saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-31 23:27:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
5801 commits 2856 branches 57 tags 1.3 GiB
Commit graph

816 commits

Author SHA1 Message Date
Edward Chen
ddb4c05852
Save graph runtime optimizations for minimal build (#9508) 
Add support for saving graph runtime optimizations in an ORT format model. The idea is to allow some optimizations to be "replayed" at runtime in a minimal build. The replaying part will be in a future change.
 2021-11-04 10:49:46 -07:00
baijumeswani
230099e482
Make ORTModule serializable (#9634) 2021-11-03 13:54:05 -07:00
groenenboomj
5c56fa0def
Miopen conv grad (#9574) 
* Add source for conv_grad

* Add sources for ROCm EP.
* Transliterate sources for conv_grad for ROCm EP.

* Add conv_grad to ROCm EP

Add conv_grad to ROCm execution
provider.

* Update ROCm EP ConvGrad

Update ConvGrad for the ROCm EP to match other EP
changes and fix a build issue.
 2021-10-31 11:19:46 -07:00
Hariharan Seshadri
b5f7bb7d10
Update ONNX (#9462) 2021-10-29 10:33:40 -07:00
Xavier Dupré
9c15c68ed4
Enable fallback when forward fails due to non contiguous tensor (#9369) 2021-10-28 13:04:54 -07:00
Thiago Crepaldi
5d5c03bcdc
Fix opset version change by not using copy of global constant (#9393) 2021-10-27 12:42:06 -04:00
satyajandhyala
f29057c7c0
Added TanhGrad. (#9507) 
* Added TanhGrad.
 2021-10-26 09:10:03 -07:00
pengwa
b125446f9c
Optimize python overhead of APEX amp (#9447) 
* optimize python overhead of _post_amp_backward

* overwrite apex amp's zero_grad for faster implementation

* move unscale_fp16_grads_into_fp32_grads into C++ impl

* improve the efficiency furthur, reducing 3.5ms to 1.7ms for unilm.

* unilm 1.7ms to 338us: 1). optimize python list <==> std::vector copy, 2). launch the kernels as long as num_elem reach thresh hold. This help reduce the CUDA idel time.

* refine the logic a bit after validating

Co-authored-by: Baiju Meswani <bmeswani@microsoft.com>
 2021-10-26 13:13:49 +08:00
ashbhandare
0270ff7951
Minor import fix (#9538) 2021-10-25 21:29:31 -07:00
Vincent Wang
fb4f7dbbb7
Call ATenOp for ReduceSum on ORTModule (#9471) 
* call ATenOp for ReduceSum

* Enable ReduceSum ATenOp for training only

* always load extension
 2021-10-26 09:48:57 +08:00
Sherlock
3ed8ade675
Use SafeInt for malloc related computation (#9503) 
* Use SafeInt for malloc related computation
 2021-10-22 16:42:12 -07:00
Wei-Sheng Chin
beddbdec5a
Fix PythonOp exporter (#9318) 
Register PythonOp exporter with the right symbol.
 2021-10-22 10:45:45 -07:00
Wei-Sheng Chin
d2d480a0db
Allow None As Autograd Context (#9315) 
* Allow none ctx

* Update orttraining/orttraining/test/python/orttraining_test_ortmodule_autograd.py

Co-authored-by: pengwa <pengwa@microsoft.com>

* Address a comment

Co-authored-by: pengwa <pengwa@microsoft.com>
 2021-10-21 20:37:36 -07:00
Jeff Daily
66ceb6926d
rehipify ROCm EP files under orttraining (#9443) 
* rehipify rocm ep files under orttraining committed to source control

* fix flake8 error
 2021-10-21 13:36:21 -07:00
Xavier Dupré
5797bd6db3
Remove one unnecessary deepcopy in unflatten_user_output (#9353) 
* Removes one unnecessary deepcopy
 2021-10-21 10:44:27 +02:00
Nick Kreeger
f1123c2fb3
Fix whitespace and style in concat.cc (#9452) 2021-10-20 12:43:46 -05:00
Changming Sun
406f1629c1
Remove Featurizers code (#9300) 2021-10-20 10:20:35 -07:00
baijumeswani
20eaed43e5
Ignore all string inputs to ORTModule AB#1310803 (#9344) 2021-10-19 16:34:47 -07:00
baijumeswani
757bc66720
Set cuda version to be None instead of an empty string (#9435) 2021-10-19 11:10:52 -04:00
baijumeswani
5da4e07daa
Make FusedAdam mathematically equivalent to Transformers AdamW (#9343) 2021-10-18 16:03:18 -07:00
pengwa
f05c285a58
Exception when duplicated autograd.Function name detected (#9351) 
* Exception when duplicated autograd.Function name detected

* reorder a bit for a bittle bit better perf

* fix a bug in previous PR :(

* correct the error message a bit
 2021-10-15 12:23:13 +08:00
Jeff Daily
c8789d3047
[ROCm] static re-hipify of CUDA EP to ROCm EP, now a shared provider (#8877) 
* re-hipify all rocm EP sources

* fix all other files affected by re-hipify

* add cuda_provider_factory.h to amd_hipify.py

* do not use cudnn_conv_algo_search in ROCm EP, missing reduce min registration

* Fix ReduceConsts template specialization introduced in #9101.

Fixes the error when building for ROCm 4.3.1:

error: too many template headers for onnxruntime::rocm::ReduceConsts<__half>::One (should be 0)

* fix flake8 error in amd_hipify.py

* speed up hipify with concurrent.futures

* flake8 fix in amd_hipify.py
 2021-10-14 15:15:51 -07:00
Abhishek Jindal
23700a15a0
Abjindal/eager windows build (#9326) 
* removing warnings which are causing errors from torch and changing flags for Windows

* adding MKL library resolution and comments

* cleaning up the code

* fixing onnxruntime_python file for windows build

* fix the include order to aovid the python_d.lib issue on win debug build

* changes for warnings, typos and other comments

* merge conflict

* adding fix for mkl library error

* Revert "adding fix for mkl library error"

This reverts commit 73b87c73c2.

* fix for dll path for windows

* typo for dll path

Co-authored-by: Cheng Tang <chenta@microsoft.com>
 2021-10-14 12:54:49 -07:00
Xavier Dupré
22e3f8bf54
Refactor TrainingManager.forward (#9354) 
* Refactor TrainingManager.forward
 2021-10-14 12:54:31 +02:00
pengwa
5ee47e3ffa
legacy_megatron-lm/deepspeed_ZERO1&2 FP16_Optimizer wrapper (#9184) 
* megatron-lm FP16_Optimizer Wrap, allow model parallelism aggregation optional

* add deepspeed zero1 and zero2 - checkoverflow & clip norm

* re-structure code and add the copyright

* update the document

* refine the code after validation
 2021-10-14 09:01:23 +08:00
Chandru Ramakrishnan
ba0cca96f0
Hooked up eager logging to ORT default logger. (#9340) 
* Hooked up eager logging to ORT default logger.
 2021-10-13 18:10:32 -04:00
Tang, Cheng
f0bc35c4ba
fix a hardcode type (#9337) 2021-10-12 13:44:46 -07:00
Tang, Cheng
48737091c0
resolve the provider options before create training session in orttrainer (#9199) 
* resolve the provider options before create training session in orttrainer

* Update orttraining/orttraining/python/orttraining_pybind_common.h

Co-authored-by: Thiago Crepaldi <thiago.crepaldi@microsoft.com>

* support clear the training ep instance pool

* fix status error

Co-authored-by: Thiago Crepaldi <thiago.crepaldi@microsoft.com>
 2021-10-12 09:30:45 -07:00
ashbhandare
52c021d1f3
Fix export of aten op for Max and Avg Pool 2D (#9330) 2021-10-12 09:03:14 -07:00
Edward Chen
79e736ed25
Make onnxruntime::Status nodiscard (#9279) 
Mark onnxruntime::Status class with [[nodiscard]] attribute.
Fix existing warnings.
 2021-10-08 17:10:31 -07:00
satyajandhyala
29379db432
Added SigmoidGrad schema and kernels. (#9244) 
* Added SigmoidGrad schema and kernels.

* Added test_sigmoid_grad function.
 2021-10-08 11:03:28 -07:00
Tang, Cheng
68601fc296
error handling ffor eager mode's data transfer (#9261) 2021-10-07 17:16:33 -07:00
ytaous
7166586d7e
Enable SkipCheck by default (#9215) 
* Enable SkipCheck by default

* fix UTs

* fix UT

* fix UTs

* fix UTs

* address comments

* fix UT

* enable skipchecks

* move _SkipCheck back

* move _SkipCheck back

* move _SkipCheck back

* Update orttraining/orttraining/python/training/ortmodule/_inference_manager.py

* Update orttraining/orttraining/python/training/ortmodule/_utils.py

Co-authored-by: Ethan Tao <ettao@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Thiago Crepaldi <thiago.crepaldi@microsoft.com>
 2021-10-07 15:47:14 -07:00
Tang, Cheng
c002dc86a3
set mpi group init flag after add group (#9293) 2021-10-07 10:09:16 -07:00
Thiago Crepaldi
52d067402a
Fix all-or-nothing fallback for bad ORTModule init (#9277) 
* Fix all-or-nothing fallback for bad ORTModule init

* Address comments
 2021-10-06 15:12:27 -04:00
baijumeswani
bcdb411c8d
Implement FusedAdam for ORT adapted from DeepSpeed (#9266) 2021-10-05 20:50:34 -07:00
ashbhandare
35c2102cfa
Fixes for GatherND, Multinomial (#9143) 
* register gathernd kernel, aten multinomial

* fix CI, add test

* review comments
 2021-10-05 14:51:58 -07:00
G. Ramalingam
0b77c9ca7c
Cleanup function definitions of contrib ops (#9265) 
* Simplify function definitions

* Simplify fast-gelu function definition

* Simplify training function op body definitions

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>

* Eliminate redundant function

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>

* Formatting changes

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>

* Minor formatting changes

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>

* Add comment

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>

* Specify int64 type for constant 1

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
 2021-10-05 11:38:42 -07:00
Thiago Crepaldi
6e2f66ee9c
Allow custom exporter args + bug fix (#9242) 2021-10-04 11:32:42 -04:00
baijumeswani
45399d5ace
Remove TORCH_WARN to avoid torch string related operations that take up time (#9238) 2021-10-01 13:56:04 -04:00
Tang, Cheng
be4d887439
Fix ONNX exporter call with latest API for ORTrainer (#9228) 
* update the exporter call with latest api in orttrainer

* use official export api instead of the private call
 2021-10-01 13:49:55 -04:00
G. Ramalingam
e79be39081
LayerNormGrad function body and LayerNorm inference/body fix (#9160) 
* Add function body for LayerNormGrad

* Fix LayerNorm schema for multiple normalization dims
 2021-09-30 12:03:08 -07:00
Thiago Crepaldi
ceb51dda4a
Support external torch cpp extensions on ORTModule (#9223) 2021-09-30 10:37:35 -04:00
satyajandhyala
278928a102
Added a test case for python gradient builder. (#9207) 
* Register Cos operator gradient using ORTModule's register_gradient and compare gradient against PyTorch.
 2021-09-29 09:24:12 -07:00
Suffian Khan
6f580f07de
Switch AMD CI pipeline to use environment image from onnxruntimecibuildenvironment (#9206) 
* shift docker image reference for amd ci pipeline

* fix service endpoint

* reduce perf tolerance
 2021-09-28 13:06:16 -07:00
ytaous
d3f859fe30
Dropout Vectorized Kernel (#9157) 
* vectorized kernel

* fix build

* re-calibrate expected loss

* fix build

* re-calibrate convergence results

* more re-calibrate on loss

* divide kernels

* adress comments

* more calibration

* calibration

* per comments

* enable sync

Co-authored-by: Ethan Tao <ettao@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
 2021-09-27 17:19:12 -07:00
Wei-Sheng Chin
1b0816859f
Only wrap sub-modules which can be wrapped as ORTModule (#9021) 2021-09-27 17:18:22 -07:00
baijumeswani
c30cc9190a
Change the agent pool for orttraining-distributed pipeline (#9179) 2021-09-26 21:26:44 -07:00
baijumeswani
fd91bf91c9
Print full stacktrace exception when exporter fails (#9169) 2021-09-24 10:24:37 -04:00
Vincent Wang
39dc6ea8a3
Fix to_dlpack Failure on PyTorch-1.10 (#9151) 
* workaround to_dlpack fail in new pt version

* add torch code link
 2021-09-24 09:48:07 +08:00