From KTIR to SpyreCode: An MLIR Dataflow Backend for IBM Spyre
PyTorch Blog1 min read
The Torch-spyre Inductor framework enables efficient kernel scheduling and tensor memory management between on-chip scratchpad and off-chip memory on IBM Spyre hardware. The backend compiler leverages MLIR to optimize dataflow execution, improving performance for PyTorch workloads on Spyre's architecture.
AI summary