fastmachinelearning/wa-hls4ml

wa-hls4ml Benchmark Dataset是一个用于硬件（FPGA）上实现的神经网络资源利用率和延迟估计的替代模型训练的基准数据集。数据集包含超过68万个完全合成的数据流模型，旨在继续扩展和扩展数据集。数据集包括从机器学习模型到HLS表示再到寄存器传输级（RTL）的所有综合链步骤，并保存完整的日志。它分为训练集、验证集和测试集，每个集包含不同类型的神经网络模型和参数。数据集还包括来自真实科学应用的示例模型，用于测试替代模型在新配置和架构中的预测能力。数据集的生成使用hls4ml工具，并提供了详细的JSON文件格式描述。

The wa-hls4ml Benchmark Dataset is a benchmark dataset for training surrogate models to estimate resource utilization and latency of neural networks implemented on hardware (FPGAs). The dataset consists of over 680,000 fully synthesized dataflow models, and aims to continue expanding over time. It includes all steps of the synthesis chain from ML model to HLS representation to register-transfer level (RTL) and saves the full logs. The dataset is split into training, validation, and test sets, each containing different types of neural network models and parameters. It also includes exemplar models from real scientific applications to test the surrogate models prediction capabilities in new configurations and architectures. The dataset generation uses the hls4ml tool and provides a detailed description of the JSON file format.