# Squeezing More from Limited Data with Recursive Transformers

arXiv: https://arxiv.org/abs/2608.26973 · HTML: https://mlsanity.com/paper/2608.26973/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 39 references that are in the map.
- First posted on arXiv 27 Aug 2026.

## Builds on (most cited first, 25 of 25)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [The LAMBADA dataset: Word prediction requiring a broad discourse context](https://mlsanity.com/paper/1606.06031/) (2016)
- [Universal Transformers](https://mlsanity.com/paper/1807.03819/) (2018)
- [Deep Equilibrium Models](https://mlsanity.com/paper/1909.01377/) (2019)
- [BLiMP: The Benchmark of Linguistic Minimal Pairs for English](https://mlsanity.com/paper/1912.00582/) (2019)
- [Adaptive Computation Time for Recurrent Neural Networks](https://mlsanity.com/paper/1603.08983/) (2016)
- [Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach](https://mlsanity.com/paper/2502.05171/) (2025)
- [Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free](https://mlsanity.com/paper/2505.06708/) (2025)
- [Query-Key Normalization for Transformers](https://mlsanity.com/paper/2010.04245/) (2020)
- [Understanding the Difficulty of Training Transformers](https://mlsanity.com/paper/2004.08249/) (2020)
- [NorMuon: Making Muon more efficient and scalable](https://mlsanity.com/paper/2510.05491/) (2025)
- [Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA](https://mlsanity.com/paper/2410.20672/) (2024)
- [Nemotron-CLIMB: CLustering-based Iterative Data Mixture Bootstrapping for Language Model Pre-training](https://mlsanity.com/paper/2504.13161/) (2025)
- [Elements of World Knowledge (EWoK): A Cognition-Inspired Framework for Evaluating Basic World Knowledge in Language Models](https://mlsanity.com/paper/2405.09605/) (2024)
- [Parcae: Scaling Laws For Stable Looped Language Models](https://mlsanity.com/paper/2604.12946/) (2026)
- [Findings of the Second BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora](https://mlsanity.com/paper/2412.05149/) (2024)
- [Understanding Transformer from the Perspective of Associative Memory](https://mlsanity.com/paper/2505.19488/) (2025)
- [GPT or BERT: why not both?](https://mlsanity.com/paper/2410.24159/) (2024)
- [AbbIE: Autoregressive Block-Based Iterative Encoder for Efficient Sequence Modeling](https://mlsanity.com/paper/2507.08567/) (2025)
- [Mask and You Shall Receive: Optimizing Masked Language Modeling For Pretraining BabyLMs](https://mlsanity.com/paper/2510.20475/) (2025)

## Abstract (arXiv)

Pre-training under limited data requires a different view of scaling than web-scale language modeling. With a fixed data budget but relatively abundant compute, increasing parameter count helps only up to an optimal scale; beyond that point, models overfit and generalization worsens. We study this behavior across 10M-100M word pre-training budgets, two corpora, and multiple downstream evaluations, and find that optimal size depends strongly on both the data budget and the downstream target. We argue that standard Transformers scale down poorly to this setting, because embeddings consume a large fraction of the parameter budget and per-token computation is tied to representational capacity. To address this coupling, we study recursive Transformers, reusing a shared block across depth to scale compute, together with factorized embeddings to reduce vocabulary-map parameters. We train three recursive models and find that they outperform standard Transformers at 10M and 100M words, while remaining competitive with BabyLM Challenge 2025 winners.