saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-10 17:37:14 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
4580 commits 2856 branches 57 tags 1.3 GiB
Commit graph

4580 commits

This branch
This branch
All branches
Author SHA1 Message Date
Zhang Lei
dbcfc4bee6
Add mlas_bench tools. Starting with sconv bench and sgemm bench. (#7139) 
* Add mlas_bench tools. Starting with sconv bench and sgemm bench.

* Some update with build related.
 2021-04-06 10:30:18 -07:00
ashari4
56b22c1c6b
Fix assert that the tensor's device type is 'cpu' #7248 2021-04-06 09:08:32 -07:00
ashbhandare
e9ffcfa247
Add cuda kernels for GreaterOrEqual, LessOrEqual, Where; modify Clip to avoid memcpy (#7187) 
* Where and Clip cuda kernel support

* GreaterOrEqual and LessOrEqual cuda kernels

* Clip input GPU mem

* review comments

* Add CPU kernel as well

* review comment

* Add kernel def hash for new op kernels

* Fix CI
 2021-04-06 09:04:38 -07:00
Derek Murray
c85657cfd7
Update test_training_model.onnx to opset 12. (#7251) 
Co-authored-by: Derek Murray <demurra@microsoft.com>
 2021-04-06 07:49:58 -07:00
Tracy Sharpe
a9dbb511fb
MLAS: fix qgemm bus error with Android + ARM32 (#7250) 2021-04-05 22:46:04 -07:00
Olivia Jain
fb40602ea2
Mem trt (#6868) 
* adding trt comparison and memory consumption

* creating separate docker file
 2021-04-05 22:16:12 -07:00
Changming Sun
2fcd69d644
Cleanup build.py (#7245) 2021-04-05 18:49:29 -07:00
Changming Sun
5bd192c439
Update ContribOperators.md (#7246) 2021-04-05 17:11:33 -07:00
Pranav Prakash
3b16afc0db
Make dW optional for convgrad (#7083) 2021-04-05 17:05:20 -07:00
Guoyu Wang
c5973fbbac
Update the build script for Android AAR package (#7229) 
* Update the build script for Android AAR package

* Address CR comments
 2021-04-05 16:37:22 -07:00
Suffian Khan
9f14af9809
Add BERT-L perf regression test on MI100 and re-enable batch size test (#7240) 
* restore bs test and add perf test

* update perf number and fix path to results
 2021-04-05 15:51:52 -07:00
Ryan Lai
10102c09b6
Add better model test error messaging (#7239) 2021-04-05 14:59:19 -07:00
Ashwini Khade
e7c5dcd572
Fix Zip-Nuget-Java Packaging Pipeline (#7208) 
* Ignore test failures due to opset support

* skip identity sequence test

* plus fixes
 2021-04-05 10:58:13 -07:00
Chun-Wei Chen
3ee9b0ec4d
Add detailed assertion error message (#7232) 2021-04-05 10:05:40 -07:00
Marek Šuppa
008065aab1
Update README.md (#7043) 
* Fix the precision type (switch from nonexistent `int32` to `fp32`).
 2021-04-05 10:03:14 -07:00
ashbhandare
2b8513539e
Div mul fusion (#7183) 
* Div mul fusion

* Change to rewrite rule

* Add to inference transformers
 2021-04-05 09:35:30 -07:00
Weixing Zhang
74ee24cf7f
rename cuda_mem_limit and hip_mem_limit to gpu_mem_limit for both CUDA EP and ROCm EP (#7226) 
With this change, differentiating CUDA EP and ROCm EP is not needed in training script when mem_limit option needs to be set.

Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
 2021-04-05 09:04:04 -07:00
baijumeswani
68b12a6179
Support for saving and loading pytorch compatible state dictionaries (#7220) 
* Override methods on torch.nn.Module to get direct access to the methods on the original module.
 2021-04-05 03:40:41 -07:00
Yufeng Li
8d737f9770
handle optional input in quant topo sort (#7223) 2021-04-02 20:42:48 -07:00
Weixing Zhang
59b57d8322
HSA_NO_SCRATCH_RECLAIM and RCCL_ALLTOALL_KERNEL_DISABLE are not needed for ROCm 4.1 (#7224) 
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
 2021-04-02 18:19:11 -07:00
Ahmad Zakaria
ba5f056b09
move trt_profile to TensorrtFuncState and reuse it (#7195) 
use unordered_set instead of unordered_map to keep track of dynamic shape tensors with shape updates

fix: insert input_name in the set of input_names

move trt_profile to TensorrtFuncState and reuse it
 2021-04-02 17:09:03 -07:00
Weixing Zhang
ef88dc912c
enable more unit tests for ROCM EP (#7222) 2021-04-02 15:57:08 -07:00
Guoyu Wang
afbbeaa30a
[NNAPI/CoreML EP] Add Onnx opset 14 support (#7211) 
* Add opset 14 support for nnapi/coreml ep

* Address CR comments
 2021-04-02 13:18:47 -07:00
Sherlock
a98c2ebb8c
Enable saving optimized models in OrtModule (#7214) 
* Enable saving optimized models in OrtModule

Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
 2021-04-02 12:37:05 -07:00
RandySheriffH
ebde320950
Add cupti path for python gpu packaging pipeline (#7200) 
* add cupti dll path for py3.8

* correct path

* add prints

* replace path join

* add all path

* restore pipeline

* format

* expand path only for python 38&39

* add all cupti path

Co-authored-by: Randy Shuai <rashuai@microsoft.com>
 2021-04-02 12:12:46 -07:00
Weixing Zhang
2d352056cf
Support SkipLayerNorm for ROCm EP (#7210) 
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
 2021-04-02 09:03:30 -07:00
Weixing Zhang
a3f17c8b0d
update lamb and GatherGrad kernel for ROCm EP (#7184) 
With ROCm4.1, the CUDA implementation of Lamb and GatherGrad can be
utilized for ROCm EP.
 2021-04-02 09:02:49 -07:00
Weixing Zhang
17f91ff410
remove un-needed header file. (#7193) 
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
 2021-04-01 21:05:58 -07:00
Ryan Hill
5a6d477625
Make IDataTransfer be directly shared with shared providers (#7215) 2021-04-01 20:39:16 -07:00
Edward Chen
0ebeaf529d
Check kernel def hashes (#7120) 
Add unit test for verifying kernel def hashes.
Add way to add new types to kernel definition without changing hash.
 2021-04-01 17:42:58 -07:00
ashbhandare
15c67ddbf0
Make output 1 of ConcatTraining Optional and place on CPU (#7199) 
* Optional input 1 on CPU ConcatTraining

* Rename output_1
 2021-04-01 16:05:17 -07:00
Jesse Benson
4543459984 MIOpen supports MIOPEN_REDUCE_TENSOR_AVG now. 2021-04-01 16:00:34 -07:00
Yufeng Li
34a8b22186
disable prepacking in training (#7201) 
* disable prepacking in training
 2021-04-01 14:03:47 -07:00
sfatimar
52bcef4d4f
Openvino ep 2021.3 (#7180) 
* Integrate openvino-ep-2021.3

* operators type

* changed the myriad as it is case sensitive

* logging information for openvino-ep-2021.3

* Unit test fix

* Resize operator added for myriad

* Fixed python tests for CPU and GPU

* data commit for loop tile and gatherelements failure

* adding checks for Where

* fixing gatherelements and loop tests

* disabling instance normalization test for now as there seems to be a
myriad bug, putting loop in ops supported only because all the tests
fail

* gather elements op test taking care of warning message

* condition needs to be an intializers

* Disabled python test for Myriad

* Disable compilation warning for MSVC windows compiler

* softmax_test, threedimaxis0 and 1 test give accuracy mismatch
tensoroptest disables test gives accuracy mismatch
gather test gives accuracy mismatch

* Updated with ov version 2021.3

* Updated with ov version 2021.3

* Updated README

* Disabling python tests for cpu

* Disabling python tests with accuracy mismatch on cpu

* Added fix for Linux CI Pipeline failure

-> Disabled tests that were throwing segfault

Co-authored-by: sfatimar <sahar.fatima@intel/com>
Co-authored-by: MaajidKhan <n.maajidkhan@gmail.com>
Co-authored-by: Aravind <aravindx.gunda@intel.com>
 2021-04-01 11:28:54 -07:00
baijumeswani
249a2c14ef
Pin version of pytorch to 1.8.1 for ORTModule CI pipeline (#7167) 
* Pin version of pytorch to 1.8.1 for ORTModule CI pipeline
* Use pytorch-lightning stable version 1.2.5
* Revert to cuda 10.1
 2021-04-01 09:37:47 -07:00
George Wu
fc6ac5bfac
dnnl fixes (#7202) 2021-04-01 07:34:18 -07:00
Scott McKay
329fd03bb4
Add int32_t as required type to some operators (#7192) 
* Updates to some operators to always support int32 and int64 based on testing of Android package build config with a minimal build.

If an operator can be used for shape manipulation (int64) it is frequently used for indices manipulation (int32), so we enable both types for that set of ops.
  - e.g. BERT models take indices as input
  - Scatter/Gather ops utilize indices

Misc. fix to python bindings to exclude call that fails in a minimal build.
 2021-04-01 19:32:34 +10:00
Edward Chen
04679e31ab
Specify CUDA compute capability 7.5 in Linux GPU build (#7203) 
Recently a build agent pool was changed to use T4 GPUs (CUDA compute capability 7.5). Updating some CUDA build options to accommodate that.
 2021-03-31 18:51:44 -07:00
Hariharan Seshadri
0e0dd50e39
Support int32 type for TopK CPU op (#7089) 2021-03-31 18:08:21 -07:00
Xavier Dupré
b370ddbf5e
Removes unnecessary transpose in operator Einsum (#7141) 
* remove one unnecessary transpose
* add more unit test
 2021-03-31 09:59:08 +02:00
Guoyu Wang
d500c5952b
Add Android AAR packaging script for ORT-Mobile (#7138) 
* Add Android aar packaging script for ORT-Mobile

* Address CR comments
 2021-03-30 18:42:18 -07:00
Yulong Wang
0fdef1bf47
[Node.js binding] upgrade y18n to v4.0.1 (#7185) 2021-03-30 16:09:04 -07:00
Negin Raoof
45cb0cae8c
Adding TorchEmbedding contrib op (#7136) 
* Adding TorchEmbedding contrib op

* Update contrib_defs.cc

* Shape fix

* Update shape_inference_test_helper.h

* Fix typo

* Fix test

* Fix for test code

* Merge

* Fix CI

* Fix for CI

* Fix CI no-contrib
 2021-03-30 14:33:25 -07:00
liqunfu
e545604499
. (#7165) 2021-03-30 13:58:30 -07:00
RandySheriffH
d880578537
Exclude cpuid.h from Mac non x86 arch (#7166) 
* add ifdef to exclude inclusion from non x86 arch

* exclude calling of __cpuid_count

Co-authored-by: Randy Shuai <rashuai@microsoft.com>
 2021-03-30 11:50:42 -07:00
Edward Chen
0ccfe6c86a
Enable type reduction for Scatter/ScatterElements CPU kernels (#7171) 
Enable type reduction for Scatter/ScatterElements CPU kernels. Some refactoring to reduce binary size.
Add MLTypeCallDispatcher methods.
Minor cleanup for Pad CPU kernel.
 2021-03-30 11:02:24 -07:00
Tang, Cheng
07201bac7a
expose session option and provider options (#7112) 
* expose session option and provider options

* merge provider_names and provider_options

* integrate into orttrainer options

* fix doc string

* fix a typo

* Update orttraining/orttraining/python/training/orttrainer.py

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

* Update orttraining/orttraining/python/training/orttrainer.py

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

* Update orttraining/orttraining/python/training/orttrainer_options.py

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

* fix the usage of provider_options

* Update orttraining/orttraining/python/training/orttrainer.py

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

* Update orttraining/orttraining/python/training/orttrainer.py

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

* update expected result in tests

* fix default provider options

* minor update to trigger rebuild

* minor update to trigger rebuild

Co-authored-by: Thiago Crepaldi <thiago.crepaldi@microsoft.com>
 2021-03-30 09:49:45 -07:00
Yufeng Li
c4ebc60870
sort quantized nodes in topo logical order (#7172) 2021-03-30 09:01:15 -07:00
Yufeng Li
4f30341253
Check the count of DequantizeLinear for matmul (#7174) 2021-03-30 09:00:48 -07:00
Tracy Sharpe
a01334ba56
MLAS: activate udot kernel on Windows ARM64 (#7169) 2021-03-29 17:56:48 -07:00