REVIEW 20 cited by
Quantifying Variance in Evaluation Benchmarks
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
Quantifying Variance in Evaluation Benchmarks
abstract
Evaluation benchmarks are the cornerstone of measuring capabilities of large language models (LLMs), as well as driving progress in said capabilities. Originally designed to make claims about capabilities (or lack thereof) in fully pretrained models, evaluation benchmarks are now also extensively used to decide between various training choices. Despite this widespread usage, we rarely quantify the variance in our evaluation benchmarks, which dictates whether differences in performance are meaningful. Here, we define and measure a range of metrics geared towards measuring variance in evaluation benchmarks, including seed variance across initialisations, and monotonicity during training. By studying a large number of models -- both openly available and pretrained from scratch -- we provide empirical estimates for a variety of variance metrics, with considerations and recommendations for practitioners. We also evaluate the utility and tradeoffs of continuous versus discrete performance measures and explore options for better understanding and reducing this variance. We find that simple changes, such as framing choice tasks (like MMLU) as completion tasks, can often reduce variance for smaller scale ($\sim$7B) models, while more involved methods inspired from human testing literature (such as item analysis and item response theory) struggle to meaningfully reduce variance. Overall, our work provides insights into variance in evaluation benchmarks, suggests LM-specific techniques to reduce variance, and more generally encourages practitioners to carefully factor in variance when comparing models.
Forward citations
Cited by 20 Pith papers
-
DataComp-VLM: Improved Open Datasets for Vision-Language Models
DataComp-VLM benchmark shows instruction-heavy data mixing outperforms filtering for VLM training, with DCVLM-Baseline achieving 63.6% on 33 tasks for 8B models (+5.4pp over FineVision).
-
Fluid Language Model Benchmarking
Fluid Benchmarking, combining IRT-based ability estimation with Fisher-information-based adaptive item selection, improves LM evaluation across efficiency, validity, variance, and saturation in pretraining settings.
-
How Much of the Routing Gap Is Real? Decomposing the Router-to-Oracle Gap into Reproducible Specialist Advantage and Single-Draw Label Noise
Under stochastic decoding, 12–36% of the LLM router-to-oracle gap is single-draw noise unreachable by any single-commit router, while the rest is recoverable specialist advantage; best-of-K sampling closes the floor.
-
The Harder Text Embedding Benchmark (HTEB): Beyond One-dimensional Static Robustness
HTEB introduces dynamic, multi-axis evaluation of text embedding robustness using LLM transformations, finding decoupled profiles across models and that scaling does not close all robustness gaps.
-
The Art of Scaling Reinforcement Learning Compute for LLMs
A 400k+ GPU-hour study shows RL scaling in LLMs follows predictable sigmoidal trajectories, with most design choices affecting efficiency rather than the performance asymptote, enabling accurate large-scale prediction...
-
Messier: A High-Resolution Corpus for Cross-Benchmark Agent Evaluation
A unified corpus of 957k trial outcomes shows frontier progress is uneven and strict all-pass aggregation obscures capability and can reorder agents.
-
DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks
Original, never-upstreamed multi-file engineering tasks with functional verifiers grade coding agents more faithfully and separate frontier models more widely than inherited-test SWE benchmarks.
-
DataComp-VLM: Improved Open Datasets for Vision-Language Models
DataComp-VLM benchmark shows instruction-heavy data mixtures outperform caption-heavy ones for VLM training, with DCVLM-Baseline reaching 63.6% on 33 tasks using 200B tokens, +5.4pp over FineVision.
-
Bounded Difference Concentration for Infinitely Exchangeable Sequences with Applications to AI Benchmark Uncertainty
Establishes concentration bounds for infinitely exchangeable sequences with cancellation for zero-sum contrasts and applies the result to distribution-free uncertainty quantification in composite AI benchmarks.
-
Validity Threats for Foundation Model Research
Maps common low-compute research strategies for foundation models onto statistical, internal, external, and construct validity threats via a causal-inference lens.
-
Resolution Diagnostics for Paired LLM Evaluation
Paired LLM leaderboard comparisons frequently lack resolution at conventional (alpha=0.05, power=0.8) levels, with a new per-pair ratio q=N/N* showing that common unpaired shortcuts underestimate required samples by r...
-
Beyond Fixed Benchmarks and Worst-Case Attacks: Dynamic Boundary Evaluation for Language Models
Dynamic Boundary Evaluation locates each LLM's performance boundary at ~50% pass probability via a calibrated item bank and Skill-Guided Boundary Search algorithm to enable unified, adaptive evaluations across safety,...
-
Beyond Fixed Benchmarks and Worst-Case Attacks: Dynamic Boundary Evaluation for Language Models
Dynamic Boundary Evaluation adaptively identifies each LLM's performance boundary on a shared difficulty scale using a calibrated item bank and a search algorithm.
-
A Study of LLMs' Preferences for Libraries and Programming Languages
Empirical study of eight LLMs finds overuse of popular libraries like NumPy in up to 45% of unnecessary cases and strong default preference for Python even when suboptimal.
-
LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws
Pretraining data determines loss-to-loss scaling laws in LLMs, while model size, optimization, tokenizer, and architecture have limited impact.
-
Laplace-PSN-IRT: Uncertainty Quantification for Neural Item Response Theory Models of LLM Benchmarks
Scoped last-layer Laplace posteriors for PSN-IRT yield calibrated ability/difficulty uncertainty, showing 64 of 66 model pairs are statistically indistinguishable and that posterior-averaged Fisher information survive...
-
AdaRoPE: Not All Attention Heads Should Rotate and Scale Equally
Head-wise learnable rotary frequencies and length-dependent attention scaling (AdaRoPE) beat uniform RoPE and YaRN schedules in pretraining and 8k-to-64k context extension up to 8B scale.
-
A Tale of Two Variances: When Single-Seed Benchmarks Fail in Bayesian Deep Learning
Single-seed CRPS estimates in limited-data BDL show high variance and peaks for heteroscedastic methods, with local variance correlating above 0.96 to single-seed error.
-
ReasonBENCH: Benchmarking the (In)Stability of LLM Reasoning
LLM reasoning benchmark scores vary substantially across repeated runs under the same model, strategy, and task, so single-run evaluation can misrank systems.
-
Correcting Prompt Dependence in LLM Benchmarks: A Bayesian Hierarchical Model with Embedding-Space Clustering
A Bayesian model that groups similar LLM test prompts into clusters gives better predictive scores than a no-clustering baseline but does not prove that it truly corrects prompt dependence.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.