Understand this story
Aravind Srinivas shared JOLT, a post-training method proposed by Perplexity in collaboration with NVIDIA. The method lets the same model serve as both teacher and student, and trains the teacher to learn from rewards from its own attempts while penalizing drift away from the student's behavior. On the MATH benchmark, JOLT reaches accuracy comparable to GRPO with 6.5 times fewer training tokens; JOLT+ additionally provides outcome-based rewards to the student model and clearly surpasses GRPO on math, coding, terminal tasks, and tool-use benchmarks. Currently this research is being shared publicly by Aravind Srinivas, with no further follow-up progress yet.
AI-generated from public reports
Latest development · Report
Perplexity and NVIDIA propose JOLT self-distillation training methodDevelopments
1 confirmed developments
Aravind Srinivas shares JOLT post-training research in collaboration with NVIDIA
Source:Aravind Srinivas
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 22
Aravind Srinivas
Perplexity and NVIDIA propose JOLT self-distillation training method
Perplexity and NVIDIA collaboratively propose a post-training method, JOLT: it lets the same model serve as both teacher and student, and trains the teacher to learn from the rewards of its own attempts while penalizing drift away from the student's behavior. On the MATH benchmark, JOLT achieves accuracy comparable to GRPO with 6.5 times fewer training tokens; JOLT+ additionally provides the student model with outcome-based rewards, clearly surpassing GRPO on math, coding, terminal tasks, and tool-use benchmarks.