes LoRA adapters for efficient fine-tuning, making it a specialized tool for code-centric tasks.
Good For
- Competitive Programming: Excels at understanding problem constraints and generating accurate solutions.
- Code Generation: Produces more efficient and correct code.
- Reasoning Tasks: Provides enhanced step-by-step reasoning for intricate coding problems.