### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
<!-- Describe your changes. -->
Introduces an ATen fallback for
`torch.nn.functional.scaled_dot_product_attention`. This operator was
introduced in torch 2.0 and, since then, has had many updates including
the implementation of memory efficient attention for V100 machines. The
current torchscript exporter exports a subgraph for attention which does
not provide the same memory savings that PyTorch's memory efficient
attention kernel provides. Allowing fallback to PyTorch ATen op for
attention helps mitigate memory spike issues for models leveraging
memory efficient attention.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Memory issues arose when integrating ONNX Runtime Training with AML
Stable Diffusion.
---------
Co-authored-by: root <prathikrao@microsoft.com>
### Description
<!-- Describe your changes. -->
There is a bug for kernel running on rocm6.0, so change ci docker image
to rocm6.1
For the torch installed in the docker image, change to rocm repo when it
is not 6.0 version.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
There are so many typos reported by the review dog, [Optional Lint]
actions (example:
https://github.com/microsoft/onnxruntime/actions/runs/9864564489/job/27239732367),
this PR is to fix some of them.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
---------
Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com>
### Description
It might be easier if we just directly include the original gsl headers.
"core/common/gsl.h" is an indirection that doesn't provide extra help.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Delete path.h and replace all occurrences of onnxruntime::Path with
std::filesystem::path.
Previously we couldn't use C++17's std::filesystem because it was not
supported in iOS 12(which was released in 2018). Now we dropped the
support for iOS 12.
### Motivation and Context
To simplify code. For example, if an EP wants to use the Path class, now
it can directly use it without going through a wrapper. And the standard
implementation can handle various path types better. (We didn't take
much consideration on UNC path, "/" as a path separator on Windows,
etc).
### Problem
Currently, the codebase contains some logics pertaining to model
re-export checks and graph_builder reinitialization checks. Ideally,
these operations should function akin to a state machine. However, upon
inspecting the implementation, it becomes apparent that certain states
are checked or set in various scattered locations. This fragmentation
makes it challenging to comprehend when a re-export or re-initialization
will be triggered. For optimal clarity and maintainability, it is
advisable to consolidate these states into a cohesive component, rather
than dispersing them within the current graph execution manager.
Furthermore, the process of model exports and post-export processing for
stage 3 support or memory-efficient gradient management introduces
considerable complexity. To enhance the codebase's structure, it would
be beneficial to extract these intricate functionalities into a
dedicated component, divorcing them from the current graph execution
manager.
As part of the effort to improve the codebase, it's essential to address
inconsistencies in handling input/output flatten/unflatten operations.
Currently, there are several functions performing these operations
recursively, each with slightly different implementations. This
inconsistency leads to varying support for input/output data types and
structures in different parts of the code. To rectify this, the proposed
pull request simplifies these operations into a set of primitive
functions, ensuring uniformity. This not only streamlines the code but
also facilitates the maintenance of consistency when introducing bug
fixes or supporting new data types. One thing to mention here: input
output handling is deeply bound to the graph transition mentioned above,
so it is difficult to make this change separately.
While acknowledging the complexity of these logics, it is reassuring
that the codebase benefits from an extensive suite of unit tests that
cover all possible branches. Despite the intricacies, ensuring the
passage of all tests has been a time-intensive but necessary aspect of
this development effort.
### Design
Introduce `GraphTransitionManager` and put all model export and
post-export processing logics in it.
1. Re-export check
2. Do export
3. Re-post-export process check
4. Do post-export process
5. Return `PostExportProcessedModelInfo`, which contains all the
information we need, to pass to ORT to build gradient graph (currently
we do the same for training or evaluating, but ideally we should not do
it for evaluating, let's keep this behavior as it is now, and make the
change later).
```
# Input names for the pre-gradient-build graph.
# This may be different with the one in ExportedGraph since we may
modify the graph inputs as needed
# for example when memory efficient gradient management is enabled.
self.onnx_graph_input_names: list[str] = onnx_graph_input_names
# A subset of onnx_graph_input_names.
# Input names that require gradients for the pre-gradient-build graph.
self.onnx_graph_input_names_require_grad: list[str] =
onnx_graph_input_names_require_grad
# Create symbolic names for each dimension of the graph input (e.g.
onnx_graph_input_names).
# The key is the input name, the value is a dict of {dim_index:
symbolic_dim_name}
# e.g. {"input1": {0: "input1_dim0", 1: "input1_dim1"}, "input2": {0:
"input2_dim0"}}
self.onnx_graph_input_dynamic_axes_map: dict[str, dict[int, str]] =
onnx_graph_input_dynamic_axes_map
self.buffer_for_ort_runs: dict[str, torch.Tensor] = OrderedDict()
self.onnx_graph_input_names_user_defined = (
onnx_graph_input_names_user_defined # The ONNX graph input names
excluding the parameters, buffers.
)
# The ONNX graph input names excluding the parameters, buffers.
self.onnx_graph_input_names_require_grad_user_defined =
onnx_graph_input_names_require_grad_user_defined
self._post_export_processed_model: onnx.ModelProto | None =
post_export_processed_model
# A function to access the input data from the args and kwargs.
# If it is not None, the length is same as onnx_graph_input_names.
# For i-th input name, we can use the i-th function to get the input
data from args and kwargs.
self.data_accessor: list[callable] | None = data_accessor
# Used for unflattening the outputs from the ORT forward run.
self.module_forward_output_schema: ORTModelInputOutputSchemaType | None
= module_forward_output_schema```
The `GraphTransitionManager` instance is a property of
`GraphExecutionManager` (e.g. `TrainingManager` or ``InferenceManager),
1. Use
'self._graph_transition_manager.use_cache_or_reconstruct_post_processed_model(inputs,
kwargs)' to check whether the PyTorch module need a re-export or
re-post-export-process.
2. Use
`self._graph_transition_manager._post_export_processed_model_info.construct_inputs`
to construct the list of inputs used for ORT runs.
3. Use
`self._graph_transition_manager._post_export_processed_model_info.restore_outputs(user_outputs)`
to restore the outputs in original PyTorch output structure.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
1. Update the functions in tensorprotoutils.h to use
std::filesystem::path instead of onnxruntime::Path. Eventually we can
remove the whole onnxruntime::Path class, but to this PR small I am not
doing that.
2. Remove the _SILENCE_EXPERIMENTAL_FILESYSTEM_DEPRECATION_WARNING macro
def when TensorRT EP is enabled.
### Description
Skip softmax BF16 test for ROCm, because BFloat16 is unsupported by
MIOpen, and `torch.cuda.is_available()` also returns `True` for ROCm.
### Description
Alternative design from #20942
Allow users to pass in a model path for the generate_artifacts API.
### Motivation and Context
- ONNX API calls such as the onnx checker + shape inference fail when
given a model > 2GB, but work if a path to a model >2GB is passed in.
### Description
change is_pod tp is_trivial
### Motivation and Context
This is commonnly needed for both linux and win c++20 upgrade.
is_trivial was introduced backed in C++11
### Release backward inputs per static graph ref count
For the output buffer marked as external output:
1. Remove the additional ref count we used for avoiding reusing buffer.
Instead, when we find reuse input/output buffer, we will make sure the
reused buffer not not generated by nodes that has external outputs.
2. Remove the ref count of pybind feed inputs, which exists all the time
until the run_backward completed. Instead, passing a mutuble feeds, and
we clean the feeds vector once that is copied into session states and
not needed any more before run the graph sequencentially.
#### Before the change:
One of the backward inputs is 3.9GB, it lives until the backward ends.
#### With the change:
The 3.9GB is released when the last node depending on that tensor
completed.
Be noted: the peak did not change though, we have more work to do to
reduce on the peak.
#### Others
It is found there are few tests that were updated to use incorrect
expected values in previous code refactoring
a81faee41e (diff-9e8fbae7d3dff24106cd17564949f320e943cb3048eae07813c7de144f140419L382).
This PR tries to fix them back, and I think now all test cases are back
to normal.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Adding bf16 support to GatherElementsGrad.
---------
Co-authored-by: Adam Louly <adamlouly@microsoft.com@h100vm-ort.kxelwkzfzxguje5bxvwxxs135a.gvxx.internal.cloudapp.net>
Adding Bfloat16 to scale op
---------
Co-authored-by: Adam Louly <adamlouly@microsoft.com@h100vm-ort.kxelwkzfzxguje5bxvwxxs135a.gvxx.internal.cloudapp.net>
### Flash attn recompute
1. Allow PythonOp(FlashAttn) can be recomputed correctly.
45879ff5c2
2. Use JSON to pass the selected-to-recompute subgraphs.
3c374da678
#### Better Memory Efficiency
Customer model can run both PyTorch SPDA and Flash Attn, this PR make it
possible to let the Flash Attn path work with ORTModule layerwise
recompute. The peak drop from 45.xGB to 32.xGB if we only compare the
layers (not including other pieces, BTW there are few more optimization
targeting other pieces as well later).
#### Better Perf
Using Flash ATTN bring additionally 16% end to end time reduction, with
highly aligned loss curve.
#### Use JSON File to pass Recompute Plans
To overcome the limitation of max length of the strings defined in
session options.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
The InsertGatherBeforeSceLoss optimization is enabled when the density
of label padding less than 90%. We need to check the density of the
label padding to decide whether enable the optimization.
Before this pr, we just check the inputs of graph and correlate one with
the SCE node by iterate graph from the SCE node back to one graph input.
This is hard to be general because there may be complicated pattern
between graph input and SCE node.
This pr check padding density by the direct input of SCE module rather
than the input of graph at the first graph execution when exporting onnx
graph.
And if the density < 90%, insert a flag PythonOp after the SCE node as:
```
SoftmaxCrossEntropy
|
PythonOp (func_name: FlagAndPrintDensity) (insert if density < 90%)
|
Following graph
```
When the InsertGatherBeforeSceLoss is invoked, it check if there is the
flag PythonOp(func_name: FlagAndPrintDensity) after the SCE node and if
it is, remove it and do the padding elimination optimization.
If the env of ORTMODULE_PRINT_INPUT_DENSITY is 1, we will print input
density each step by the PythonOp (func_name: FlagAndPrintDensity). In
this case the PythonOp will not be removed.
### Improve perf for mem efficient grad mgmt
When memory efficient gradient mangement feature is enabled, the weight
retrieval PythonOp for every layers will be launched at the beginning of
the forward, which would make GPU stream idle for few milliseconds. The
reason is the ReversedDFS ordering cannot ALWAYS handle such input
branching well, so we introduce a distantance-to-input_leaf concepts
when doing the reversedDFS, which not only move the problematical
PythonOp to the place where it is needed, but also those Cast ops
following the weight retrieval to the place where it is needed.
Main branch: 102.19 - 26.35s = 75.84s for 260 steps(4627samples),
61.04sample/second
This PR: 100.28s - 25.10s = 75.18s for 260 steps. 61.54samples/second
(+0.8% gains)
Main branch:
This PR:
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Create numpy arrays based on the native buffers of returned OrtValues.
Hold on to the OrtValue until the numpy array is garbage collected.
### Motivation and Context
This saves cpu on tensor copies and addresses customer concerns.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
<!-- Describe your changes. -->
Fix some misc build warnings from x86 Windows build
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Add bf16 support for below ops:
ConstantOfShape
Exp
Erf
convolution
PythonOp
### Motivation and Context
phimm model works on bf16, ORT need support bf16 on previous ops to work
with phimm on bf16
### Description
In Deepspeed's Pipeline Parallel Implementation, there is a class used
to instantiate the object after it's moved to the device and assigned in
a stage.
This approach helps reduce peak memory usage.
In this PR, we're adding support to ORT for wrapping this LayerSpec.
### Description
<!-- Describe your changes. -->
Fix comparison that was not updated when the threshold was converted to
bytes.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Fix CI failure
### Introduce memory efficient topo sort (for training)
~~and laze initialize Priority-Based and Memory-Efficient topo sort.
Because in most cases, they are not needed, so we free the overheads of
GraphViewer construction for most use cases.~~
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
<!-- Describe your changes. -->
Add ability to store initializer data in an external file.
Update training checkpoint code to use external file if data > ~2GB.
I don't see a way for the flatbuffers 64-bit offsets to be used, as they
don't support storing 'table' types with 64-bit offsets (and our Tensor
is a 'table' type not a simple struct).
0cfb7eb80b/tests/64bit/test_64bit.fbs (L38-L39)
Allowing a Tensor to have its raw_data in an external file should
hopefully work with the least friction. As it's an extra field it's
backwards compatible.
Please feel free to suggest alternative approaches.
Side note: the diffs in the generated *.fbs.h files are unexpectedly
large. Maybe they weren't re-generated when the new flatbuffers version
was checked in. I updated by running:
`python .\compile_schema.py -f <build output
dir>\_deps\flatbuffers-build\Debug\flatc.exe`
from onnxruntime\core\flatbuffers\schema which I thought was the correct
way but maybe that's out of date.
I think you can ignore all the diffs in the generated files and just
worry about the changes to the .fbs files in
onnxruntime/core/flatbuffers/schema. Basically start at the bottom of
the files changed and work up as all the 'real' diffs are there.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
---------
Co-authored-by: carzh <wolfivyaura@gmail.com>
### Description
Introducing a new class ORTPipelineModule to handle wrapping layers in
DeepSpeed pipeline parallel.
### Motivation and Context
To support pipeline parallelism on ORTModule.
This PR will include an initial support of deepspeed Pipeline
parallelism.
- [x] Support Pipeline parallel where layers are nn Modules in
Sequential.
- [ ] Support LayerSpec and TiedLayerSpec
- [ ] Enable partitioning to accept List
- [ ] Full-GPU Graph Consolidation
- [ ] Subgraph Merging for Inference
Previous implementation used numpy array and numpy data_type to store
constant value and data type, which is not support BFloat16 natively.
This PR is to switch to use torch tensor which supports BFloat16.
### Description
Background:
User save large model with initializer data in external file. e.g:
onnx.save_model(onnx_model, "path/to/save/the/model.onnx", save_as_external_data=True, all_tensors_to_one_file=True,
location="filename", size_threshold=1024).
In that case, Ort loads the model, get the external initializer information (external file name, offset, length) and use the model path to find the external file, and locate to the tensor data via the offset and length.
But it won't work if user load the model from memory, since Ort lost track of the model path.
This PR adds API/session option to let user provide a table with external initializer file name as the key, the pointer to the loaded external file in memory and the buffer length as value. So that
1. user can load the model from memory buffer with external initializers in memory buffer too.
2. the initializers can be shared across sessions, for different EPs.
3. user can load the file in any way they want, e.g mmap.
Internally,
1. at session creation time, Ort goes through the external initializers in the graph, gets the file name, offset, data length of the external initializers from Tensorproto .
2. With the file name, Ort get the file in memory buffer and buffer length from the table user provided.
4. Ort locates the tensor buffer from file in memory buffer (user provided) using the offset and data length (from Tensorproto ).
5. Ort creates the Tensor and replace the existing Tensor in the graph.
### Motivation and Context
https://github.com/onnx/onnx/blob/main/docs/ExternalData.md
For a model with external data, the Tensorproto may have initializer data in a separate file. The external file location is set via the file path relative to the model path. With the API to load model from memory buffer, it lost track of the
model path. So it causes error if the model has external data. By adding a session option to set the external data buffer, Ort can find the external data correctly if model loaded from memory buffer.
### Description
- Adds a patch that fixes a shape inference bug that caused a segfault:
https://github.com/onnx/onnx/pull/6080
- Fix documentation describing why QLinearMatMul tests are currently
being skipped.
### Motivation and Context
The [PR for integrating with ONNX
1.16.0](https://github.com/microsoft/onnxruntime/pull/19745) disabled
various python quantization tests due to a shape inference bug. This PR
applies the ONNX fix as a patch. We still can't enable the tests because
some of our CIs pip install onnx-1.16.0, which doesn't include the fix.
### Description
update with ONNX 1.16.0 branch according to
https://github.com/microsoft/onnxruntime/blob/main/docs/How_To_Update_ONNX_Dev_Notes.md
ONNX 1.16.0 release notes:
https://github.com/onnx/onnx/releases/tag/v1.16.0
#### Updated ops for CPU EP:
- DequantizeLinear(21)
- Added int16 and uint16 support + various optimizer tests
- Missing int4 and uint4 support
- Missing block dequantization support
- QuantizeLinear(21)
- Added int16 and uint16 support + various optimizer tests
- Missing int4 and uint4 support
- Missing block quantization support
- Cast(21)
- Missing int4 and uint4 support
- CastLike(21)
- Missing int4 and uint4 support
- ConstantOfShape(21)
- Missing int4 and uint4 support
- Identity(21)
- Missing int4 and uint4 support
- If(21)
- Missing int4 and uint4 support
- Loop(21)
- Missing int4 and uint4 support
- Reshape(21)
- Missing int4 and uint4 support
- Scan(21)
- Missing int4 and uint4 support
- Shape(21)
- Missing int4 and uint4 support
- Size(21)
- Missing int4 and uint4 support
- Flatten(21)
- Missing float8e4m3fnuz, float8e5m2, float8e5m2fnuz, int4, and uint4
support
- Pad(21)
- Missing float8e4m3fnuz, float8e5m2, float8e5m2fnuz, int4, and uint4
support
- Squeeze(21)
- Missing float8e4m3fnuz, float8e5m2, float8e5m2fnuz, int4, and uint4
support
- Transpose(21)
- Missing float8e4m3fnuz, float8e5m2, float8e5m2fnuz, int4, and uint4
support
- Unsqueeze(21)
- Missing float8e4m3fnuz, float8e5m2, float8e5m2fnuz, int4, and uint4
support
#### Unimplemented opset 21 features/ops
- int4 and uint4 data type
- QLinearMatMul(21)
- GroupNormalization(21)
- ai.onnx.ml.TreeEnsemble(5)
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Disabled tests
#### ORT Training
orttraining/orttraining/test/python/orttraining_test_ort_apis_py_bindings.py
- test_ort_custom_ops: Potential shape inference bug for custom ops
#### Python quantization unit tests
test/onnx/python/quantization (shape inference bug)
- test_op_conv_transpose.py: test_quantize_conv_transpose_u8u8_fp16
- test_op_conv_transpose.py: test_quantize_conv_transpose_s8s8_fp16
- test_op_gemm.py: test_quantize_qop_gemm_s8s8
- test_op_gemm.py: test_quantize_qop_gemm_e4m3fn_same
- test_op_gemm.py: test_quantize_qop_gemm_e4m3fn_p3
- test_op_matmul.py: test_quantize_matmul_u8u8_f16
- test_op_matmul.py: test_quantize_matmul_s8s8_f16
- test_op_matmul.py: test_quantize_matmul_s8s8_f16_entropy
- test_op_matmul.py: test_quantize_matmul_s8s8_f16_percentile
- test_op_matmul.py: test_quantize_matmul_s8s8_f16_distribution
- test_op_relu.py: test_quantize_qop_relu_s8s8
#### ONNX tests
- test_maxpool_2d_ceil_output_size_reduce_by_one: ONNX 1.16.0 fixed a
maxpool output size bug and added this test. Enable this test when [ORT
PR](https://github.com/microsoft/onnxruntime/pull/18377) is merged.
Refer to original [ONNX PR](https://github.com/onnx/onnx/pull/5741).
- test_ai_onnx_ml_tree_ensemble_set_membership_cpu: new unimplemented op
ai.onnx.ml.TreeEnsemble
- test_ai_onnx_ml_tree_ensemble_single_tree_cpu: same
- test_ai_onnx_ml_tree_ensemble_set_membership_cuda: same
- test_ai_onnx_ml_tree_ensemble_single_tree_cuda: same
- test_cast_INT4_to_FLOAT_cpu: ORT Cast(21) impl doesn't support int4
yet
- test_cast_INT4_to_INT8_cpu: same
- test_cast_UINT4_to_FLOAT_cpu: same
- test_cast_UINT4_to_UINT8_cpu: same
- test_cast_INT4_to_FLOAT_cuda
- test_cast_INT4_to_INT8_cuda
- test_cast_UINT4_to_FLOAT_cuda
- test_cast_UINT4_to_UINT8_cuda
- test_constantofshape_float_ones_cuda: ConstantOfShape(21) not
implemented for cuda
- test_constantofshape_int_shape_zero_cuda: same
- test_constantofshape_int_zeros_cuda: same
- test_flatten_axis0_cuda: Flatten(21) not implemented for cuda
- test_flatten_axis1_cuda: same
- test_flatten_axis2_cuda: same
- test_flatten_axis3_cuda: same
- test_flatten_default_axis_cuda: same
- test_flatten_negative_axis1_cuda: same
- test_flatten_negative_axis2_cuda: same
- test_flatten_negative_axis3_cuda: same
- test_flatten_negative_axis4_cuda: same
- test_qlinearmatmul_2D_int8_float16_cpu: QLinearMatMul(21) for onnx not
implemented in ORT yet
- test_qlinearmatmul_2D_int8_float32_cpu: same
- test_qlinearmatmul_2D_uint8_float16_cpu: same
- test_qlinearmatmul_2D_uint8_float32_cpu: same
- test_qlinearmatmul_3D_int8_float16_cpu: same
- test_qlinearmatmul_3D_int8_float32_cpu: same
- test_qlinearmatmul_3D_uint8_float16_cpu: same
- test_qlinearmatmul_3D_uint8_float32_cpu: same
- test_qlinearmatmul_2D_int8_float16_cuda: same
- test_qlinearmatmul_2D_int8_float32_cuda: same
- test_qlinearmatmul_2D_uint8_float16_cuda: same
- test_qlinearmatmul_2D_uint8_float32_cuda: same
- test_qlinearmatmul_3D_int8_float16_cuda: same
- test_qlinearmatmul_3D_int8_float32_cuda: same
- test_qlinearmatmul_3D_uint8_float16_cuda: same
- test_qlinearmatmul_3D_uint8_float32_cuda: same
- test_size_cuda: Size(21) not implemented for cuda
- test_size_example_cuda: same
- test_dequantizelinear_blocked: Missing implementation for block
dequant for DequantizeLinear(21)
- test_quantizelinear_blocked_asymmetric: Missing implementation for
block quant for QuantizeLinear(21)
- test_quantizelinear_blocked_symmetric: Missing implementation for
block quant for QuantizeLinear(21)
---------
Signed-off-by: liqunfu <liqun.fu@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Co-authored-by: Ganesan Ramalingam <grama@microsoft.com>
Co-authored-by: George Wu <jywu@microsoft.com>
Co-authored-by: adrianlizarraga <adlizarraga@microsoft.com>
### Description
The PaddingElimination optimization is enabled when the density of
embedding padding less than 90%. We need to check the density of the
embedding padding to decide whether enable the optimization.
Before this pr, we just check the inputs of graph and correlate one with
the embedding node by iterate graph from the embedding node back to one
graph input.
This is hard to be general because there may be complicated pattern
between graph input and embedding node.
This pr check padding density by the direct input of embedding module
rather than the input of graph at the first graph execution when
exporting onnx graph.
And if the density < 90%, insert a flag PythonOp after the embedding
node as:
```
Embedding
|
PythonOp (func_name:_FlagPaddingElimination) (insert if density < 90%)
|
Following graph
```
When the PaddingElimination is invoked, it check if there is the flag
PythonOp(func_name:_FlagPaddingElimination) after the Embedding node and
if it is, remove it and do the padding elimination optimization.
### Prompt layer-wise when applicable
Give explicit prompts in export failures to users to enable layer-wise
memory optimization if we found the checkpoint function is used.
- Using checkpoint function is a strong indicator that the model is too
large to fit in GPU memory.
- If we don't override the checkpoint function here, mostly ONNX export
will be failed. 1. For old version PyTorch, when handling gradient
checkpoint feature, we just throw an exception. 2. For new version
PyTorch, an export failure happens.
- But both failures did not give users explicitly "HOW" to mitigate.
This PR did that.
``
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Support more ops for recompute
To cover Mistral model, and support padding elimination ops.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Fix transformer layer detection for recompute
Originally logic miss detecting the layer boudary node in Mistral model.
This PR simplifies the searching, by using more strong pattern's match,
to make sure it is flexible enough to cover different transformer
variants.
Also add a UT.
Add a warning when user enable layerwise recompute but no layer boudary
nodes are found.
### Keep original name during fusion
This could be helpful to know where the fused node coming from, I feel
this is very useful when debugging the execution order issues between
different transformer layers.
For example:
- A node named
`/_original_module/model/layers.1/self_attn/MatMul/MatmulTransposeFusion//MatMulScaleFusion/`
goes through two fusion paths in the 1st transformer layer - e.g.
`MatmulTransposeFusion` and `MatMulScaleFusion`.
-
`/_original_module/model/layers.2/post_attention_layernorm/Mul_1/SimplifiedLayerNormFusion/`
node is a fused node by `SimplifiedLayerNormFusion`.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
in fp16 input, the softmax will return nan in some case,
the reason is because in float16 dtype,
std::numeric_limits<float16>::infinity() will return 0 instead of inf
### Fix memory stats printing
The mmeory stats printing is failed when module is in eval mode, doing
ORTModule wrap. At that time, runtime inspector for training manager
should have training model being true, but got a false (because existing
logic get the boolean from module.training). Runtime inspector as part
of training manager or inference manager should know it is serving
training or not explicitly, so we cannot depend on the stat of
module.training during ORTModule initialization.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Why we need to define softmax export logic here?
For the usage `nn.functional.softmax(attn_weights, dim=-1,
dtype=torch.float32)` in the model,
76a33a1092/src/transformers/models/mistral/modeling_mistral.py (L302)
If dtype is specified, the input tensor is casted to dtype before the
operation is performed.
This is useful for preventing data type overflows. While existing ONNX
exporter do the cast after the operation, which is not correct.
(cf06189a2d/torch/onnx/symbolic_opset13.py (L27)).
This override can be a workaround before PyTorch fix the issues in
coming releases.
(TODO: pengwa - add PyTorch versions when the issue is fixed).
@thiagocrepaldi We may need a fix in PyTorch repo as well.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
