🛠 Layr Labs' Laguna MLX Fast Challenge
A competition has been launched to optimize the inference of the Poolside Laguna XS 2.1 model on Apple Silicon. Participants must optimize the Swift runtime and Metal kernels to accelerate prefix and sequential decoding on M5 Max chips.
🌍 It stimulates the development of libraries for Apple Silicon, pushing the boundaries of performance through custom Metal kernels and the optimization of MoE (Mixture-of-Experts) architectures.
👤 An opportunity to participate in the race and test the local execution speed of modern LLMs on Mac via the MLX framework.
Source 1: https://mlx.fast/ Source 2: https://github.com/Layr-Labs/mlxfast-challenge