H Company 发布 Holo2-235B-A22B 预览版模型，在 UI 本地化方面取得领先
H Company's new Holo2 model takes the lead in UI Localization
H Company 发布了 Holo2-235B-A22B 预览版模型，该模型在 Screenspot-Pro 和 OSWorld G 上分别达到了 78.5% 和 79.0% 的 SOTA 记录。该模型专注于 UI 元素本地化，并使用 SkyPilot 进行大规模训练。
原文给出了能力变化和开放入口，读者可以据此判断它会怎样改变现有工作流。
Two months since releasing our first batch of Holo2 models, H Company is back with our largest UI localization model yet: Holo2-235B-A22B Preview. This model achieves a new State-of-the-Art (SOTA) record of 78.5% on Screenspot-Pro and 79.0% on OSWorld G.
Available on Hugging Face, Holo2-235B-A22B Preview is a research release focused on UI element localization.
Agentic Localization
High-resolution 4K interfaces are challenging for localization models. Small UI elements can be difficult to pinpoint on a large display. With agentic localization, however, Holo2 can iteratively refine its predictions, improving accuracy with each step and unlocking 10-20% relative gains across all Holo2 model sizes.
Holo2-235B-A22B's Performance on ScreenSpot-Pro
Holo2-235B-A22B Preview reaches 70.6% accuracy on ScreenSpot-Pro in a single step. In agent mode, it achieves 78.5% within 3 steps, setting a new state-of-the-art on the most challenging GUI grounding benchmark.
Trained with SkyPilot
Training Holo2 models at scale requires coordinating workloads across multiple cloud providers. H Company uses SkyPilot as a unified interface for launching training jobs on our clusters with Kubernetes (k8s). By abstracting away infrastructure complexity, SkyPilot lets researchers focus on model development instead of managing k8s manifests or maintaining separate deployment scripts.
来源：Hugging Face Blog · huggingface.co