Add generic fx-graph serializer (NFXG)#21032
Conversation
Summary: Serialize a torch.fx graph to a standalone flatbuffer (native_graph.fbs, id "NFXG") and back: a topological node list with a tagged Argument union, tensor metadata side table, and by-reference constants (ConstantRef.fqn) whose bytes are shipped separately. Covers input/output classification, symbolic dims in list args, and a validate_graph() completeness check. Constants are never inlined. Reviewed By: SS-JIA Differential Revision: D111753288
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21032
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New FailuresAs of commit 065febf with merge base 8077bb6 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@metascroy has exported this pull request. If you are a Meta employee, you can view the originating Diff in D111753288. |
SS-JIA
left a comment
There was a problem hiding this comment.
Review automatically exported from Phabricator review in Meta.
This PR needs a
|
Summary:
Serialize a torch.fx graph to a standalone flatbuffer (native_graph.fbs, id
"NFXG") and back: a topological node list with a tagged Argument union, tensor
metadata side table, and by-reference constants (ConstantRef.fqn) whose bytes are
shipped separately. Covers input/output classification, symbolic dims in list
args, and a validate_graph() completeness check. Constants are never inlined.
Reviewed By: SS-JIA
Differential Revision: D111753288