onnxruntime
Annotation based partitioning along with resource accounting
#27595
Merged
Merged
Annotation based partitioning along with resource accounting #27595
tianleiwu
merged
63 commits into
main
from
yuslepukhin/layering
Introduce annotation query function
18b5d2bf
Wire annotations to partitioning interface.
1eeba1a1
Fix up annotations with Transpose Optimizer
debc8dd2
Add ORT_EXTENDED_MINIMAL build
8f0ff86e
Move rules and matcher inside the index
f7a422ed
Add Update with tests
1ef4078c
TODO: Consider not removing annotations
828eca39
Clear annotations after partitioning
1626d5c0
Merge branch 'main' into yuslepukhin/layering
f33bab1a
Address accountant bug
b3ecb39f
Annotate tiny_gpt2_beamsearch by layers
b5ea1c6e
Refactor Graph_GetGraphView to make it a utility
9a422bae
Introduce a graph utility to create an IndexedSubgraph
a1caf934
Merge branch 'main' into yuslepukhin/layering
e1b1c4f8
Fix lint in python script.
acec4024
Merge branch 'main' into yuslepukhin/layering
31dd7a84
Merge branch 'main' into yuslepukhin/layering
9fa48492
Merge branch 'main' into yuslepukhin/layering
50c58c90
Fix build errors and address Copilt comments
e445b608
Reject duplicate rules
358f7df6
Move methods to .cc
653fb8b1
Remove code duplication
23a8ecfa
Add missing include
ef1227e5
Fix matching bug
b0b23966
Change index parsing
b9e13cfa
Remove wrong comment
add0227e
Address minimal build issues
17e35254
Fix unused arg
1b1a7dbb
adrianlizarraga
commented
on 2026-03-13
Add logging
88c2c479
Make sure the annotation is copied on node copy
9b0b5298
Adjust error message
dab76bc0
Copy Annotations when copying nodes and inlining functions
b39a487d
Update LayeringIndex after function inlining
4e260bce
Add intermediate buffers accounting + temp coefficient
52143507
Merge branch 'main' into yuslepukhin/layering
3fb1d1e3
tianleiwu
commented
on 2026-03-24
Address MakeNodeUnassigned feedback
cd73b56d
Address InlineNodes feedback
dfe4d139
Fix underaccounting for shared weights in fused nodes
62f3d114
yuslepukhin
changed the title
tianleiwu
commented
on 2026-03-25
Update onnxruntime/python/tools/layering/layer_annotate.py
3cab988f
github-actions
commented
on 2026-03-26
Lint
e6cb75f1
Flip = prefix to exact match
b2ef9a22
Adjust comments for duplicate annotations
fde63000
Remove bad comment
7871afa0
Adjust EpWithNoLayeringRulesSeesAllUnassignedNodes
16ec921b
Throw on multiple annotations
4da5c3b4
Make sure annotations are propagated on function inlining
01e4506d
Update include/onnxruntime/core/session/onnxruntime_session_options_c…
3e52f145
Update onnxruntime/core/framework/graph_partitioner.cc
24a46e8c
Update onnxruntime/core/graph/graph_utils.h
e745e02a
Fix issues in python
59b5ccd1
Address undercounting problem
054f8941
Add copyright header
09967c3f
Update onnxruntime/core/framework/graph_partitioner.cc
d23ee080
Adjust doc and implementaton for fetching layering ann
9e8be7a4
Make GetContainingGraph public
f636138d
Adjust accounting for fused node and remove stray local var
fcda5245
Address flaky test
2da1394a
Update onnxruntime/core/providers/cuda/cuda_execution_provider.cc
c0c5e519
Update onnxruntime/core/graph/graph_utils.cc
c55bb6ec
Address review issues
cdd9faa8
Fix potential perf issue
67a947b4
Address review comments
44c69046
tianleiwu
commented
on 2026-03-27
tianleiwu
commented
on 2026-03-27
Add documentation for ann and ep propagation. Fix L1 optimizers, add …
927a0efa
tianleiwu
approved these changes
on 2026-03-30
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub