Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit9f5153b

Browse files
yushangdipytorchmergebot
authored andcommitted
Preserve GrpahModule node stack trace after torch package deserializaion re-tracing (#155638)
Summary:urrently the node.meta["stack_trace"] is not preserved when we torch package/load GraphModule, which means the original stack trace is lost. When we re-trace the packaged graph module, we just get a stack trace like fx-generated._0......Adding the node.meta["stack_trace"] to torch packaged graph moduleTest Plan:```buck2 run @//mode/dev-nosan fbcode//caffe2/test:package -- -r TestPackageFX```Rollback Plan:Differential Revision: D76379692Pull Requestresolved:#155638Approved by:https://github.com/angelayi
1 parentce9ba07 commit9f5153b

File tree

2 files changed

+39
-0
lines changed

2 files changed

+39
-0
lines changed

‎test/package/test_package_fx.py‎

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -187,6 +187,27 @@ def forward(self, a):
187187
input=torch.rand(2,3)
188188
self.assertEqual(loaded_traced(input),traced(input))
189189

190+
deftest_package_gm_preserve_stack_trace(self):
191+
classSimpleTest(torch.nn.Module):
192+
defforward(self,x):
193+
returntorch.relu(x+3.0)
194+
195+
st=SimpleTest()
196+
traced=symbolic_trace(st)
197+
198+
fornodeintraced.graph.nodes:
199+
node.meta["stack_trace"]=f"test_{node.name}"
200+
201+
f=BytesIO()
202+
withPackageExporter(f)aspe:
203+
pe.save_pickle("model","model.pkl",traced)
204+
205+
f.seek(0)
206+
pi=PackageImporter(f)
207+
loaded_traced=pi.load_pickle("model","model.pkl")
208+
fornodeinloaded_traced.graph.nodes:
209+
self.assertEqual(f"test_{node.name}",node.meta["stack_trace"])
210+
190211

191212
if__name__=="__main__":
192213
run_tests()

‎torch/fx/graph_module.py‎

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -204,6 +204,14 @@ def is_leaf_module(self, _: torch.nn.Module, __: str) -> bool:
204204
tracer_extras=body.get("_tracer_extras", {})
205205
graph=KeepModules().trace(com,**tracer_extras)
206206

207+
# Recover node.meta["stack_trace"] after re-tracing
208+
node_meta_stack_trace=body.get("_graphmodule_graph_node_meta_stack_trace",None)
209+
ifnode_meta_stack_traceisnotNone:
210+
delbody["_graphmodule_graph_node_meta_stack_trace"]
211+
fornodeingraph.nodes:
212+
ifnode_meta_stack_trace.get(node.name,None)isnotNone:
213+
node.meta["stack_trace"]=node_meta_stack_trace[node.name]
214+
207215
# Manually set Tracer class on the reconstructed Graph, to avoid
208216
# referencing the private local subclass KeepModules.
209217
graph._tracer_cls=tracer_cls
@@ -859,6 +867,16 @@ def __reduce_package__(self, exporter: PackageExporter):
859867
dict_without_graph["_graphmodule_cls_name"]=self.__class__.__name__
860868
deldict_without_graph["_graph"]
861869

870+
# Store node.meta["stack_trace"] so we can recover them after re-tracing during deserialization
871+
node_meta_stack_trace= {
872+
node.name:node.meta["stack_trace"]
873+
fornodeinself.graph.nodes
874+
if"stack_trace"innode.meta
875+
}
876+
dict_without_graph[
877+
"_graphmodule_graph_node_meta_stack_trace"
878+
]=node_meta_stack_trace
879+
862880
generated_module_name=f"fx-generated._{exporter.get_unique_id()}"
863881
python_code=self.recompile()
864882
import_block=_format_import_block(python_code.globals,exporter.importer)

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp