AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

From Growing to Looping: A Unified View of Iterative Computation in LLMs

Ferdinand Kapl, Emmanouil Angelis, Kaitlin Maile, Johannes von Oswald, Stefan Bauer

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
grow the model in depth by middle-layer duplication (LIDAS)SmolLM 1.7B · Reasoning Primitivesaccuracy: 34.62 -> 48.02Improved
use MIDAS instead of LIDAS growth operator (ablation)SmolLM 1.7B · Reasoning Primitivesaccuracy: 48.02 -> 40.88Hurt
loop a 12-layer block twice instead of standard 12-layer model (iso-param)SmolLM 1.7B · Reasoning Primitivesaccuracy: 35.42 -> 41.22Improved
fully loop a 4-layer block 6 times vs standard iso-inference baselineSmolLM 1.7B · holdout NLL / knowledgeNLL: 1.96 -> 2.12Hurt
keep unique encoder and decoder layers and loop only middle block (Loop 4-4x4-4) (ablation)looped 1.7B · layer-swap robustnessrelative accuracyImproved
repeat a 4-layer middle block at inference time without training to loopMIDAS and LIDAS · Copy Real Wordsaccuracy: up to 2xImproved
repeat a middle block more than twice at inference (ablation)grown models · reasoning primitivesaccuracyNo clear effect
more in-context examples for looped or grown modelsLIDAS, Loop models · Reasoning Primitives few-shotaccuracyImproved
use Nemotron-CC-Math-4+ instead of FineMath-4+ for 20% math cooldown (ablation)LIDAS 360M · Math Word Problemsaccuracy: 16.54 -> 25.64Improved
add 20% Nemotron-CC-Math-4+ cooldownSmolLM 1.7B LIDAS · GSM8Kaccuracy: 5.61 -> 13.34Improved
retrofit one extra loop of layers 8-11 during math cooldownLIDAS 1.7B · GSM8Kaccuracy: 13.34 -> 17.36Improved
retrofit loop of layers 12-15 instead of 8-11 (ablation)LIDAS 1.7B · Reasoning Primitivesaccuracy: 54.84 -> 56.60 on primitives; 17.36 -> 15.47 on GSM8KMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy