AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Smoothie: Smoothing Diffusion on Token Embeddings for Text Generation

Alexander Shabalin, Viacheslav Meshchaninov, Dmitry Vetrov

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 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
use smoothing diffusion space instead of simplex space (ablation)diffusion text model (~100M) · XSumROUGE-L: 23.0 -> 26.0Improved
use smoothing diffusion space instead of embedding space (ablation)diffusion text model (~100M) · XSumROUGE-L: 24.6 -> 26.0Improved
use smoothing diffusion space instead of embedding space (ablation)diffusion text model (~100M) · ROCStoriesMAUVE: 41.5 -> 80.7Improved
use smoothing diffusion space instead of simplex space (ablation)diffusion text model (~100M) · ROCStoriesMAUVE: 15.2 -> 80.7Improved
use smoothing diffusion space instead of embedding space (ablation)diffusion text model (~100M) · QQPBLEU: 31.3 -> 30.8No clear effect
use Eq. 9 solver (ignores previous latent) instead of DDPM (ablation)Smoothie · XSum / Quasar-T / QQP / ParaDetoxROUGE-L / J-Score: 25.6 -> 26.0 (XSum R-L)Improved
use DDIM solver instead of Eq. 9 solver (ablation)Smoothie · Quasar-TROUGE-L: 36.5 -> 31.8Hurt
lower reverse-process noise delta-tilde (ablation)Smoothie unconditional generation · ROCStoriesperplexity vs diversityMixed
increase number of denoising steps beyond 200 (ablation)Smoothie · XSumBERTScore: 68.8 -> 68.4 (200 -> 500 steps)Hurt
increase number of denoising steps (ablation)Smoothie · Quasar-TBERTScore: 63.1 at 25 to 500 stepsNo clear effect
use cross-entropy loss instead of MSE objective (ablation)Smoothie · seq2seq tasksgeneration qualityHurt
use embedding diffusion instead of smoothing for speeddiffusion text model · generation timeseconds: 2.897 -> 1.642Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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