[GRPO Explained] DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
Yannic Kilcher · 69:00
This video walks through the April 2024 DeepSeekMath paper. It argues that a 7B model reached GPT-4/Gemini-Ultra–level math benchmark scores for two reasons: an iteratively mined 120B-token math corpus from Common Cra...