nilarnabdebnath/qwen3-1.7b-gsm8k-sft
This model is a 2 billion parameter Qwen3-1.7B variant, fine-tuned by nilarnabdebnath specifically for mathematical reasoning tasks. It achieves 77.2% accuracy on the GSM8K benchmark, a significant improvement over its base model, by utilizing a chain-of-thought reasoning format. Optimized for grade school math problems, it also shows proficiency in competition-level math, particularly in algebra and prealgebra.
Loading preview...
Overview
This model, nilarnabdebnath/qwen3-1.7b-gsm8k-sft, is a specialized fine-tuned version of the 2 billion parameter Qwen3-1.7B base model. Its primary focus is enhancing mathematical reasoning capabilities, particularly for grade school math problems. The model was trained using a two-stage process on a combined dataset of GSM8K and MetaMathQA examples, totaling nearly 250,000 entries, and leverages a crucial <think>...</think> chain-of-thought reasoning format.
Key Capabilities & Performance
- Exceptional GSM8K Performance: Achieves 77.2% accuracy on the GSM8K benchmark, a substantial 57 percentage point increase over the base Qwen3-1.7B model.
- Mathematical Reasoning: Demonstrates strong performance on competition-level math problems (MATH-500 benchmark at 55.2% accuracy), with particular strengths in Algebra (71.8%) and Prealgebra (68.3%).
- Chain-of-Thought: Utilizes a specific chain-of-thought reasoning format, which is critical for its improved math problem-solving abilities.
- Efficient Training: Fine-tuned in approximately 7 hours on an NVIDIA H100 GPU.
When to Use This Model
This model is ideal for applications requiring robust mathematical problem-solving, especially those involving grade school and intermediate algebra-level word problems. Its fine-tuning on GSM8K makes it a strong candidate for educational tools, automated problem solvers, or any system where accurate, step-by-step mathematical reasoning is paramount. While optimized for GSM8K, it also shows good transferability to harder math problems.