pub fn cpu_allocator() -> Arc<dyn ExecutionProvider>Expand description
The shared CPU execution provider as an ExecutionProvider trait object.
Exposed so callers building a Tensor from borrowed host memory (e.g.
the Python binding’s zero-copy DLPack import) can supply the allocator
Tensor::from_borrowed_parts_with_guard requires. Because a borrowed
buffer is never actually freed by the EP, any CPU provider suffices.