DeepMath: A lightweight math reasoning Agent with SmolAgents

DeepMath is a math reasoning agent built on the Qwen3-4B Thinking model and fine-tuned using Group Relative Policy Optimization (GRPO). Instead of verbose text, the model generates small Python snippets for intermediate calculations, which are run in a secure sandbox. The results are then integrated back into the model's reasoning process to solve mathematical problems. This approach significantly reduces output length and arithmetic errors, leading to improved accuracy on various math benchmarks.