# xlite-dev/LeetCUDA

Modern CUDA Learn Notes with PyTorch for Beginners, 200+ CUDA Kernels, Tensor Cores, HGEMM, FA-2 MMA.

Repository: https://github.com/xlite-dev/LeetCUDA
Canonical: https://ross.abutalabs.com/products/leetcuda
Homepage: https://github.com/xlite-dev/LeetCUDA
Language: Cuda
License: GPL-3.0
License Family: copyleft
Topics: cuda, hgemm, cuda-kernels, cuda-toolkit, flash-attention, cuda-demo, learn-cuda, leet-cuda, cuda-kernel, cuda-library, cuda-12, cuda-cpp
Last push: 2026-08-17T03:15:23+00:00

## Health v2 (maintenance only)
Score: 93/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 98, release rhythm 86, longevity 96
- inputs: {"age_days": 1355, "days_push": 16, "days_rel": 16, "gap_med": 4, "n_releases_24m": 62}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 11836, forks 1244 (observed 2026-08-28T04:10:50.138463+00:00)

## What it is
A collection of modern CUDA learning notes with PyTorch integration for beginners, featuring 200+ CUDA kernels, 100+ LLM/CUDA blogs, and high-performance implementations like HGEMM (98-100% of cuBLAS TFLOPS) and Flash Attention-2 using Tensor Cores with pure MMA PTX. It serves as both an educational resource and a reference for GPU kernel optimization techniques.

## Use cases
- learn CUDA programming from scratch with PyTorch integration
- study high-performance HGEMM kernel implementations
- understand Flash Attention-2 with Tensor Core MMA PTX
- find reference CUDA kernel implementations for common operations
- prepare for CUDA/GPU kernel engineering interviews
- benchmark custom CUDA kernels against cuBLAS
- learn about TF32, FP16, BF16, and FP8 tensor core usage

## When to choose
- you are a beginner wanting structured CUDA learning materials with runnable PyTorch examples
- you need reference implementations of optimized kernels like HGEMM or Flash Attention
- you are preparing for GPU kernel engineering interviews
- you want to study Tensor Core programming with MMA PTX instructions

## When to avoid
- you need a production-ready kernel library for deployment rather than educational notes
- you need a general-purpose CUDA math library like cuBLAS or CUTLASS
- you work on non-NVIDIA GPU platforms
- you need stable, versioned APIs with long-term support guarantees

## Facets
- artifact type: learning-resource
- maturity: active
- function: gpu-computing, developer-tools, benchmarking, machine-learning
- domain: gpu-computing, developer-tools, tutorials, machine-learning, deep-learning
- platform: python
- tags: cuda-kernels, tensor-cores, hgemm, flash-attention, ptx, pytorch, cuda-cpp, awesome-lists, linux, cuda, gpu

## Member repositories
- xlite-dev/LeetCUDA (main) score 93

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:50.138463+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 2026-08-29T17:15:17.479079+00:00, confidence not recorded.
  - readme: https://github.com/xlite-dev/LeetCUDA (fetched 2026-08-28T04:10:50.138463+00:00, sha ec3ee4f9bad9)
  - homepage: https://github.com/xlite-dev/LeetCUDA (fetched 2026-08-29T08:13:09.578299+00:00, sha f2ea7085c0c0)
- Data as of 2026-08-30T08:39:29.467469+00:00.