LLM Digest

Story

arxiv_cs_lg · Sep 21, 2026 · paper

Source brief

iSDFT: Information-Proximal Self-Distillation for Continual Learning in LLMs

arxiv.orgSep 21, 2026
original source linked

In brief

On-policy self-distillation fine-tuning (SDFT) learns new skills from demonstrations while reducing forgetting, but it always distils toward the full demonstration-conditioned teacher. This fixes teacher influence at...

Feed lens

evaluation

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

iSDFT: Information-Proximal Self-Distillation for Continual Learning in LLMs

Earlier in this thread 4 items

Benchmarking World Models for Continual Learning on Compositional Tasks

One Adapter, Many Tasks: Task-Conditioned Feature Transformations for Continual Learning

A Domain Incremental Continual Learning Benchmark for ICU Time Series Model Transportability

Fine-Tuning Regimes Define Distinct Continual Learning Problems