Causal LLM Routing: End-to-End Regret Minimization from Observational Data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tsiourvas, Asterios, Sun, Wei, Perakis, Georgia
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912798859067392
author Tsiourvas, Asterios
Sun, Wei
Perakis, Georgia
author_facet Tsiourvas, Asterios
Sun, Wei
Perakis, Georgia
contents LLM routing aims to select the most appropriate model for each query, balancing competing performance metrics such as accuracy and cost across a pool of language models. Prior approaches typically adopt a decoupled strategy, where the metrics are first predicted and the model is then selected based on these estimates. This setup is prone to compounding errors and often relies on full-feedback data, where each query is evaluated by all candidate models, which is costly to obtain and maintain in practice. In contrast, we learn from observational data, which records only the outcome of the model actually deployed. We propose a causal end-to-end framework that learns routing policies by minimizing decision-making regret from observational data. To enable efficient optimization, we introduce two theoretically grounded surrogate objectives: a classification-based upper bound, and a softmax-weighted regret approximation shown to recover the optimal policy at convergence. We further extend our framework to handle heterogeneous cost preferences via an interval-conditioned architecture. Experiments on public benchmarks show that our method outperforms existing baselines, achieving state-of-the-art performance across different embedding models.
format Preprint
id arxiv_https___arxiv_org_abs_2505_16037
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Causal LLM Routing: End-to-End Regret Minimization from Observational Data
Tsiourvas, Asterios
Sun, Wei
Perakis, Georgia
Artificial Intelligence
Computation and Language
Machine Learning
Optimization and Control
LLM routing aims to select the most appropriate model for each query, balancing competing performance metrics such as accuracy and cost across a pool of language models. Prior approaches typically adopt a decoupled strategy, where the metrics are first predicted and the model is then selected based on these estimates. This setup is prone to compounding errors and often relies on full-feedback data, where each query is evaluated by all candidate models, which is costly to obtain and maintain in practice. In contrast, we learn from observational data, which records only the outcome of the model actually deployed. We propose a causal end-to-end framework that learns routing policies by minimizing decision-making regret from observational data. To enable efficient optimization, we introduce two theoretically grounded surrogate objectives: a classification-based upper bound, and a softmax-weighted regret approximation shown to recover the optimal policy at convergence. We further extend our framework to handle heterogeneous cost preferences via an interval-conditioned architecture. Experiments on public benchmarks show that our method outperforms existing baselines, achieving state-of-the-art performance across different embedding models.
title Causal LLM Routing: End-to-End Regret Minimization from Observational Data
topic Artificial Intelligence
Computation and Language
Machine Learning
Optimization and Control
url https://arxiv.org/abs/2505.16037