Perplexity and NVIDIA Unveil JOLT to Boost AI Training Efficiency
Researchers from Perplexity and NVIDIA have introduced JOLT, an innovative reinforcement learning method where a single model acts as both teacher and student. The approach achieved matching accuracy on the MATH benchmark while utilizing 6.5 times fewer training tokens, significantly reducing the computational resources typically required for advanced artificial intelligence model post-training.
Source: MarkTechPost