ProductInductor
PyTorch Inductor
The default compiler backend of PyTorch, responsible for compiling computational graphs into efficient low-level code, with support for CPU and GPU code generation and memory optimization.
Timeline (last 90 days)
Oct 19
On PyTorch Inductor, 22 of the 54 CPU configurations had actual peak memory deviate from the predicted value, with 9 of those configurations even performing worse than the original program.
Unverified50%