YuvrajSingh9886/jetson-non-reasoning-benchmark-25w
收藏资源简介:
这是一个针对Jetson Orin Nano 8GB设备的Tiny LLM基准测试数据集,用于评估小规模语言模型在边缘AI场景下的推理性能。数据集包含多种模型(如gemma3-1b、lfm2.5-1.2b、qwen2.5-0.5b等)在不同量化配置(如Q4_K_M、Q8_0)下的性能指标,测试参数包括输入token长度(如128、512、1024、2048)和生成token长度(如64、128、256)。性能指标包括时间到首次token(TTFT)、token到token延迟(T2T)、吞吐量(Tok/s)、功率消耗(W)以及能效(Tok/J)。数据集基于llama.cpp CUDA后端运行,旨在为边缘设备上的LLM推理提供基准参考。
This is a Tiny LLM benchmark dataset for Jetson Orin Nano 8GB devices, designed to evaluate the inference performance of small-scale language models in edge AI scenarios. The dataset includes performance metrics for various models (e.g., gemma3-1b, lfm2.5-1.2b, qwen2.5-0.5b) under different quantization configurations (e.g., Q4_K_M, Q8_0), with test parameters covering input token lengths (e.g., 128, 512, 1024, 2048) and generation token lengths (e.g., 64, 128, 256). Performance metrics include time to first token (TTFT), token-to-token latency (T2T), throughput (Tok/s), power consumption (W), and energy efficiency (Tok/J). The dataset is based on the llama.cpp CUDA backend and aims to provide benchmark references for LLM inference on edge devices.




