cdreetz/

triton-sft-dataset

This dataset contains 5958 examples for supervised fine-tuning (SFT) of models to generate Triton GPU kernels.

The dataset consists of two types of examples:

Synthetic queries (60%): Generated queries asking for Triton kernels for various operations Convert queries (40%): PyTorch code conversion requests to Triton kernels

Each example contains:

prompt: The instruction/query asking for a Triton… See the full description on the dataset page: https://huggingface.co/datasets/cdreetz/triton-sft-dataset.