[QNN-EP] Implement file mapped weights feature#26952
Merged
edgchen1 merged 19 commits intomicrosoft:mainfrom Jan 27, 2026
Merged
[QNN-EP] Implement file mapped weights feature#26952edgchen1 merged 19 commits intomicrosoft:mainfrom
edgchen1 merged 19 commits intomicrosoft:mainfrom
Conversation
Member
|
The observation is not entirely true. ORT memory map external weights. You have an ability to request a weight as an ORT Value from the EP. If the weight is external it will be memory mapped. See |
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
onnxruntime/core/providers/qnn/builder/qnn_file_mapping_callback_interface.h
Outdated
Show resolved
Hide resolved
Member
|
I suggest not to implement QNN specific mapping, but re-use code in ORT. |
Member
|
Discussed offline, the EP maps initializers form the binary context, not from the external weights files. |
added 3 commits
January 14, 2026 13:56
- Create file mapping callback interface class - Android expected to have support in the future - Implement Windows callbacks in WindowsFileMapper - New option disable_file_mapped_weights - Feature is enabled by default with retry logic
quic-calvnguy
force-pushed
the
dev/calvnguy/file_mapped_weights
branch
from
f55dc78 to
2e451ae
Compare
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
Member
|
Please, avoid force pushes. |
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
yuslepukhin
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
Contributor
|
/azp run Linux QNN CI Pipeline,Windows ARM64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
Member
|
Please, comment on all Copilot review issues before resolving them. |
edgchen1
reviewed
Jan 15, 2026
onnxruntime/core/providers/qnn/builder/qnn_file_mapping_callback_interface.h
Outdated
Show resolved
Hide resolved
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
Contributor
|
/azp run Linux QNN CI Pipeline,Windows ARM64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
edgchen1
reviewed
Jan 24, 2026
QnnBackendManager::SetupBackend if file mapping is not available
edgchen1
reviewed
Jan 24, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
onnxruntime/core/providers/qnn/builder/qnn_file_mapping_interface.h
Outdated
Show resolved
Hide resolved
Contributor
|
/azp run Linux QNN CI Pipeline,Windows ARM64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
yuslepukhin
reviewed
Jan 24, 2026
onnxruntime/core/providers/qnn/builder/qnn_windows_file_mapper.cc
Outdated
Show resolved
Hide resolved
yuslepukhin
reviewed
Jan 24, 2026
Make file mapping callbacks more thread safe Do not destruct file_mapper_ until session destruction
Contributor
|
/azp run Linux QNN CI Pipeline,Windows ARM64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
edgchen1
reviewed
Jan 26, 2026
and unnecessary functions relating to file_mapped_weights_enabled_
Contributor
|
/azp run Linux QNN CI Pipeline,Windows ARM64 QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows GPU Doc Gen CI Pipeline |
|
Azure Pipelines successfully started running 4 pipeline(s). |
edgchen1
approved these changes
Jan 26, 2026
tianleiwu
pushed a commit
that referenced
this pull request
Jan 27, 2026
Description Enables the file mapping of weights as well as the overall context bin. This feature is currently only enabled for ARM64 WIN devices Motivation and Context Currently, when reading the context bin, ORT allocates a large buffer on the heap. Assuming the same model is used, each ORT session will allocate a buffer for the context bin. This is incredibly wasteful when large models are used. Instead, WIN file mapping can be leveraged to map the context bin, then every time a context needs to be created with the context bin, the pointer to the context bin can be retrieved and used instead of some pre-allocated buffer, thus making QNN EP more memory-efficient. In the case of multiple ORT sessions, the context bin will only be loaded once for all sessions, increasing memory efficiency and overall initialization performance. This is very useful regarding the use of LLMs going forward. --------- Co-authored-by: quic_calvnguy <quic_calvnguy@quic_inc.com>
tianleiwu
added a commit
that referenced
this pull request
Jan 27, 2026
| Commit | Commit Title | Author | | :--- | :--- | :--- | | `11dde2d9e` | [NV TensorRT RTX EP] Fix external tensorrt_plugins load path (#26814) | keshavv27 | | `080d96818` | Move model compatibility checks ahead of session initialization (#27037) | adrastogi | | `ec4f6bfa1` | [QNN EP] Fix error messages being logged as VERBOSE instead of ERROR (#24931) | Copilot | | `0432e7125` | perftest: support plugin eps for compile_ep_context (#27121) | Jaskaran Singh Nagi | | `727db0d3d` | Engine compatibility validity API implementation (#26774) | umangb-09 | | `27013522f` | Deprecate transformers model examples (#27156) | Jambay Kinley | | `f83d4d06e` | [QNN-EP] Implement file mapped weights feature (#26952) | quic-calvnguy | --------- Co-authored-by: keshavv27 <165012837+keshavv27@users.noreply.github.com> Co-authored-by: adrastogi <aditya.rastogi@microsoft.com> Co-authored-by: Aditya Rastogi <adityar@ntdev.microsoft.com> Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com> Co-authored-by: vraspar <51386888+vraspar@users.noreply.github.com> Co-authored-by: yuslepukhin <11303988+yuslepukhin@users.noreply.github.com> Co-authored-by: Dmitri Smirnov <yuslepukhin@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Dmitri Smirnov <dmitrism@microsoft.com> Co-authored-by: Jaskaran Singh Nagi <jaskaran.singh.nagi@intel.com> Co-authored-by: umangb-09 <umangb@nvidia.com> Co-authored-by: Jambay Kinley <jambaykinley@microsoft.com> Co-authored-by: quic-calvnguy <quic_calvnguy@quicinc.com> Co-authored-by: quic_calvnguy <quic_calvnguy@quic_inc.com>
tianleiwu
added
cherry-picked
milpuz01
pushed a commit
to milpuz01/onnxruntime
that referenced
this pull request
Feb 4, 2026
Description Enables the file mapping of weights as well as the overall context bin. This feature is currently only enabled for ARM64 WIN devices Motivation and Context Currently, when reading the context bin, ORT allocates a large buffer on the heap. Assuming the same model is used, each ORT session will allocate a buffer for the context bin. This is incredibly wasteful when large models are used. Instead, WIN file mapping can be leveraged to map the context bin, then every time a context needs to be created with the context bin, the pointer to the context bin can be retrieved and used instead of some pre-allocated buffer, thus making QNN EP more memory-efficient. In the case of multiple ORT sessions, the context bin will only be loaded once for all sessions, increasing memory efficiency and overall initialization performance. This is very useful regarding the use of LLMs going forward. --------- Co-authored-by: quic_calvnguy <quic_calvnguy@quic_inc.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Enables the file mapping of weights as well as the overall context bin. This feature is currently only enabled for ARM64 WIN devices
Motivation and Context
Currently, when reading the context bin, ORT allocates a large buffer on the heap. Assuming the same model is used, each ORT session will allocate a buffer for the context bin. This is incredibly wasteful when large models are used. Instead, WIN file mapping can be leveraged to map the context bin, then every time a context needs to be created with the context bin, the pointer to the context bin can be retrieved and used instead of some pre-allocated buffer, thus making QNN EP more memory-efficient. In the case of multiple ORT sessions, the context bin will only be loaded once for all sessions, increasing memory efficiency and overall initialization performance. This is very useful regarding the use of LLMs going forward.