Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 22 additions & 0 deletions docs/helion-hackathon.md
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,28 @@ Each problem directory (e.g. `causal_conv1d_py/`) contains:
- `task.py` -- type definitions (`input_t`, `output_t`)
- `task.yml` -- input shapes, test cases, and benchmark configs

## Testing Locally

You can test and benchmark your submissions locally on your own GPU without submitting to KernelBot. This is useful for fast iteration during development.

From the `reference-kernels/problems/helion` directory, run:

```bash
# Correctness test (validates your kernel via CUDA graph capture)
python eval.py test causal_conv1d_py/

# Benchmark (measures kernel performance with L2 cache flushing)
python eval.py benchmark causal_conv1d_py/

# Both test and benchmark in one go
python eval.py both causal_conv1d_py/

# Profile (generates PyTorch profiler trace)
python eval.py profile causal_conv1d_py/
```

Replace `causal_conv1d_py/` with any problem directory.

## Writing a Helion Submission

Your submission must be a single Python file that defines `custom_kernel(data: input_t) -> output_t`. To use Helion, write a `@helion.kernel` decorated function and call it from `custom_kernel`.
Expand Down
Loading