Perplexity and NVIDIA researchers have introduced JOLT, a new post-training method for language models that assigns a single model two roles at once: teacher and student. In this setup, the teacher has access to privileged information while the student does not. The main change is that the teacher is also trained, which the researchers say keeps its feedback usable during the process.
According to MarkTechPost, JOLT matched the accuracy of GRPO on the MATH benchmark while using 6.5x fewer training tokens. The team also presented JOLT+, a version that adds rewards based on the student’s own outcomes. In the reported tests, JOLT+ posted the best average score across all settings.
On AppWorld, JOLT+ reached a pass rate of 72.09%, compared with 62.20% for GRPO, a gap of 9.9 percentage points. The method has been validated on multiple models, including Qwen3 and DeepSeek-V2-Lite.
Perplexity and NVIDIA researchers have released JOLT, a new post-training method for language models. The approach lets a single model act as both teacher and student, with the teacher receiving privileged information that is not available to the student.
The key change in JOLT is that the teacher model is also trained, which is intended to keep its feedback usable.
6.5x fewer training tokens on MATH at the same accuracy as GRPO
According to MarkTechPost, JOLT reached the same accuracy as GRPO on the MATH benchmark while reducing the number of training tokens processed by 6.5x.
JOLT+ posted the top average score across test settings
The researchers also introduced JOLT+, a version that adds rewards tied to the student’s own results. In the reported experiments, JOLT+ achieved the highest average score across all test settings.
On AppWorld, JOLT+ recorded a 72.09% pass rate, versus 62.20% for GRPO, a lead of 9.9 percentage points.
Validated on multiple models
The method has been validated on several models, including Qwen3 and DeepSeek-V2-Lite.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
