mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158055 Approved by: https://github.com/AlannaBurke, https://github.com/anijain2305 ghstack dependencies: #157985 Co-authored-by: Svetlana Karslioglu <svekars@meta.com>
867 B
867 B
Working with Graph Breaks
As you might remember from (Dynamo Core Concepts)[programming_model.dynamo_core_concepts] that Dynamo performs a graph break when
it encounters code that can't be traced. In the default torch.compile settings, Dynamo compiles the FX graph
that has been determined up to that point, executes the unsupported code in regular Python, and then resumes tracing.
Graph breaks enable Dynamo to trace through arbitrary Python code and carve out functional subgraphs that can each be individually optimized.
However, graph breaks may cause unexpected slowness in torch.compile.
If you're not seeing the expected speedups, we recommend checking for graph breaks and removing them.
The following sections outline strategies for addressing graph breaks.
programming_model.fullgraph_true
programming_model.common_graph_breaks