strict_fusion¶

class torch.jit.strict_fusion[source][source]¶

Give errors if not all nodes have been fused in inference, or symbolically differentiated in training.

Example: Forcing fusion of additions.

@torch.jit.script
def foo(x):
    with torch.jit.strict_fusion():
        return x + x + x

Docs