Perplexity and NVIDIA's JOLT matches GRPO accuracy on MATH with 6.5x fewer training tokens
Perplexity and NVIDIA researchers introduced JOLT, a post-training method that trains a single model as both teacher and student. According to the report, it matched GRPO's accuracy on MATH while processing 6.5x fewer training tokens.
Perplexity and NVIDIA researchers have introduced JOLT, a post-training method that trains one model as both teacher and student. On MATH, JOLT matched the accuracy of GRPO while processing 6.5x fewer training tokens, according to a MarkTechPost report.
What was announced
JOLT is described as a joint teacher-student reinforcement learning method. Rather than using separate models for the teacher and student roles, it trains a single model to fill both.
Reported result
- Benchmark: MATH
- Baseline: GRPO
- Outcome: JOLT matched GRPO's accuracy
- Efficiency: 6.5x fewer processed training tokens
What this means
The reported result concerns training efficiency: reaching GRPO-level accuracy on MATH while processing 6.5x fewer tokens. It also points to a design in which one model handles both the teacher and student roles during post-training. The comparison reported is against GRPO on MATH.
Related Articles
Microsoft's ThinkingBox-Training lifts Qwen3.8-27B fully reliable task completions from 38 to 107 of 507
Microsoft and Hugging Face released ThinkingBox-Training, which turns executable database-state checks into RL rewards. After 50 GRPO steps, Qwen3.8-27B's pass@1 rose from 51.70% to 60.78%, and tasks passed on all 20 attempts rose from 38 to 107, according to the authors.
NVIDIA and AI4Bharat announce Nemotron-pretraining-Indic-v1: 2.8T tokens across 22 Indian languages
NVIDIA and IIT Madras' AI4Bharat lab announced Nemotron-pretraining-Indic-v1, a corpus of roughly 2.8 trillion tokens covering all 22 languages in India's Constitution. It is due for release soon, alongside the open GPU-accelerated pipeline used to generate it.
Anthropic: Claude found 493 attack-surface issues across 25 open-source projects, led by OpenSSH fixes
Anthropic published early results on using Claude to map and reduce software attack surface. Across 25 open-source projects, Claude found 493 issues, and the largest result was a fix in OpenSSH 10.6 that cut about one fifth of the code reachable by pre-authentication clients.
Anthropic Says Claude Proved Security Properties for 239 Entry Points Across 123 Libraries
Anthropic published research on October 22, 2026, reporting that Claude produced machine-checked formal proofs of security properties at the RISC-V assembly level. The proofs cover 239 fuzz-test entry points across 123 open-source libraries, in a collaboration with Theorem.
Comments
Loading...