# Investigating ReLoRA: Effects on the Learning Dynamics of Small Language Models

arXiv: https://arxiv.org/abs/2509.12960 · HTML: https://mlsanity.com/paper/2509.12960/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 23 of its 41 references that are in the map.
- First posted on arXiv 16 Sep 2025.

## Builds on (most cited first, 23 of 23)

- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets](https://mlsanity.com/paper/2311.15127/) (2023)
- [GLU Variants Improve Transformer](https://mlsanity.com/paper/2002.05202/) (2020)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuning](https://mlsanity.com/paper/2012.13255/) (2020)
- [Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey](https://mlsanity.com/paper/2403.14608/) (2024)
- [Bias and Fairness in Large Language Models: A Survey](https://mlsanity.com/paper/2309.00770/) (2023)
- [Green AI](https://mlsanity.com/paper/1907.10597/) (2019)
- [OLMo: Accelerating the Science of Language Models](https://mlsanity.com/paper/2402.00838/) (2024)
- [BLiMP: The Benchmark of Linguistic Minimal Pairs for English](https://mlsanity.com/paper/1912.00582/) (2019)
- [What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language models](https://mlsanity.com/paper/1907.13528/) (2019)
- [ReLoRA: High-Rank Training Through Low-Rank Updates](https://mlsanity.com/paper/2307.05695/) (2023)
- [Are Large Pre-Trained Language Models Leaking Your Personal Information?](https://mlsanity.com/paper/2205.12628/) (2022)
- [Signal Propagation in Transformers: Theoretical Perspectives and the Role of Rank Collapse](https://mlsanity.com/paper/2206.03126/) (2022)
- [Paloma: A Benchmark for Evaluating Language Model Fit](https://mlsanity.com/paper/2312.10523/) (2023)
- [What Kind of Language Is Hard to Language-Model?](https://mlsanity.com/paper/1906.04726/) (2019)
- [Holistically Evaluating the Environmental Impact of Creating Language Models](https://mlsanity.com/paper/2503.05804/) (2025)
- [Why do small language models underperform? Studying Language Model Saturation via the Softmax Bottleneck](https://mlsanity.com/paper/2404.07647/) (2024)

## Built on by (most cited first, 1 of 1)

- [Reversible Foundations: Training a 120B Sparse MoE through State-Preserving Scaling](https://mlsanity.com/paper/2606.07404/) (2026)

## Abstract (arXiv)

Parameter-efficient methods like LoRA have revolutionised large language model (LLM) fine-tuning. ReLoRA extends this idea to pretraining by repeatedly merging and reinitialising low-rank adapters, increasing cumulative rank while keeping updates cheap. This aligns well with observations that high-capacity models learn through locally low-rank trajectories that expand over time. By contrast, recent work suggests that small language models (SLMs) exhibit rank deficiencies and under-utilise their available dimensionality. This raises a natural question: can ReLoRA's rank-expanding update rule textitsteer SLMs toward healthier learning dynamics, mitigating rank bottlenecks in a capacity-constrained regime? We argue SLMs are an ideal testbed: they train quickly, enable controlled ablations, and make rank phenomena more measurable. We present the first systematic study of ReLoRA in SLMs (11M-66M parameters), evaluating both performance and learning dynamics. Across loss, Paloma perplexity, and BLiMP, we find that ReLoRA underperforms full-rank training, with gaps widening at larger scales. Analysis of proportional effective rank and condition numbers shows that ReLoRA amplifies existing rank deficiencies and induces ill-conditioned updates early in training. Our results suggest that while ReLoRA's merge-and-restart strategy can expand ranks in larger models, it does not straightforwardly translate to capacity-limited SLMs, motivating adaptive-rank or hybrid-rank approaches for low-compute pretraining.