philipturner/metal-flash-attention

FlashAttention (Metal Port)

/ 100

Emerging

This project helps machine learning engineers efficiently train large language models on Apple Silicon. It takes model architecture details and training data, performing the complex 'attention' calculations at the core of these models. The output is a significantly faster training process, especially for the backward pass, tailored for Apple's M-series chips.

589 stars. No commits in the last 6 months.

Use this if you are developing or training large AI models and need to maximize the performance of attention mechanisms on Apple Silicon (M1, M2, M3, M4 chips) to reduce training time and memory usage.

Not ideal if you are working with AI models on non-Apple hardware (like NVIDIA GPUs) or if your model doesn't heavily rely on the attention mechanism.

AI-model-training large-language-models Apple-Silicon-optimization deep-learning-performance transformer-architectures

Stale 6m No Package No Dependents

Maintenance 0 / 25

Adoption 10 / 25

Maturity 16 / 25

Community 14 / 25

How are scores calculated?

Stars

589

Forks

Language

Swift

License

MIT

Higher-rated alternatives

triton-inference-server/server

The Triton Inference Server provides an optimized cloud and edge inferencing solution.

gpu-mode/Triton-Puzzles

Puzzles for learning Triton

hailo-ai/hailo_model_zoo

The Hailo Model Zoo includes pre-trained models and a full building and evaluation environment

open-mmlab/mmdeploy

OpenMMLab Model Deployment Framework

hyperai/tvm-cn

TVM Documentation in Chinese Simplified / TVM 中文文档

Explore ML Frameworks

All categories Trending ML Framework directory Insights