Paper Detail

Building Multi-Task Agentic LLMs via Two-Phase Distillation

Huaijie Wang, Shusheng Xu, Yi Wu, Kaifeng Lyu

arxiv Score 8.8

Published 2026-06-29 · First seen 2026-06-30

General AI

Abstract

A key step toward artificial general intelligence is to train models that can perform multiple tasks. In this paper, we study how to build such models by first training separate RL experts for individual tasks and then consolidating them via distillation, as an alternative to directly training a single model on mixed tasks. We show that off-policy distillation degrades in multi-task settings due to the mode-covering nature of forward KL: aggregating data from multiple tasks introduces a large number of behavioral modes that can exceed the student's capacity, forcing it to average across behaviors and leading to degraded performance. In contrast, on-policy distillation is mode-seeking but requires strong initialization. Inspired by these observations, we propose a two-phase approach: off-policy distillation followed by on-policy refinement. Evaluation across conversational agents and text-based games confirms that this two-phase approach matches single-task RL expert performance for each individual task, whereas off-policy or on-policy distillation alone fails to match this performance.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
soon
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{wang2026building,
  title = {Building Multi-Task Agentic LLMs via Two-Phase Distillation},
  author = {Huaijie Wang and Shusheng Xu and Yi Wu and Kaifeng Lyu},
  year = {2026},
  abstract = {A key step toward artificial general intelligence is to train models that can perform multiple tasks. In this paper, we study how to build such models by first training separate RL experts for individual tasks and then consolidating them via distillation, as an alternative to directly training a single model on mixed tasks. We show that off-policy distillation degrades in multi-task settings due to the mode-covering nature of forward KL: aggregating data from multiple tasks introduces a large nu},
  url = {https://arxiv.org/abs/2606.30044},
  keywords = {cs.LG},
  eprint = {2606.30044},
  archiveprefix = {arXiv},
}

Metadata

{}