Sepolian/qwen2.5-0.5B-math
Sepolian/qwen2.5-0.5B-math is a fine-tuned Qwen2.5-0.5B model developed by Sepolian, specifically optimized for mathematical reasoning and problem-solving tasks. This 0.5 billion parameter model demonstrates improved performance on benchmarks like GSM8K and TheoremQA compared to its base version. It is designed for applications requiring accurate arithmetic, word problem reasoning, and competition-level math capabilities.
Loading preview...
Sepolian/qwen2.5-0.5B-math: Math-Optimized Qwen2.5 Fine-tune
This model is a specialized fine-tuned version of the Qwen/Qwen2.5-0.5B base model, developed by Sepolian, with a strong focus on enhancing mathematical reasoning and problem-solving abilities. It has been trained on a diverse, small subset of multiple math datasets totaling approximately 25,000 examples.
Key Capabilities & Performance
The fine-tuning significantly improves the model's performance on various mathematical benchmarks:
- GSM8K: Achieves 30.93% strict-match accuracy, a substantial increase from the base model's 0.00%.
- TheoremQA: Shows an improved score of 14.12% compared to the base model's 10%.
- MATH-500: While exact match decreased slightly, the
math_verifyscore remained consistent at 17.60%.
Training Details
The model was trained using a mixed dataset approach, incorporating several high-quality math datasets:
openai/gsm8k: For foundational arithmetic and word problem reasoning.AI-MO/NuminaMath-CoT: To cover competition-level math problems.TIGER-Lab/MathInstruct(CoT-only): For diverse math reasoning and theorem-style supervision.hendrycks/competition_math(L3-L5): Targeting higher-difficulty competition math.TIGER-Lab/TheoremQA-aligned slice: Providing exposure to basic theorem application.
Ideal Use Cases
This model is particularly well-suited for applications requiring robust mathematical capabilities, including:
- Solving arithmetic and word problems.
- Assisting with competition-level mathematics.
- Tasks involving theorem application and mathematical reasoning.