Efficient Pre-Training with Token Superposition

Hugging Face Daily Papers Papers

Summary

Token-Superposition Training (TST) improves LLM pre-training efficiency by combining contiguous tokens into bags during a superposition phase with a multi-hot cross-entropy objective, achieving up to 2.5x reduction in training time without architectural changes.

Pre-training of Large Language Models is often prohibitively expensive and inefficient at scale, requiring complex and invasive modifications in order to achieve high data throughput. In this work, we present Token-Superposition Training (TST), a simple drop-in method that significantly improves the data throughput per FLOPs during pre-training without modifying the parallelism, optimizer, tokenizer, data, or model architecture. TST is done in two phases: (i) A highly efficient superposition phase where we combine many contiguous tokens into one bag and train using a multi-hot cross-entropy (MCE) objective, and (ii) a recovery phase where we revert back to standard training. We extensively evaluate TST on the scale of 270M and 600M parameters and validate on 3B and a 10B A1B mixture of experts model, demonstrating that it is highly robust in different settings. Ultimately, TST consistently outperforms baseline loss and downstream evaluations, and under equal-loss settings, TST yields up to a 2.5x reduction in total pre-training time at the 10B A1B scale.
Original Article
View Cached Full Text

Cached at: 05/13/26, 04:13 PM

Paper page - Efficient Pre-Training with Token Superposition

Source: https://huggingface.co/papers/2605.06546

Abstract

Token-Superposition Training (TST) improves pre-training efficiency by combining contiguous tokens into bags during a superposition phase with multi-hot cross-entropy objective, achieving faster training times without architectural changes.

Pre-trainingof Large Language Models is often prohibitively expensive and inefficient at scale, requiring complex and invasive modifications in order to achieve highdata throughput. In this work, we presentToken-Superposition Training(TST), a simple drop-in method that significantly improves thedata throughputperFLOPsduringpre-trainingwithout modifying theparallelism,optimizer,tokenizer, data, ormodel architecture. TST is done in two phases: (i) A highly efficientsuperposition phasewhere we combine many contiguous tokens into one bag and train using amulti-hot cross-entropy(MCE) objective, and (ii) arecovery phasewhere we revert back to standard training. We extensively evaluate TST on the scale of 270M and 600M parameters and validate on 3B and a 10B A1B mixture of experts model, demonstrating that it is highly robust in different settings. Ultimately, TST consistently outperforms baseline loss and downstream evaluations, and under equal-loss settings, TST yields up to a 2.5x reduction in totalpre-trainingtime at the 10B A1B scale.

View arXiv pageView PDFAdd to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.06546 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.06546 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.06546 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

SuperThoughts: Reasoning Tokens in Superposition

arXiv cs.LG

SuperThoughts compresses consecutive chain-of-thought tokens into latent representations and decodes two tokens per step, achieving ~20–30% CoT length reduction with minimal accuracy loss on math reasoning benchmarks, while doubling inference throughput.

Stage-adaptive Token Selection for Efficient Omni-modal LLMs

Hugging Face Daily Papers

SEATS is a training-free, stage-adaptive token selection method that reduces computational overhead in omni-modal LLMs by progressively pruning redundant visual and audio tokens, achieving a 9.3x FLOPs reduction and 4.8x prefill speedup while preserving 96.3% performance.