Model Overview

GetSoloTech/Qwen3-Code-Reasoning-4B is a 4 billion parameter causal language model, specifically LoRA-finetuned from the Qwen3-4B-Thinking-2507 base model. Developed by GetSoloTech, this model is engineered to excel in competitive programming and complex code reasoning tasks. It utilizes a 4096-token context length, with configurability up to 262,144 tokens, and inherits advanced "thinking capabilities" from its base.

Key Capabilities

• Enhanced Code Reasoning: Trained on the GetSoloTech/Code-Reasoning dataset, which comprises high-quality competitive programming problems from sources like TACO, APPS, CodeContests, and Codeforces.
• High-Quality Solutions: Focuses on generating solutions that previously achieved at least a 50% test case pass rate, ensuring practical effectiveness.
• Structured Output: Optimized for producing well-reasoned and comprehensive programming solutions.
• Efficient Training: Utilizes LoRA (Low-Rank Adaptation) with Unsloth and QLoRA for efficient finetuning.

Good For

• Competitive Programming: Understanding problem constraints and generating accurate, efficient solutions.
• Code Generation: Producing high-quality code with improved accuracy.
• Reasoning Tasks: Providing enhanced step-by-step reasoning for intricate coding challenges.
• Solution Completeness: Handling edge cases and delivering more comprehensive solutions.