Benchmark · LLMs · GPU databases
DataKernelBench
Can LLMs optimize database queries on GPUs?
A benchmark for translating SQL-derived PyTorch programs into optimized CUDA and Triton kernels, evaluated for correctness and execution speed.
GPU-native data systems
Data processing,
down to the kernel.
KernelDF is a home for research and open-source systems that use AI to generate, validate, and optimize GPU kernels for real data workloads.
Projects