cs.LGOct 30, 2026

There Is No Free Loop: When Does Looping a Trained Model Pay for Its Inference Compute?

Authors: Rhea Sanjay Sukthanker, Pashmina Cameron

Organizations: Microsoft Applied Sciences Group

Abstract

Looping re-applies a block of layers so that each token passes through the block more than once. Adding a loop to a trained model is attractive because it leaves the parameter count, and hence the disk size, the memory footprint and the training cost unchanged. Unfortunately, every generated token pays for the loop. Re-applying m of N layers makes each token about (N+m)/N times as expensive. Loops that a model learns during large-scale pretraining justify this price, but whether loops added after pretraining justify this cost remains unclear. We compare a looped model with two simple baselines. We first compare it with the same model trained identically without the loop (equal parameters). Secondly, we compare it with the best unlooped use of the same inference floating-point operations (equal FLOPs). Across Qwen3 and Qwen3.5 models from 0.8B to 9B parameters, only the looped 0.8B model beats both baselines. At 4B and 9B, the looped model does not beat the first baseline, i.e. the same model trained without the loop. This is not because the benchmarks are too easy, since on the same benchmarks the unlooped 9B model still scores up to 13 points higher than the 4B model, showing that there is ample room for improvement. Introducing loops in an already pretrained model therefore only pays for small models with a fixed parameter budget.

Figures & tables

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 30, 2026

Closing the Loop: Practical Training Recipes for Looped Language Models

Looped language models increase effective depth by repeatedly applying a shared block of layers, but existing large-scale recipes require multi-stage training over trillions of tokens, while the benefits of recurrence remain difficult to separate from differences in data and training. In this work, we establish practical training recipes for looped language models, with three main results. (1) We develop a compute-efficient from-scratch pipeline that reduces the training budget from 7.7T tokens in Ouro to 310B tokens while retaining strong reasoning performance. Pretraining followed by high-quality mid-training, together with learning-rate warmup and stronger exit-gate regularization, enables stable recurrent training without prior multi-stage schedules. (2) Under controlled comparisons, our 1.4B LoopLM outperforms a parameter-matched dense model trained on the same data and token budget on all 12 evaluated benchmarks, including +14 points on GSM8K, +10 on MATH, and +22 on DROP. At matched inference compute, it approaches a 3.9B dense model on mathematical reasoning and reading comprehension while using only 36% as many parameters. (3) We introduce a minimal recipe for converting pretrained dense models into looped ones: a single learned input-mixing scalar and a smoothed exit loss, with no step-specific parameters. Applied to Qwen3-1.7B-Base, Looped Qwen improves over an identically continued dense baseline on every evaluated benchmark across two data regimes, with statistically clear gains on GSM8K, MATH, and MMLU-Pro on the curated mixture. Together, these results make looped language models substantially cheaper to train from scratch and practical to introduce into existing pretrained checkpoints, while isolating the gains due to recurrence itself.
Oct 29, 2026

How to Loop Your Transformer: Loop Scaling, Specialization, and Parallel Inference

Looping in transformer models reuses weights by repeatedly applying the same transformer, or parts of it, making the model deeper without increasing its parameter count. Whether looping is useful depends on which constraint dominates: model size, training compute, or autoregressive inference efficiency. We study these trade-offs jointly in dense decoder-only transformers trained from scratch, varying which parts of the model are looped, how many loops are applied, and how information is passed between loops. Our experiments span models with up to 1B parameters, trained on up to 155B tokens with up to 1.07×10211.07\times10^{21} FLOPs and 32 loops. We find that looping the full transformer generally gives the strongest quality at a fixed parameter budget, whereas looping a middle block of layers gives a better return when training FLOPs are constrained. Adding loops is not uniformly beneficial: gains eventually saturate, reverse, or become unstable. This motivates asking whether forcing every loop to use the same weights limits the benefit of deeper looping. We address this by adding small loop-specific low-rank adapters that improve quality, with larger gains as more loops are added. Looped transformers also suffer a decoding slowdown proportional to the number of loops. Parallel Loop Transformers address this by passing hidden states between loops across adjacent token positions, allowing loop stages to run concurrently. However, in our experiments the original formulation does not consistently retain the quality gains of ordinary looping. We find that preserving the current-token embedding and separately normalizing it and the previous-token hidden state before combining them recovers strong performance across model scales and training budgets. Together, these results show that effective looping requires matching the design to the resource constraint and inference setting.
Oct 18, 2026

UpLoop: Upcycling Large Language Models with Gated Loop

Looped large language models trade parameter storage for recurrent computation, but most are trained from scratch. In practice, pretraining with looped depth is expensive, making post-training upcycling of a pretrained non-looped LLM an attractive alternative. We introduce UpLoop, a post-training recipe based on layer-wise recurrence. Instead of looping a whole model or a middle block, UpLoop loops each selected layer locally, preserving the pretrained depth hierarchy and avoiding feature-space mismatch. To make loop execution useful, we add hidden-wise gates on the attention and MLP residual branches, so later loops act as progressive refinements rather than redundant rewrites. Empirically, UpLoop breaks the commonly observed saturation at two loops (R=2R=2) in prior looped models. Starting from Qwen3-4B Base and applying SFT, UpLoop improves the average accuracy across 20 benchmarks by 6.17% at R=3R=3. We use R=3R=3 as the main setting for the best accuracy--cost trade-off, while both native and extrapolated R=4R=4 yield further gains.