LLM Digest

Story

arxiv_cs_ai · Oct 23, 2026 · paper

Source brief

NgramRL: Accelerating RL Post-Training for Structured Generation via Rollout-Aware N-gram Speculative Decoding

arxiv.orgOct 23, 2026
original source linked

In brief

Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but its cost is dominated by autoregressive rollout generation, which can exceed 70% of training time. Existing s...

Read the original at arxiv.org →Open in live feed Read that day’s brief

NgramRL: Accelerating RL Post-Training for Structured Generation via Rollout-Aware N-gram Speculative Decoding

Earlier in this thread 4 items

Training Parallel Speculative Draft Models by Directly Minimizing Expected Decoding Rounds

Secure Speculative Decoding for Large Language Models

AsymSpec: Context-Asymmetric Speculative Decoding for Agentic LLMs

Exploring Speculative Decoding in vLLM on AMD GPUs