# ELS-RD/kernl

Kernl lets you run PyTorch transformer models several times faster on GPU with a single line of code, and is designed to be easily hackable.

Repository: https://github.com/ELS-RD/kernl
Canonical: https://ross.abutalabs.com/products/kernl
Homepage: http://www.kernl.ai
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Topics: cuda, cuda-kernel, pytorch, triton, transformer
Archived: true
Last push: 2026-01-28T15:12:32+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 64, release rhythm 8, longevity 100
- inputs: {"age_days": 1489, "days_push": 217, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1584, forks 98 (observed 2026-08-28T04:05:07.367884+00:00)

## What it is
Kernl is an open-source PyTorch model optimization library that accelerates transformer inference on GPUs with a single line of code, using OpenAI Triton kernels, CUDA Graphs, and TorchDynamo. It is now archived, with its Triton debugger upstreamed to the official Triton repository.

## Use cases
- speed up pytorch transformer inference on gpu
- optimize huggingface models without rewriting them
- learn how flash attention works with tutorials
- write and debug custom triton gpu kernels
- avoid tensorrt or onnx export for model acceleration

## When to choose
- you want one-line GPU acceleration for PyTorch transformer models
- you want readable, hackable Triton kernels instead of CUDA C
- you want educational tutorials on matmul, online softmax, and flash attention

## When to avoid
- you need actively maintained software - the project is archived
- you need production inference serving beyond transformer optimization
- you do not have an NVIDIA GPU

## Facets
- artifact type: library
- maturity: abandoned
- function: machine-learning, llm-inference, gpu-computing, benchmarking
- domain: machine-learning, deep-learning, gpu-computing, developer-tools
- platform: python
- tags: triton, cuda-kernels, pytorch-optimization, flash-attention, kernel-fusion, archived, gpu, linux

## Member repositories
- ELS-RD/kernl (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:07.367884+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:55:53.102390+00:00, confidence not recorded.
  - readme: https://github.com/ELS-RD/kernl (fetched 2026-08-28T04:05:07.367884+00:00, sha fadc8bfed8ca)
  - homepage: http://www.kernl.ai (fetched 2026-08-29T11:26:23.527343+00:00, sha 2fea8d6943db)
- Data as of 2026-08-30T08:39:29.467469+00:00.