Research

Paper

AI LLM March 18, 2026

Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing

Authors

Raghavv Goel, Mukul Gagrani, Mingu Lee, Chris Lott

Abstract

Large language models (LLMs) exhibit latent multi-token prediction (MTP) capabilities despite being trained solely for next-token generation. We propose a simple, training-free MTP approach that probes an LLM using on-the-fly mask tokens drawn from its embedding space, enabling parallel prediction of future tokens without modifying model weights or relying on auxiliary draft models. Our method constructs a speculative token tree by sampling top-K candidates from mask-token logits and applies a lightweight pruning strategy to retain high-probability continuations. During decoding, candidate predictions are verified in parallel, resulting in lossless generation while substantially reducing the number of model calls and improving token throughput. Across benchmarks, our probing-based MTP consistently outperforms existing training-free baselines, increasing acceptance length by approximately 12\% on LLaMA3 and 8--12\% on Qwen3, and achieving throughput gains of up to 15--19\%. Finally, we provide theoretical insights and empirical evidence showing that decoder layers naturally align mask-token representations with next-token states, enabling accurate multi-step prediction without retraining or auxiliary models.

Metadata

arXiv ID: 2603.17942
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-18
Fetched: 2026-03-19 06:01

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.17942v1</id>\n    <title>Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing</title>\n    <updated>2026-03-18T17:14:01Z</updated>\n    <link href='https://arxiv.org/abs/2603.17942v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.17942v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Large language models (LLMs) exhibit latent multi-token prediction (MTP) capabilities despite being trained solely for next-token generation. We propose a simple, training-free MTP approach that probes an LLM using on-the-fly mask tokens drawn from its embedding space, enabling parallel prediction of future tokens without modifying model weights or relying on auxiliary draft models. Our method constructs a speculative token tree by sampling top-K candidates from mask-token logits and applies a lightweight pruning strategy to retain high-probability continuations. During decoding, candidate predictions are verified in parallel, resulting in lossless generation while substantially reducing the number of model calls and improving token throughput. Across benchmarks, our probing-based MTP consistently outperforms existing training-free baselines, increasing acceptance length by approximately 12\\% on LLaMA3 and 8--12\\% on Qwen3, and achieving throughput gains of up to 15--19\\%. Finally, we provide theoretical insights and empirical evidence showing that decoder layers naturally align mask-token representations with next-token states, enabling accurate multi-step prediction without retraining or auxiliary models.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-18T17:14:01Z</published>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Raghavv Goel</name>\n    </author>\n    <author>\n      <name>Mukul Gagrani</name>\n    </author>\n    <author>\n      <name>Mingu Lee</name>\n    </author>\n    <author>\n      <name>Chris Lott</name>\n    </author>\n  </entry>"
}