AIBerkeley AI Researchabout 2 months ago
K-Search Transfers CUDA Kernel Expertise to Apple Silicon
10 min read
Berkeley AI Research extended its K-Search framework with a CUDA-to-MLX translation layer, enabling automatic transfer of GPU kernel optimization expertise to Apple Silicon. The approach achieves near-expert performance, reaching 0.97x speedup versus native MLX Attention and up to 20x prefill speedup on the Mamba SSM kernel.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyApple