triton-cuda-examples-torch

What problem does it solve?

The collection provides complete Triton CUDA kernel examples integrated with PyTorch, enabling developers to implement and experiment with high-performance GPU kernels within PyTorch models.

Core Features & Use Cases

• Vectorized Triton kernels for common ops such as vector_add, matmul, layer_norm, and softmax, plus multi-kernel workflows.
• Demonstrates how to wrap Triton kernels inside PyTorch models using patterns akin to torch.autograd.Function for custom ops.
• Use cases include research prototyping, performance tuning of kernels, and education for learning Triton integration with PyTorch.

Quick Start

Run the provided PyTorch + Triton examples to study and integrate high-performance CUDA kernels into your models.