RightNow-AI / AutoMegaKernel
An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode, paper: https://arxiv.org/abs/2606.09682
Why this job: works in the areas we cover, but does not say which job it does
What it does
AMK's auto-tuned int8 (W8A16, near-lossless) megakernel beats CUDA-graphed cuBLAS bf16 at batch-1 decode across NVIDIA's datacenter inference-class GPUs, found autonomously by AMK's own search and correctness-gated (argmax-exact). Ratio = cuBLAS / AMK, so > 1 means AMK is faster.
Get started
No single install command. Read the steps in the README (opens in new tab)
- Maintenance
- Active
- Last commit 4 months ago
- Stars
- 163
- +2 this week
- Licence
- MIT
- Allows business use
- Downloads for
- None listed
- No release files
- Latest version
- None yet
- No releases
- Built with
- Python
- 82% of the code
Is it looked after?
- Contributors
- 0
- Last 90 days
- Issues
- 0% closed
- 2 still open
- Pull requests
- 0 merged
- 2 waiting
- Releases
- None
- Never published
- Contributing guide (opens in new tab)
- Security policy (opens in new tab)
- No code of conduct
Similar repos
| Repo | Stars | Last commit | Licence | Downloads for |
|---|---|---|---|---|
| NVIDIA / cuml | 5,321 | 6 days ago | Apache-2.0 | No downloads |
| areal-project / AReaL | 5,907 | yesterday | Apache-2.0 | No downloads |
| HPMLL / BurstGPT | 295 | 4 months ago | CC-BY-4.0 | No downloads |
More: languages, topics and project record
Languages
- Python 82%
- Cuda 15%
- C++ 1%
- C 1%
Topics
- agent-harness
- cuda
- gpu
- gpu-programming
- kernel-fusion
- llm-inference
- machine-learning
- megakernel
- mlsys
Project record
- JobOther AI tools
- StatusListed
- Commits22
- Forks16
- Watching3
- Stars in 28 days+11
- CreatedJun 8, 2026
- First seen by usNov 7, 2026
- CheckedNov 7, 2026