cublya/GPT-OSS-Code-Reasoning-20B
cublya/GPT-OSS-Code-Reasoning-20B is a 20 billion parameter language model, fine-tuned from openai/gpt-oss-20b, specifically optimized for competitive programming and algorithmic reasoning tasks. It excels at generating Python and C++ solutions and explanations for complex coding problems. Trained on the nvidia/OpenCodeReasoning-2 dataset, this model is designed to provide efficient and correct code solutions within a 32768 token context length.
Loading preview...
Model Overview
cublya/GPT-OSS-Code-Reasoning-20B is a 20 billion parameter model, built upon openai/gpt-oss-20b, and specialized through supervised fine-tuning for competitive programming and algorithmic reasoning. It leverages the nvidia/OpenCodeReasoning-2 dataset, which combines Python and C++ problem-solution pairs, to enhance its code generation and explanation capabilities. The model supports a context length of 32768 tokens and was trained using LoRA SFT via TRL SFTTrainer.
Key Capabilities
- Competitive Programming Solutions: Generates Python and C++ solutions for algorithmic problems.
- Algorithmic Reasoning: Provides reasoning alongside code solutions, which can be adjusted for 'low', 'medium', or 'high' effort.
- Chat Format Optimization: Best performance is achieved using a specific chat template, including a system prompt for an "expert competitive programmer."
Intended Use Cases
- Generating Code: Ideal for producing correct and efficient Python/C++ code for competitive programming challenges.
- Problem Solving Assistance: Can be used to understand and solve complex algorithmic tasks by generating solutions and explanations.
Limitations
- Not for Safety-Critical Applications: The model is not intended for use in safety-critical systems.
- Potential for Hallucinations: May occasionally produce incorrect or inefficient code, requiring user verification.