Forecasting Downstream Performance of LLMs With Proxy Metrics

• URL: http://arxiv.org/abs/2605.18607v1
• Date: Mon, 18 May 2026 16:17:15 GMT
• Title: Forecasting Downstream Performance of LLMs With Proxy Metrics
• Abstract summary: We propose to construct proxy metrics by aggregating token-level statistics, such as entropy, top-k accuracy, and expert token rank.<n>Across three settings, our proxies consistently outperform loss- and compute-based baselines.
• Score: 37.61408486519996
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Progress in language model development is often driven by comparative decisions: which architecture to adopt, which pretraining corpus to use, or which training recipe to apply. Making these decisions well requires reliable performance forecasts, yet the two commonly used signals are fundamentally limited. Cross-entropy loss is poorly aligned with downstream capabilities, and direct downstream evaluation is expensive, sparse, and often uninformative at early training stages. Instead, we propose to construct proxy metrics by aggregating token-level statistics, such as entropy, top-k accuracy, and expert token rank, from a candidate model's next token distribution over expert-written solutions. Across three settings, our proxies consistently outperform loss- and compute-based baselines: 1) For cross-family model selection, they rank a heterogeneous population of reasoning models with mean Spearman Rho = 0.81 (vs. Rho = 0.36 for cross-entropy loss); 2) For pretraining data selection, they reliably rank 25 candidate corpora for a target model at roughly $10{,}000\times$ less compute than direct evaluation, pushing the Pareto frontier beyond existing methods; and 3) for training-time forecasting, they extrapolate downstream accuracy across an $18\times$ compute horizon with roughly half the error of existing alternatives. Together, these results suggest that expert trajectories are a broadly useful source of signal for assessing model capabilities, enabling reliable performance forecasting throughout the model development life cycle.

Related papers

• Rethinking One-Shot Federated Graph Learning: Training-Free Statistical Estimation [12.66442044096499]
  One-shot federated graph learning generally aims to train Graph Neural Networks (GNNs) across clients with disconnected subgraphs in a single communication round.<n>Under extreme non-IID conditions, local GNN training suffers from severe cross-client representation misalignment.<n>We propose SPEAR, a training-free framework that directly computes topology-smoothed class prototypes from local graphs.
  arXiv  Detail & Related papers  (2026-09-05T15:55:22Z)
• RIDE: An Open Dataset and Benchmark for Train Delay Prediction [1.7793982512455193]
  RIDE is an open dataset and benchmark for train delay prediction built at nationwide scale over the Belgian railway network.<n>It covers 94.5M train events, 3.6M journeys, and 35.7M weather records from 2023 to 2025.<n>It is organized as a layered data pipeline from raw railway and weather sources to two public releases: a reusable intermediate relational dataset and model-ready benchmark datasets.
  arXiv  Detail & Related papers  (2026-06-03T16:29:50Z)
• How Faithful Is Trajectory-Based Data Attribution? Error Sources, Remedies, and Practical Guidelines [21.57090069950487]
  Trajectory-based data attribution methods estimate the influence of training samples on model predictions by unrolling the training trajectory.<n>There is a lack of comprehensive error analysis of these methods, raising concerns about method faithfulness and hindering reliable deployment.<n>We provide the first systematic analysis of error sources in trajectory-based data attribution, together with concrete remedies to mitigate them and practical guidelines for downstream use.
  arXiv  Detail & Related papers  (2026-05-12T09:50:45Z)
• Prescriptive Scaling Reveals the Evolution of Language Model Capabilities [22.14002750185524]
  We estimate capability boundaries, high conditional quantiles of benchmark scores as a function of log pre training FLOPs.<n>We validate the temporal reliability by fitting on earlier model generations and evaluating on later releases.<n>We introduce an efficient algorithm that recovers near full data frontiers using roughly 20% of evaluation budget.
  arXiv  Detail & Related papers  (2026-02-17T03:13:51Z)
• Revisiting the Scaling Properties of Downstream Metrics in Large Language Model Training [11.179110411255708]
  We propose a direct framework to model the scaling of benchmark performance from the training budget.<n>Our results show that the direct approach extrapolates better than the previously proposed two-stage procedure.<n>We release the complete set of pretraining losses and downstream evaluation results.
  arXiv  Detail & Related papers  (2025-12-09T18:33:48Z)
• Beyond Scaling: Measuring and Predicting the Upper Bound of Knowledge Retention in Language Model Pre-Training [68.94373533768501]
  We model knowledge retention, the capacity of a pre-trained language model to memorize factual information from its corpus, and introduce a principled method to estimate it prior to training.<n>We propose Size-dependent Mutual Information (SMI), an information-theoretic predictor that integrates knowledge frequency, knowledge specificity, and model size to forecast closed-book question answering (QA) accuracy.
  arXiv  Detail & Related papers  (2025-02-06T13:23:53Z)
• Rejection via Learning Density Ratios [50.91522897152437]
  Classification with rejection emerges as a learning paradigm which allows models to abstain from making predictions.<n>We propose a different distributional perspective, where we seek to find an idealized data distribution which maximizes a pretrained model's performance.<n>Our framework is tested empirically over clean and noisy datasets.
  arXiv  Detail & Related papers  (2024-05-29T01:32:17Z)
• DE$^3$-BERT: Distance-Enhanced Early Exiting for BERT based on Prototypical Networks [43.967626080432275]
  We propose a novel Distance-Enhanced Early Exiting framework for BERT (DE$3$-BERT) We implement a hybrid exiting strategy that supplements classic entropy-based local information with distance-based global information. Experiments on the GLUE benchmark demonstrate that DE$3$-BERT consistently outperforms state-of-the-art models.
  arXiv  Detail & Related papers  (2024-02-03T15:51:17Z)
• Towards Motion Forecasting with Real-World Perception Inputs: Are End-to-End Approaches Competitive? [93.10694819127608]
  We propose a unified evaluation pipeline for forecasting methods with real-world perception inputs. Our in-depth study uncovers a substantial performance gap when transitioning from curated to perception-based data.
  arXiv  Detail & Related papers  (2023-06-15T17:03:14Z)
• How to Estimate Model Transferability of Pre-Trained Speech Models? [84.11085139766108]
  "Score-based assessment" framework for estimating transferability of pre-trained speech models. We leverage upon two representation theories, Bayesian likelihood estimation and optimal transport, to generate rank scores for the PSM candidates. Our framework efficiently computes transferability scores without actual fine-tuning of candidate models or layers.
  arXiv  Detail & Related papers  (2023-06-01T04:52:26Z)
• Learning Sample Difficulty from Pre-trained Models for Reliable Prediction [55.77136037458667]
  We propose to utilize large-scale pre-trained models to guide downstream model training with sample difficulty-aware entropy regularization. We simultaneously improve accuracy and uncertainty calibration across challenging benchmarks.
  arXiv  Detail & Related papers  (2023-04-20T07:29:23Z)
• Surrogate uncertainty estimation for your time series forecasting black-box: learn when to trust [2.0393477576774752]
  Our research introduces a method for uncertainty estimation. It enhances any base regression model with reasonable uncertainty estimates. Using various time-series forecasting data, we found that our surrogate model-based technique delivers significantly more accurate confidence intervals.
  arXiv  Detail & Related papers  (2023-02-06T14:52:56Z)
• Leveraging Unlabeled Data to Predict Out-of-Distribution Performance [63.740181251997306]
  Real-world machine learning deployments are characterized by mismatches between the source (training) and target (test) distributions. In this work, we investigate methods for predicting the target domain accuracy using only labeled source data and unlabeled target data. We propose Average Thresholded Confidence (ATC), a practical method that learns a threshold on the model's confidence, predicting accuracy as the fraction of unlabeled examples.
  arXiv  Detail & Related papers  (2022-01-11T23:01:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.