Paper Detail

ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging

Neha Verma, Nikhil Mehta, Shao-Chuan Wang, Naijing Zhang, Alicia Tsai, Li Wei, Lukasz Heldt, Lichan Hong, Ed Chi, Xinyang Yi

arxiv Score 19.5

Published 2026-05-12 · First seen 2026-05-13

Research Track A · General AI

Abstract

Despite the rapid advancements in large language model (LLM) development, fine-tuning them for specific tasks often results in the catastrophic forgetting of their general, language-based reasoning abilities. This work investigates and addresses this challenge in the context of the Generative Retrieval (GenRetrieval) task. During GenRetrieval fine-tuning, we find this forgetting occurs rapidly and correlates with the distance between the fine-tuned and original model parameters. Given these observations, we propose ORBIT, a novel approach that actively tracks the distance between fine-tuned and initial model weights, and uses a weight averaging strategy to constrain model drift during GenRetrieval fine-tuning when this inter-model distance exceeds a maximum threshold. Our results show that ORBIT retains substantial text and retrieval performance by outperforming both common continual learning baselines and related regularization methods that also employ weight averaging.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
now
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{verma2026orbit,
  title = {ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging},
  author = {Neha Verma and Nikhil Mehta and Shao-Chuan Wang and Naijing Zhang and Alicia Tsai and Li Wei and Lukasz Heldt and Lichan Hong and Ed Chi and Xinyang Yi},
  year = {2026},
  abstract = {Despite the rapid advancements in large language model (LLM) development, fine-tuning them for specific tasks often results in the catastrophic forgetting of their general, language-based reasoning abilities. This work investigates and addresses this challenge in the context of the Generative Retrieval (GenRetrieval) task. During GenRetrieval fine-tuning, we find this forgetting occurs rapidly and correlates with the distance between the fine-tuned and original model parameters. Given these obse},
  url = {https://arxiv.org/abs/2605.12419},
  keywords = {cs.CL, cs.IR, cs.LG},
  eprint = {2605.12419},
  archiveprefix = {arXiv},
}

Metadata

{}