Piper Kernels

Reusable PyTorch inference operators and optimized kernels for the Piper ecosystem and other consumers.

Piper Kernels requires Python 3.14 or newer.

The package owns operator semantics, portable PyTorch references, tensor subclasses, and optimized backends. It deliberately does not know about model repositories, checkpoint metadata, pipeline frameworks, or device-offloading policy.

Planned operators

Package	Role
`piper_kernels.convrot`	ConvRot quantized tensors and linear operators; INT8 today, INT4 planned
`piper_kernels.attention`	Attention operators, including a future SageAttention backend

ConvRot INT8

Quantize a dense weight, or wrap existing checkpoint storage without dequantizing it, then use the resulting tensor as a normal linear weight:

import torch

from piper_kernels.convrot import ConvRotInt8Tensor

weight = ConvRotInt8Tensor.from_hp(dense_weight, group_size=64)
checkpoint_weight = ConvRotInt8Tensor.from_packed(qdata, scale, group_size=64)
output = torch.nn.functional.linear(activation, weight, bias)

# In-place low-rank update with the standard Tensor.addmm_ contract.
weight.addmm_(lora_b, lora_a, alpha=lora_strength)

addmm_ computes weight = beta * weight + alpha * (mat1 @ mat2) and requantizes the result. It preserves the ConvRot tensor and packed storage identities, allowing offload integrations to keep their existing buffers. Repeated updates are lossy, so reload a pristine base weight before changing or removing a previously merged adapter. This is an inference operation and does not support autograd.

The operator selects its Triton implementation on supported CUDA devices and otherwise uses the portable PyTorch reference. Install the tensor format and optimized backend with piper-kernels[convrot,triton]. The base package does not require TorchAO or Triton, so future attention-only consumers do not inherit quantization-specific dependencies.

Dependency direction

Applications such as Piper consume this package. Integrations such as torch-offload may optionally recognize its tensor types, but piper-kernels does not depend on either project.

Development

uv sync --dev
uv run pytest
uv run ruff check .
uv run pyright
uv build

GPU tests use the gpu pytest marker. The pre-commit test hook hides CUDA so commits run the portable suite; run uv run pytest directly to exercise installed GPU backends.

Releases

Releases follow the compatibility and release policy in VERSIONING.md. Distribution artifacts are built from version tags and published to PyPI by GitHub Actions using Trusted Publishing; maintainers do not upload releases from local environments.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

piper_kernels-0.1.0.tar.gz (14.4 kB view details)

Uploaded Aug 4, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

piper_kernels-0.1.0-py3-none-any.whl (18.8 kB view details)

Uploaded Aug 4, 2026 Python 3

File details

Details for the file piper_kernels-0.1.0.tar.gz.

File metadata

Download URL: piper_kernels-0.1.0.tar.gz
Upload date: Aug 4, 2026
Size: 14.4 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for piper_kernels-0.1.0.tar.gz
Algorithm	Hash digest
SHA256	`ed7eb738869353fa08c2e67637000fd5419895f1c2e9b10d21b2d245f68fcd83`
MD5	`e4c47880b60b88b3d866c450a1aac877`
BLAKE2b-256	`f1624cfbff5778ed2939eea0955f54e1b1c4ac2585ebd1465e4756f3c6af4046`

See more details on using hashes here.

Provenance

The following attestation bundles were made for piper_kernels-0.1.0.tar.gz:

Publisher: release.yml on Boffee/piper-kernels

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: piper_kernels-0.1.0.tar.gz
- Subject digest: ed7eb738869353fa08c2e67637000fd5419895f1c2e9b10d21b2d245f68fcd83
- Sigstore transparency entry: 2338178561
- Sigstore integration time: Aug 4, 2026
Source repository:
- Permalink: Boffee/piper-kernels@3354a000c5089e4aa39a0840aa8d81dba546acf1
- Branch / Tag: refs/tags/v0.1.0
- Owner: https://github.com/Boffee
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: release.yml@3354a000c5089e4aa39a0840aa8d81dba546acf1
- Trigger Event: push