carlosqsw/ONNX-Bench
收藏Hugging Face2025-10-12 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/carlosqsw/ONNX-Bench
下载链接
链接失效反馈官方服务：
资源简介：
ONNX-Bench是一个包含649,596个神经网络架构的大型基准数据集，这些架构采用统一的ONNX格式，并伴有性能标签。数据集跨越多个NAS搜索空间，支持在CIFAR-10上进行一致性评估，并具有丰富的架构多样性，包括基于单元和分层的架构。数据集经过预简化处理，适合用于性能预测、零样本迁移和通用代理模型训练。
ONNX-Bench is a large-scale benchmark of 649,596 neural network architectures represented in a unified ONNX format with performance labels. It spans multiple NAS search spaces, supports consistent evaluation on CIFAR-10, and features rich architectural diversity including cell-based and hierarchical structures. The dataset is pre-simplified for stable parsing and is suitable for performance prediction, zero-shot transfer, and universal surrogate model training.
提供机构：
carlosqsw
