kernel-requirements.md

Kernel requirements

Kernels on the Hub must fulfill the requirements outlined on this page. You can use kernel-builder to build conforming kernels.

Directory layout

A kernel repository on the Hub must contain a build directory. This directory contains build variants of a kernel in the form of directories following the template <framework><version>-cxx<abiver>-<cu><cudaver>-<arch>-<os>. For example build/torch26-cxx98-cu118-x86_64-linux. The currently recommended build variants are:

• torch25-cxx11-cu118-x86_64-linux
• torch25-cxx11-cu121-x86_64-linux
• torch25-cxx11-cu124-x86_64-linux
• torch25-cxx98-cu118-x86_64-linux
• torch25-cxx98-cu121-x86_64-linux
• torch25-cxx98-cu124-x86_64-linux
• torch26-cxx11-cu118-x86_64-linux
• torch26-cxx11-cu124-x86_64-linux
• torch26-cxx11-cu126-x86_64-linux
• torch26-cxx98-cu118-x86_64-linux
• torch26-cxx98-cu124-x86_64-linux
• torch26-cxx98-cu126-x86_64-linux

This list will be updated as new PyTorch versions are released. Kernels that are in pure Python (e.g. Triton kernels) only need to provide a single build variant:

• torch-universal

Each variant directory should contain a single directory with the same name as the repository (replacing - by _). For instance, kernels in the kernels-community/activation repository have a directories like build/<variant>/activation. This directory must be a Python package with an __init__.py file.

Versioning

Kernels are versioned on the Hub using Git tags. Version tags must be of the form v<major>.<minor>.<patch>. Versions are used by locking to resolve the version constraints.

Native Python module

Kernels will typically contain a native Python module with precompiled compute kernels and bindings. This module must fulfill the following requirements:

• Use ABI3/Limited API for compatibility with Python 3.9 and later.

• Compatible with manylinux_2_28. This means that the extension must not use symbols versions higher than:

  • GLIBC 2.28
  • GLIBCXX 3.4.24
  • CXXABI 1.3.11
  • GCC 7.0.0

  These requirement can be checked with the ABI checker (see below).

• No dynamic library dependencies outside:

  • Torch;
  • CUDA/ROCm libraries installed as dependencies of Torch.

The manylinux_2_28 and Python ABI 3.9 version requirements can be checked with kernel-abi-check:

$ cargo install kernel-abi-check
$ kernel-abi-check result/relu/_relu_e87e0ca_dirty.abi3.so
🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9
✅ No compatibility issues found

Torch extension

Torch native extension functions must be registered in torch.ops.<namespace>. Since we allow loading of multiple versions of a module in the same Python process, namespace must be unique for each version of a kernel. Failing to do so will create clashes when different versions of the same kernel are loaded. Two suggested ways of doing this are:

• Appending a truncated SHA-1 hash of the git commit that the kernel was built from to the name of the extension.
• Appending random material to the name of the extension.

Note: we recommend against appending a version number or git tag. Version numbers are typically not bumped on each commit, so users might use two different commits that happen to have the same version number. Git tags are not stable, so they do not provide a good way of guaranteeing uniqueness of the namespace.

Layers

A kernel can provide layers in addition to kernel functions. A layer from the Hub can replace the forward method of an existing layer for a certain device type. This makes it possible to provide more performant kernels for existing layers. See the layers documentation for more information on how to use layers.

Writing layers

To make the extension of layers safe, the layers must fulfill the following requirements:

• The layers are subclasses of torch.nn.Module.
• The layers are pure, meaning that they do not have their own state. This means that:
  • The layer must not define its own constructor.
  • The layer must not use class variables.
• No other methods must be defined than forward.
• The forward method has a signature that is compatible with the forward method that it is extending.

This is an example of a pure layer:

class SiluAndMul(nn.Module):
    def forward(self, x: torch.Tensor):
        d = x.shape[-1] // 2
        output_shape = x.shape[:-1] + (d,)
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
        ops.silu_and_mul(out, x)
        return out

For some layers, the forward method has to use state from the adopting class. In these cases, we recommend to use type annotations to indicate what member variables are expected. For instance:

class LlamaRMSNorm(nn.Module):
    weight: torch.Tensor
    variance_epsilon: float

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        return rms_norm_fn(
            hidden_states,
            self.weight,
            bias=None,
            residual=None,
            eps=self.variance_epsilon,
            dropout_p=0.0,
            prenorm=False,
            residual_in_fp32=False,
        )

This layer expects the adopting layer to have weight and variance_epsilon member variables and uses them in the forward method.

Exporting layers

To accommodate portable loading, layers must be defined in the main __init__.py file. For example:

from . import layers

__all__ = [
  # ...
  "layers"
  # ...
]

Python requirements

• Python code must be compatible with Python 3.9 and later.

• All Python code imports from the kernel itself must be relative. So, for instance if in the example kernel example, module_b needs a function from module_a, import as:

  from .module_a import foo

  Never use:

  # DO NOT DO THIS!
  
  from example.module_a import foo

  The latter would import from the module example that is in Python's global module dict. However, since we allow loading multiple versions of a module, we uniquely name the module.

• Only modules from the Python standard library, Torch, or the kernel itself can be imported.