Buba Shrimp

README excerpt

## Reference Kernels

This repo holds reference kernels for the KernelBot which hosts regular competitions on [discord.gg/gpumode](discord.gg/gpumode).

You can see what's going on [gpumode.com](https://www.gpumode.com/)

## Competitions
1. [PMPP practice problems](https://github.com/gpu-mode/reference-kernels/tree/main/problems/pmpp_v2)
2. [AMD $100K kernel competition](problems/amd)
3. [BioML kernels](problems/bioml)
4. [AMD $100K distributed kernel competition](problems/amd_distributed)
5. [NVIDIA Blackwell NVFP4 competition](problems/nvidia)
6. [AMD $1.1M competition](problems/amd_202602)
7. [Helion IRL hackathon](problems/helion)
8. [Linear Algebra Problems](problems/linalg)

We also work with universities on hosting the infrastructure for their classes:
- [Stanford CS149 assignment 5 kernels](https://github.com/stanford-cs149/asst5-kernels)
- [Tri Dao's Princeton parallel programming class](problems/princeton)

## Contributing New Problems

To add a new problem, create a new folder in the `problems/glory` directory where you need to add the following files:
- `reference.py` - This is the PyTorch reference implementation of the problem.
- `task.yml` - This is the problem specification that will be used to generate test cases for different shapes
- `task.py` - Specifies the schema of the inputs and outputs for the problem

For NVIDIA problems, follow [Adding Nsight Compute support](docs/ncu-profiling.md)
to implement the evaluator profile path and verify captured kernels