Description
Speeding up inference and training usually means hand-written kernel fusion or heavy code changes. MagiCompiler is a plug-and-play AI compiler delivering free-lunch optimizations for inference and training.
It integrates with minimal changes for LLMs, multimodal and video generation models.
Plug and play:Minimal changes.
Inference:Faster automatically.
Training:Also faster.
Models:Language, multimodal and video.
It integrates with minimal changes for LLMs, multimodal and video generation models.
Features
Plug and play:Minimal changes.
Inference:Faster automatically.
Training:Also faster.
Models:Language, multimodal and video.
