AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Token Distillation: Attention-Aware Input Embeddings for New Tokens

Konstantin Dobler, Desmond Elliott, Gerard de Melo

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Language model pretraining and induction · 4 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
add hidden-state distillation instead of subtoken mean initsubtoken mean initialization · biomedical benchmarks avgaccuracy: 60.8 -> 64.6Improved
use token distillation instead of next-token prediction tuningNTP · biomedical benchmarks avgaccuracy: 63.0 -> 64.6Improved
use token distillation instead of ZeTT hyper-networkZeTT · biomedical Llama3-8Baccuracy: 66.1 -> 67.3Improved
freeze original embeddings and optimize only new tokens (NTP) (ablation)NTP tune all embeddings · biomedical benchmarks avgaccuracy: 59.8 -> 63.0Improved
sum NTP loss with distillation loss (ablation)Token Distillation · biomedical benchmarks avgaccuracy: 64.6 -> 64.3Hurt
scale NTP loss by autoscaling factor alpha (ablation)Token Distillation + NTP · biomedical benchmarks avgaccuracy: 64.3 -> 64.7Improved
add autoscaled NTP to token distillation (ablation)Token Distillation · biomedical Llama3.2-3Baccuracy: 56.2 -> 57.9Improved
generate contexts with model instead of retrieving from corpus (ablation)Token Distillation · biomedical benchmarks avgaccuracy: 64.6 -> 64.3Hurt
start from subtoken mean instead of random init (ablation)Token Distillation · biomedical Mistral-7Baccuracy: 62.5 -> 63.5Improved
start from ZeTT embeddings instead of subtoken mean (ablation)Token Distillation · biomedical Llama3-8Baccuracy: 67.3 -> 67.9Improved
use KL or logit-MSE instead of hidden-state MSE (ablation)Token Distillation · biomedical benchmarksaccuracyNo clear effect
target an earlier layer than the last (ablation)Token Distillation · biomedical Llama3-8B-InstructaccuracyImproved

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