YuvrajSingh9886/bonsai-jetson-benchmark-15w
收藏资源简介:
该数据集是一个基准测试结果数据集,用于评估Bonsai系列模型(包括Bonsai-1.7B、Bonsai-4B、Bonsai-8B、Ternary-Bonsai-1.7B、Ternary-Bonsai-4B和Ternary-Bonsai-8B)在NVIDIA Jetson Orin Nano Super 8GB平台上的推理性能。测试在15W功率模式下进行,使用llama.cpp后端和CUDA加速,通过不同提示词长度(256、512、1024、2048个词元)和生成词数(128、256、512个词元)的组合进行性能扫描。数据集包含多项性能指标,如平均首次词元时间(TTFT)、词元间延迟(ITL)、词元每秒(Tok/s)、功耗(Power)、每焦耳词元数(Tok/J)以及温度数据。数据集旨在为边缘AI设备上的模型推理效率和能耗提供详细基准测试结果。
This dataset is a benchmarking results dataset that evaluates the inference performance of the Bonsai model series (including Bonsai-1.7B, Bonsai-4B, Bonsai-8B, Ternary-Bonsai-1.7B, Ternary-Bonsai-4B, and Ternary-Bonsai-8B) on the NVIDIA Jetson Orin Nano Super 8GB platform. The tests were conducted in a 15W power mode using the llama.cpp backend with CUDA acceleration, sweeping through combinations of different prompt token lengths (256, 512, 1024, 2048 tokens) and generation token counts (128, 256, 512 tokens). The dataset includes various performance metrics such as average Time to First Token (TTFT), Inter-Token Latency (ITL), tokens per second (Tok/s), power consumption (Power), tokens per joule (Tok/J), and thermal data. The dataset aims to provide detailed benchmarking results for model inference efficiency and energy consumption on edge AI devices.




