NorthstarAurora/strix-halo-bench-data

该数据集包含在AMD Strix Halo硬件（具体为Ryzen AI MAX+ 395处理器、Radeon 8060S集成显卡/gfx1151架构，配备128 GB统一内存）上运行的大型语言模型（LLM）推理基准测试的复现数据。数据集中提供了来自生产基准测试运行的原始llama-bench和llama-server输出日志，旨在作为论文、博客文章和论坛讨论中引用Strix Halo推理性能数据的稳定引用目标。数据集内容涵盖多个基准测试比较，包括ROCm与Vulkan后端性能对比、不同构建标志的敏感性分析，以及针对特定模型（如Qwen3.6系列）的基准测试结果。数据集还详细记录了硬件配置、软件环境、方法论（如运行命令、重复测试、环境变量设置）和使用指南，便于复现和引用。This dataset contains reproduction data for Large Language Model (LLM) inference benchmarks run on AMD Strix Halo hardware, specifically the Ryzen AI MAX+ 395 processor, Radeon 8060S integrated GPU with gfx1151 architecture, and 128 GB of unified memory. The dataset provides raw output logs of llama-bench and llama-server from production benchmark runs, serving as a stable citation target for referencing Strix Halo inference performance data in academic papers, blog posts, and forum discussions. The dataset covers multiple benchmark comparisons, including performance comparisons between ROCm and Vulkan backends, sensitivity analysis of different build flags, and benchmark results for specific models such as the Qwen3.6 series. It also comprehensively documents hardware configurations, software environments, methodologies (e.g., execution commands, repeated test runs, environment variable settings) and usage guidelines to facilitate reproduction and proper citation.