AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Disentangling Geometry, Performance, and Training in Language Models

Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models

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
higher effective rank of the unembedding matrixOLMo-style models (4M-75M) · Pile-10k ID losslossMixed
use effective rank instead of ID loss to predict OOD lossOLMo-style models · Paloma Dolma-100 programming subsetOOD lossNo clear effect
rank collapse in small models under post-training quantization (GPTQ)OLMo-4M, OLMo-28M · post-quantization ID losslossHurt
train OLMo-14M matching Pythia-14M config (sequential instead of parallel attention) (ablation)OLMo-14M · late-stage pretraining saturationID lossImproved
reduce token budget to 8B and batch size to 32 for OLMo-14M (ablation)OLMo-14M · saturationID lossNo clear effect
increase batch size (ablation)OLMo-style pretraining · Pile ID losseffective rank and lossMixed
increase weight decay (0.01 to 0.5) (ablation)OLMo-style pretraining · Pile ID losseffective rank and lossMixed
scale learning rate by 0.1, 10, 100 (ablation)OLMo-style pretraining · Pile ID losslossMixed
anneal learning rate to 10% instead of 0-1% of peak (ablation)OLMo-style pretraining · Pile ID losseffective rank and lossHurt
train on more tokens (8B to 128B) (ablation)OLMo-style pretraining · Pile ID lossloss and effective rankImproved
fine-tune models with varied initial effective rank on StarCoder-PythonOLMo-style models · fine-tuning and forgetting on Pile-10klossNo clear effect
control for training hyperparameters (partial Spearman) when relating geometry to loss (ablation)geometric metrics · Pile-10k ID lossSpearman correlation: -0.699 -> -0.130 (effective rank)Hurt

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