![[2510.03574] Efficient Test-Time Scaling for Small Vision-Language Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.03574] Efficient Test-Time Scaling for Small Vision-Language Models
Summary
The paper presents efficient test-time scaling strategies for small vision-language models (VLMs) to enhance their performance without compromising computational efficiency.
Why It Matters
As the demand for efficient AI models grows, this research addresses the limitations of small VLMs, which often struggle with generalization and performance. By proposing novel test-time scaling techniques, the study offers a pathway to improve these models' effectiveness in resource-constrained environments, making them more viable for practical applications.
Key Takeaways
- Introduces two efficient strategies: Test-Time Augmentation (TTAug) and Test-Time Adaptation (TTAdapt).
- Demonstrates consistent performance improvements across nine benchmarks.
- Maintains computational efficiency suitable for resource-constrained environments.
- Shows generality across different model scales and VLMs without additional tuning.
- Addresses the trade-off between model size and performance effectively.
Computer Science > Machine Learning arXiv:2510.03574 (cs) [Submitted on 3 Oct 2025 (v1), last revised 16 Feb 2026 (this version, v2)] Title:Efficient Test-Time Scaling for Small Vision-Language Models Authors:Mehmet Onurcan Kaya, Desmond Elliott, Dim P. Papadopoulos View a PDF of the paper titled Efficient Test-Time Scaling for Small Vision-Language Models, by Mehmet Onurcan Kaya and 2 other authors View PDF Abstract:Small Vision-Language Models (VLMs) provide a computationally efficient alternative to larger models, at the cost of weaker generalization abilities and downstream task performance. These shortcomings could be addressed by test-time scaling techniques, but existing methods are typically computationally demanding, contradicting the resource-efficient design goals of small models. To address these limitations, we propose two novel and efficient test-time scaling strategies that leverage the model-internal features rather than external supervision: (i) Test-Time Augmentation (TTAug), which generates multiple augmented inputs and aggregates outputs at the token level without parameter updates, and (ii) Test-Time Adaptation (TTAdapt), which adapts model parameters during inference using consensus-based pseudolabels from TTAug. Through extensive experiments across nine benchmarks, we demonstrate consistent performance improvements while maintaining computational efficiency suitable for resource-constrained environments. The generality of our approach is demonstrated...