helion

library1.1.0pypypiunverified

A Python-embedded DSL for writing high-performance ML kernels that compile to efficient GPU code. Version 1.1.0, actively developed under PyTorch. Monthly releases.

ai-ml devops

pip install helion

github ↗homepage ↗

INSTALL

IMPORT

SIG · HELION

helion

ai-mlpythonv1.1.0

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

helion

✓ import helion

✗ import helion

Basic kernel definition using helion's DSL, launched with CUDA grid syntax.

import helion import torch @helion.kernel def add_kernel(a: helion.Tensor[float, 1], b: helion.Tensor[float, 1], c: helion.Tensor[float, 1]): i = helion.thread_id() c[i] = a[i] + b[i] x = torch.randn(1024, device='cuda') y = torch.randn(1024, device='cuda') z = torch.zeros(1024, device='cuda') add_kernel[(1024,)](x, y, z) print(z.sum())

Debug

Known issues

breakinghelion.kernel decorator requires the function to be a standalone function, not a method or lambda.

fix

Define kernels as top-level functions only.

affects: >=1.0.0

deprecatedUsing helion.Tensor without specifying dimensions (e.g., helion.Tensor[float]) is deprecated; always specify exact shape or broadcast.

fix

Use helion.Tensor[float, 1] for 1D, etc.

affects: >=1.1.0

gotchaGarbage collection of GPU tensors inside kernel loops can cause silent memory leaks; always manage tensor scopes explicitly.

fix

Avoid creating temporary helion.Tensor inside kernel loops; preallocate and reuse.

affects: >=1.0.0

gotchaKernel launch grid dimensions must be a tuple of ints or a single int; using list or other iterable raises a TypeError.

fix

Pass grid as (N,) or (N, M) tuple; not [N].

affects: >=1.0.0

breakingIn version 1.0.0, the 'hello' module was renamed to 'helion'; old imports break.

fix

Update 'import hello' to 'import helion'.

affects: <1.0.0

Upgrade

Version history

1.1.0latest on PyPI · released May 20, 2026

Audit

Dependencies

torchrequiredRequired for kernel compilation and integration with PyTorch.

numpyrequiredUsed for array operations inside kernel code.

Agent activity

30 hits · last 30 days

node

Amazon

Resources

githubhttps://github.com/pytorch/helion ↗

homepagehttps://github.com/pytorch/helion ↗