Skip to content

LuminousYuzu/KernelFlow

Repository files navigation

KernelFlow

Automated CUDA kernel optimization and CI/CD deployment platform for LLM inference.


Overview

KernelFlow is a production-style MLOps platform built around fused CUDA kernels for LLM inference. A researcher pushes a fused kernel to GitHub — the system automatically builds, validates numerical correctness, benchmarks against an unfused baseline, and deploys it as a PyTorch Extension. No manual steps.

The pipeline mirrors NVIDIA's internal kernel development infrastructure (cuDNN / FlashInfer style): every kernel must clear a defined speedup gate and numerical error budget before it can be promoted to main and packaged.

Kernels implemented:

Milestone Kernel Speedup Gate
1 Fused RMSNorm + RoPE ≥ 1.5× vs baseline
2 Fused SiLU × Elementwise Multiply ≥ 1.3× vs baseline
3 Fused Attention (Flash Attention simplified) ≥ 2.0× vs baseline

Tech Stack

Compute

Build & CI/CD

Quality & Observability


License

Apache 2.0 — consistent with NVIDIA open-source projects (CUTLASS, TensorRT-LLM, cuDF).

About

Automated CUDA kernel optimization platform — fused LLM inference kernels with end-to-end CI/CD: Jenkins, Kubernetes, and GPU benchmarking gates from push to deploy.

Resources

License

Stars

1 star

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors