REVIEW 6 cited by
Language Model Cascades: Token-level uncertainty and beyond
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Recent advances in language models (LMs) have led to significant improvements in quality on complex NLP tasks, but at the expense of increased inference costs. Cascading offers a simple strategy to achieve more favorable cost-quality tradeoffs: here, a small model is invoked for most "easy" instances, while a few "hard" instances are deferred to the large model. While the principles underpinning cascading are well-studied for classification tasks - with deferral based on predicted class uncertainty favored theoretically and practically - a similar understanding is lacking for generative LM tasks. In this work, we initiate a systematic study of deferral rules for LM cascades. We begin by examining the natural extension of predicted class uncertainty to generative LM tasks, namely, the predicted sequence uncertainty. We show that this measure suffers from the length bias problem, either over- or under-emphasizing outputs based on their lengths. This is because LMs produce a sequence of uncertainty values, one for each output token; and moreover, the number of output tokens is variable across examples. To mitigate this issue, we propose to exploit the richer token-level uncertainty information implicit in generative LMs. We argue that naive predicted sequence uncertainty corresponds to a simple aggregation of these uncertainties. By contrast, we show that incorporating token-level uncertainty through learned post-hoc deferral rules can significantly outperform such simple aggregation strategies, via experiments on a range of natural language benchmarks with FLAN-T5 models. We further show that incorporating embeddings from the smaller model and intermediate layers of the larger model can give an additional boost in the overall cost-quality tradeoff.
Forward citations
Cited by 6 Pith papers
-
Routing Is Least Learnable Where It Is Most Valuable: Bounds on Representation Routing for Web Agents
Per-task routing between text, image, and hybrid observations of a browser page does not currently beat one fixed choice, because the labels needed to learn routing exist only where the agent already succeeds; only a ...
-
HACO: Hedged Agent Computing for Reliable LLM Systems
HACO adaptively selects a reliability-constrained hedge set of candidate LLM agent instances for each role invocation, improving robustness under changing deployment conditions at lower token/latency cost than full-po...
-
Constraint-Anchored Reasoning Traces
CART interleaves language-model reasoning with verifiable symbolic anchors and backtracking, reducing error snowballing in open-source MLLMs.
-
BEST-Route: Adaptive LLM Routing with Test-Time Optimal Compute
A routing system that chooses both the model and the number of samples per query to meet a quality threshold, yielding up to 60% cost savings.
-
AdaptiveLLM: A Framework for Selecting Optimal Cost-Efficient LLM for Code-Generation Based on CoT Length
AdaptiveLLM routes coding problems to cost-effective LLMs by clustering chain-of-thought lengths into three difficulty levels, improving pass@1 by 7.86% over ComplexityNet while cutting inference cost by 88.9%.
-
Maximizing Confidence Alone Improves Reasoning
Reinforcement learning with a reward equal to the negative entropy of the model's own token distribution improves benchmark accuracy without any labeled answers.
Discussion (0). Continue with ORCID to comment.