APIUp to 25% cheaper than official pricesTry the API →
Hermes
APIModelsCases
Back to News
newsPost TrainingReasoning

Perplexity and NVIDIA's JOLT Beats GRPO Using 6.5x Fewer Training Tokens

Perplexity and NVIDIA introduce JOLT, a self-distillation post-training method that matches GRPO accuracy on MATH using 6.5x fewer training tokens.

Perplexity and NVIDIA's JOLT Beats GRPO Using 6.5x Fewer Training Tokens
Source
Perplexity
Published
Author
AlphaSignal Newsroom
Read
1 min read

Perplexity and NVIDIA introduce JOLT, a self-distillation post-training method that matches GRPO accuracy on MATH using 6.5x fewer training tokens.

Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.

Read original report

Next reads

LoopRPT Teaches AI to Think Smarter With Every Recurrent Pass
AlphaSignal · paper

LoopRPT Teaches AI to Think Smarter With Every Recurrent Pass

Shanghai AI Lab's RePro Cuts Reasoning Tokens 47% While Boosting Math Scores
AlphaSignal · paper

Shanghai AI Lab's RePro Cuts Reasoning Tokens 47% While Boosting Math Scores

Proximal KL Beats GRPO by 2.58% When AI Training Data Goes Stale
AlphaSignal · paper

Proximal KL Beats GRPO by 2.58% When AI Training Data Goes Stale