LLM Digest

Story

arxiv_cs_lg · Oct 30, 2026 · paper

Source brief

A Better Critic Matters for Distribution Matching Distillation

arxiv.orgOct 30, 2026
original source linked

In brief

Distribution Matching Distillation (DMD) trains a few-step student using the difference between a teacher's score and an online critic's estimate of the student distribution's score. We investigate what the critic needs, in capacity and in information, to provide a reliable train

Read the original at arxiv.org →Open in live feed Read that day’s brief

A Better Critic Matters for Distribution Matching Distillation

Earlier in this thread 4 items

DecisionRank: On-Policy Distillation for Decision-Model Rerankers

Where You Inject Matters: The Trust Region Effect in LLM Agents

Learning When to Commit: Schedule Distillation for Fast Diffusion Language Models

FOCUS: Active Critic Learning for Long-Horizon Language Model Agents