sidnarsipur/onnx-inference

该数据集来源于ONNX-Predict项目，用于收集ONNX模型推理性能测量数据，基于ONNX模型库中托管的模型。它包含2300多个ONNX模型的推理数据，每个模型被优化为四个变体，并在16台具有不同CPU核心数、内存大小、缓存配置和CPU提供商的机器上收集推理测量。数据集每行结合了三个特征组：图特征（描述ONNX模型结构和估计的数据移动）、硬件特征（描述用于推理的机器）和推理测量（描述观察到的运行时性能）。具体特征包括133个图特征（127个节点计数特征和6个数据移动与计算估计特征）、9个硬件特征和4个推理测量列（平均延迟、标准差延迟、最小延迟和最大延迟，均以毫秒为单位）。图特征涵盖卷积操作、矩阵乘法操作、元素操作、归约和池化操作、归一化操作、数据移动操作等。硬件特征包括内存大小、缓存配置、CPU提供商等。数据集基于MIT许可证发布。This dataset is derived from ONNX-Predict, a dataset of ONNX model inference performance measurements collected from models hosted on the ONNX Model Zoo. It contains inference data for 2,300+ ONNX models. Each model was optimized into four variants, and inference measurements were collected across 16 machines with different CPU core counts, memory sizes, cache configurations, and CPU providers. Each row in the dataset combines three groups of features: graph features describing the ONNX model structure and estimated data movement, hardware features describing the machine used for inference, and inference measurements describing observed runtime performance. The dataset includes 133 graph features (127 node-count features and 6 data movement and compute estimates), 9 hardware features, and 4 inference columns (average_ms, stddev_ms, min_ms, max_ms, all in milliseconds). Graph features cover convolution operations, matrix multiplication operations, elementwise operations, reduction and pooling operations, normalization operations, data movement operations, and others. Hardware features include memory_mb, cache configurations, cpu_provider, etc. The dataset is released under the MIT License.