AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Late-to-Early Training: LET LLMs Learn Earlier, So Faster and Better

Ji Zhao, Shitong Shao, Yufei Gu, Xun Zhou, liang xiang, Zeke Xie

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 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
Align early layers of target LLM to a small pretrained model's late layer (LET) vs standard training1.4B LLM · 9 downstream tasks, one-shotaverage accuracy: 41.6 -> 43.6Improved
Align early layers to a small pretrained model (LET) vs standard training7B LLM · 9 downstream tasks, one-shotaverage accuracy: 43.3 -> 45.5Improved
Use reverse knowledge distillation from small teacher (RKD)LLM pretraining · 9 downstream tasks, one-shotaverage accuracy: 41.6 -> 41.4 (1.4B); 43.3 -> 42.2 (7B)Hurt
Use LET instead of SALTLLM pretraining · 9 downstream tasks, one-shotaverage accuracy: 42.9 -> 43.6 (1.4B); 44.7 -> 45.5 (7B)Improved
Train LET for only 67% of steps1.4B LLM · 9 downstream tasks, one-shotaverage accuracy: 41.6 -> 42.5Improved
Align middle layer of T instead of final layer (ablation)LET · average downstream accuracyaverage accuracyHurt
Align T's last layer to early layer of M (L2E) instead of middle or last layer of M (ablation)LET · downstream accuracy and perplexityaverage accuracyImproved
Set lambda above 0.1 (ablation)LET · average downstream accuracyaverage accuracyHurt
Use lambda=0.01 instead of 0.1 (ablation)LET · average downstream accuracyaverage accuracyHurt
Replace cosine alignment loss with logsum loss (ablation)LET · 9 downstream tasksaverage accuracy: 43.6 -> 43.7Improved
Replace cosine alignment with CCA (ablation)LET · 9 downstream tasksaverage accuracy: 43.6 -> 42.7Hurt
Tokenizer mismatch between T and M (ablation)LET · 9 downstream tasksaverage accuracy: 43.6 -> 42.3Hurt

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