Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Sampling

• URL: http://arxiv.org/abs/2606.12370v1
• Date: Wed, 10 Jun 2026 17:36:45 GMT
• Title: Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Sampling
• Abstract summary: Reinforcement learning (RL) has become a key component in modern large language models, yet the rollout stage remains the key bottleneck in RL training pipelines.<n>Although Multi-Token Prediction (MTP) offers a natural solution to accelerate rollouts through speculative decoding, many studies have observed that MTP acceptance rates degrade significantly during RL training.<n>We present Bebop, a systematic study of MTP in LLM post-training, and offer practical recipes to integrate MTP into large-scale RL pipelines.
• Score: 87.16803442525755
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement learning (RL) has become a key component in modern large language models, yet the rollout stage remains the key bottleneck in RL training pipelines. Although Multi-Token Prediction (MTP) offers a natural solution to accelerate rollouts through speculative decoding, many studies have observed that MTP acceptance rates degrade significantly during RL training, leading to limited speedup performance. To address this bottleneck, we present Bebop, a systematic study of MTP in LLM post-training, and offer practical recipes to integrate MTP into large-scale RL pipelines. First, we reveal that the MTP acceptance rate is fundamentally bounded by the fluctuation of model entropy, which demonstrates a clear negative linear relationship with the rise of entropy in the RL stage. Second, we show that probabilistic rejection sampling largely alleviates the disturbance introduced by entropy in RL compared to greedy draft sampling. We further identify that the conventional MTP training objectives (cross-entropy or KL) are suboptimal in such settings, and therefore we propose a novel end-to-end TV loss that directly optimizes multi-step rejection sampling acceptance rate, yielding ~10% acceptance rate improvements, achieving up to 95% acceptance rates and up to 25% extra inference throughput gains across mathematical reasoning, code generation, and agentic tasks. Third, we test various online MTP training strategies during RL and show that pre-RL MTP training with e2e TV loss and rejection sampling achieves a consistent acceptance rate and speedup throughout the entire RL, eliminating the need for costly online MTP updating. We provide extensive experiments and analysis that validate our findings. Experimental results show our method achieves up to 1.8x end-to-end acceleration in async RL training of Qwen3.5, Qwen3.6, and Qwen3.7 models.

Related papers

• Entropy-KL Divergence-based Token Masking: A Novel Approach for Selective Fine-tuning of Large Language Models [52.11240605311707]
  Supervised fine-tuning followed by reinforcement learning has become a standard post-training paradigm for large language models.<n>We propose EKSFT (Entropy-KL Selective Fine-Tuning), which selectively masks tokens that exhibit either high entropy or high KL divergence from a reference model.<n> Empirical evaluations on mathematical reasoning benchmarks demonstrate that EKSFT consistently outperforms standard SFT.
  arXiv  Detail & Related papers  (2026-05-28T03:36:05Z)
• Joint Training of Multi-Token Prediction in Reinforcement Learning via Optimal Coefficient Calibration [61.46060073417047]
  Multi-Token Prediction (MTP) has been a widely adopted module in pretraining.<n>We show that the per-step effect of MTP on the RL objective can be decomposed into two terms: a first-order correlation and a second-order penalty.<n>We propose an adaptive scheme that tracks the optimal coefficient online via a log-probability proxy at negligible cost.
  arXiv  Detail & Related papers  (2026-05-27T09:07:06Z)
• Self-Distillation for Multi-Token Prediction [27.12687724768597]
  Multi-Token Prediction (MTP) could accelerate inference by predicting multiple future tokens in parallel.<n>We propose MTP-D, a simple yet effective self-distillation method with minimal additional training cost.<n>We also introduce a looped extension strategy for MTP-D, enabling effective and economical MTP head extension.
  arXiv  Detail & Related papers  (2026-03-25T04:00:29Z)
• Efficient and Stable Reinforcement Learning for Diffusion Language Models [59.75789436018925]
  Reinforcement Learning (RL) is crucial for unlocking the complex reasoning capabilities of Diffusion-based Large Language Models (dLLMs)<n>Applying to dLLMs faces unique challenges in efficiency and stability.<n>We propose Spatio-Temporal Pruning (STP), a framework designed to simultaneously improve the efficiency and stability of RL for dLLMs.
  arXiv  Detail & Related papers  (2026-02-09T17:04:23Z)
• Good SFT Optimizes for SFT, Better SFT Prepares for Reinforcement Learning [8.550698116833123]
  Post-training of reasoning LLMs typically consists of an offline SFT stage followed by an online reinforcement learning stage.<n>We show that, after identical RL training, models from stronger SFT checkpoints can significantly underperform those from weaker ones.<n>We propose PEAR, an SFT-stage method that corrects this mismatch and better prepares the model for RL.
  arXiv  Detail & Related papers  (2026-02-01T06:53:45Z)
• FastMTP: Accelerating LLM Inference with Enhanced Multi-Token Prediction [11.691960175716163]
  This paper introduces FastMTP, a method that improves multi-step draft quality by aligning MTP training with its inference pattern.<n>Our approach fine-tunes a single MTP head with position-shared weights on self-distilled data, enabling it to capture dependencies among consecutive future tokens.<n> Experimental results across seven diverse benchmarks demonstrate that FastMTP achieves an average of 2.03x speedup compared to standard next token prediction.
  arXiv  Detail & Related papers  (2025-09-16T07:36:26Z)
• Ring-lite: Scalable Reasoning via C3PO-Stabilized Reinforcement Learning for LLMs [51.21041884010009]
  Ring-lite is a Mixture-of-Experts (MoE)-based large language model optimized via reinforcement learning (RL)<n>Our approach matches the performance of state-of-the-art (SOTA) small-scale reasoning models on challenging benchmarks.
  arXiv  Detail & Related papers  (2025-06-17T17:12:34Z)
• On multi-token prediction for efficient LLM inference [0.36681882674260474]
  We first show that such models inherently possess MTP capabilities via numerical marginalization over intermediate token probabilities.<n>We then explore the challenges of integrating MTP heads into frozen LLMs and find that their hidden layers are strongly specialized for NTP.
  arXiv  Detail & Related papers  (2025-02-13T15:42:44Z)
• Fair and Efficient Distributed Edge Learning with Hybrid Multipath TCP [62.81300791178381]
  The bottleneck of distributed edge learning over wireless has shifted from computing to communication. Existing TCP-based data networking schemes for DEL are application-agnostic and fail to deliver adjustments according to application layer requirements. We develop a hybrid multipath TCP (MP TCP) by combining model-based and deep reinforcement learning (DRL) based MP TCP for DEL.
  arXiv  Detail & Related papers  (2022-11-03T09:08:30Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.