analiza aiqbench
This paper argues that common evaluations of visual token pruning in vision-language models, which rely on aggregate accuracy and retained token counts, can miss whether a method selects meaningful tokens and whether it delivers real speedup. It proposes PAA-Bench, a plug-and-play framework with new metrics (Random-Pass@k and Preservation-Constrained Speedup). For readers building or deploying efficient VLMs, the key takeaway is that pruning rankings can shift once selection difficulty and end-to-end cost are measured directly, so headline efficiency claims deserve scrutiny.
Are We Evaluating VLM Token Pruning Right? Looking Beyond Accuracy and Token Count
arXiv:2610.24740v1 Announce Type: new Abstract: Visual token pruning aims to reduce inference cost while preserving the capabilities of the original VLM. Yet existing evaluations rely on aggregate task accuracy and retained token count as proxies for pruning quality and actual efficiency. We reveal two limitations: 1) Aggregate accuracy can obscure whether a pruning method truly identifies critical visual tokens; many examples remain solvable under arbitrary token removal and models can exploit textual shortcuts when visual evidence is degraded. 2) Retained token count is an unreliable proxy for efficiency, as identical token budgets can incur different end-to-end costs depending on where and how pruning is performed. Consequently, comparisons based on aggregate accuracy and retained token counts fail to recover the global performance-efficiency Pareto frontier across methods. We introduce PAA-Bench, a plug-and-play evaluation framework for existing benchmarks, to evaluate whether pruning succeeds through meaningful token selection and whether this success translates into real speedup. To determine when token choice actually matters, we propose Random-Pass@k to measure how often an example remains correct under random pruning, enabling comprehensive evaluation across different levels of selection difficulty. To compare efficiency fairly, we introduce Preservation-Constrained Speedup (PCS) to report the maximum end-to-end speedup each method achieves under a shared preservation target. Experiments across 8 methods, 3 VLM backbones, and 9 benchmarks show that stronger token compression does not necessarily yield greater speedup, method differences widen on examples where token selection matters, and efficiency rankings change with the preservation target.
Title:Are We Evaluating VLM Token Pruning Right? Looking Beyond Accuracy and Token Count
Abstract:Visual token pruning aims to reduce inference cost while preserving the capabilities of the original VLM. Yet existing evaluations rely on aggregate task accuracy and retained token count as proxies for pruning quality and actual efficiency. We reveal two limitations: 1) Aggregate accuracy can obscure whether a pruning method truly identifies critical visual tokens; many examples remain solvable under arbitrary token removal and models can exploit textual shortcuts when visual evidence is degraded. 2) Retained token count is an unreliable proxy for efficiency, as identical token budgets can incur different end-to-end costs depending on where and how pruning is performed. Consequently, comparisons based on aggregate accuracy and retained token counts fail to recover the global performance-efficiency Pareto frontier across methods. We introduce PAA-Bench, a plug-and-play evaluation framework for existing benchmarks, to evaluate whether pruning succeeds through meaningful token selection and whether this success translates into real speedup. To determine when token choice actually matters, we propose Random-Pass@k to measure how often an example remains correct under random pruning, enabling comprehensive evaluation across different levels of selection difficulty. To compare efficiency fairly, we introduce Preservation-Constrained Speedup (PCS) to report the maximum end-to-end speedup each method achieves under a shared preservation target. Experiments across 8 methods, 3 VLM backbones, and 9 benchmarks show that stronger token compression does not necessarily yield greater speedup, method differences widen on examples where token selection matters, and efficiency rankings change with the preservation target.
Artificial Intelligence (cs.AI)
arXiv:2610.24740 [cs.AI]
(or arXiv:2610.24740v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.24740 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 343 cuvinte
Subiecte conexe
When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs
arXiv · 1 sept. 2026
In open RLVR, “improvement” depends on the instrument — a small GRPO testbed separating what training optimizes, measures, and teaches
LessWrong · 15 iun. 2026
Evaluating multi-agent systems for explainability and helpfulness with Amazon Bedrock AgentCore
AWS · 5 oct. 2026
Beyond Accuracy: Counterfactual Fragility and Demographic Bias in Clinical Evaluation of LLMs
arXiv · 29 sept. 2026