kernelbot-data

This is the dataset that was created from the first and second AMD $100K kernel competitions, containing roughly 110K kernels for fp8-gemm, moe, mla, all2all, gemm+reducescatter, and allgather+gemm optimized to run on MI300. Learn more at gpumode.com/v2/news

To see the full list of kernel competitions we've ran and are running you can checkout https://github.com/gpu-mode/reference-kernels which also contains details on reference kernels and their input shapes and distributions

We are planning on adding kernels optimized for NVFP4 on Blackwell next

If you use this dataset in your work, please cite:

```bibtex
@inproceedings{
zhang2025kernelbot,
title={KernelBot: A Competition Platform for Writing Heterogeneous {GPU} Code},
author={Alex L Zhang and Matej Sirovatka and Erik Schultheis and Benjamin Horowitz and Mark Saroufim},
booktitle={Championing Open-source DEvelopment in ML Workshop @ ICML25},
year={2025},
url={https://openreview.net/forum?id=bq9U4dmuyJ}
}
```

本数据集源自首届与第二届AMD 10万美元算力核（kernel）竞赛，包含约11万个针对FP8通用矩阵乘法（fp8-gemm）、混合专家模型（MoE）、矩阵注意力（MLA）、全对全通信（all2all）、gemm+归约散射（reducescatter）以及全收集+通用矩阵乘法（allgather+gemm）的优化内核，可在MI300架构GPU上高效运行。更多详情可访问gpumode.com/v2/news。

若需查看我们已举办及正在开展的全部算力核竞赛列表，可前往https://github.com/gpu-mode/reference-kernels，该页面同时包含参考内核及其输入形状与分布的详细说明。

我们计划后续新增针对Blackwell架构的NVFP4优化内核。

若您在研究工作中使用本数据集，请引用如下文献：
bibtex
@inproceedings{
zhang2025kernelbot,
title={KernelBot: A Competition Platform for Writing Heterogeneous {GPU} Code},
author={Alex L Zhang and Matej Sirovatka and Erik Schultheis and Benjamin Horowitz and Mark Saroufim},
booktitle={Championing Open-source DEvelopment in ML Workshop @ ICML25},
year={2025},
url={https://openreview.net/forum?id=bq9U4dmuyJ}
}

搜集汇总

背景与挑战

背景概述

该数据集包含KernelBot竞赛平台上的GPU内核提交，针对不同硬件目标（如AMD MI300、NVIDIA Blackwell等）进行了优化。数据集提供了多个子集和辅助脚本，便于用户分析和使用这些优化的GPU内核代码。