← All models

DeepSeekMath 7B RL

DeepSeek · 2024-02-05 · 6.9B parameters

DeepSeekMath 7B RL is DeepSeekMath 7B Instruct further trained with Group Relative Policy Optimization (GRPO), the reinforcement-learning method the DeepSeekMath paper introduced, on about 144K GSM8K- and MATH-related questions. Released February 5, 2024, it has 6,910,365,696 parameters and a 4,096-token context.

Benchmark scores