https://github.com/Microsoft/CNTK

sort by:
Revision Author Date Message Commit Date
47aa821 Adding support for exporting CNTK's Sequence::IsFirst and Sequence::IsLast nodes 30 November 2018, 19:16:55 UTC
d1d1133 ConvTranspose asymmetric padding * Temporarily reverse the extra padding location in case of SAME_UPPER vs SAME_LOWER for convTranspose to match with onnxruntime. * In case of importing asymmetric padding convTranspose, use symmetric pads by alter the output_shape and pads, and attach a slice node afterwards to enable cudnn. * Fix a bug in slice/squeeze attribute axes export. 27 November 2018, 17:33:15 UTC
196df1a Update image layout in readme. 26 November 2018, 20:49:56 UTC
eeae2bb Add netron news and update iteration. 26 November 2018, 19:57:39 UTC
8c15683 Merge branch 'vadimma/binmlf' 24 November 2018, 09:12:41 UTC
0f1b4e0 Merge pull request #3512 from changeworld/patch-1 Fix typo 24 November 2018, 03:44:06 UTC
b566e23 Fix typo Ture -> True 23 November 2018, 15:25:07 UTC
0c1d283 update latest onnx and onnxruntime, fix shape inference 22 November 2018, 16:56:07 UTC
631392e Update bin mlf reader 15 November 2018, 15:24:40 UTC
d7101a2 Clean up bin mlf test 12 November 2018, 20:34:56 UTC
8d87948 Ensure mlf reader is included in the test 12 November 2018, 20:19:37 UTC
72200ca Add bin mlf reader tests 11 November 2018, 22:29:27 UTC
7ef57de Add dim label to bin mlf reader 11 November 2018, 05:32:28 UTC
22c7c3c Update BlockMomentumDistributedLearner.h 08 November 2018, 22:32:13 UTC
69df29e Merge branch 'liqun/seqopsStage' 08 November 2018, 22:14:44 UTC
2405457 Update BlockMomentumDistributedLearner.h 08 November 2018, 20:27:55 UTC
ab4bee2 Support RNN ops in a Scan loop Update with latest ONNX Update with latest ONNX graph IR Support sequence ops - Sequence::Gather, Sequence::PastValue, Sequence::FutureValue, etc. 08 November 2018, 02:36:20 UTC
b51e8c2 Expose binmlf in python 06 November 2018, 00:32:59 UTC
3f46cf0 Updates on several ONNX exports. * ConvTranspose outputShape: now pads values are always exported even when outputShape is given. The reason is that CNTK and ONNX have different padding specs. * Flatten: in CNTK flatten does not affect batch axis, this should be preserved in ONNX. 03 November 2018, 00:18:07 UTC
a19ce9e Resolve forward_backward docstring Sphinx error. 02 November 2018, 17:43:24 UTC
29818ff Merge branch 'bowbao/onnxruntime_ci_stage2' 01 November 2018, 06:58:24 UTC
0cd2fae Merge branch 'pull/3476' 31 October 2018, 23:43:42 UTC
fca1396 Add onnx_test_runner verification in CI. * onnx_test_runner.exe will be called on win64 GPU tests to verify if the output data produced by CNTK in onnx_op_test and onnx_model_test(cntk_model_test) matches in onnxruntime. 31 October 2018, 22:46:56 UTC
a0d6f14 Fixed type and error in identifying the deserialiser 31 October 2018, 13:05:27 UTC
cdd18d2 Update LSTM/Truncated-Kaldi CPU baseline 29 October 2018, 18:16:42 UTC
3cef1fd fix python 27 onnx test 28 October 2018, 17:22:07 UTC
1ff5548 Set linux pip install future version to 0.16.0 * This was not set explicitly. Lately the version was updated to 0.17.0 and is causing failure in CI. 26 October 2018, 17:40:01 UTC
0b41602 Update more baseline affected by fixing InvStdDev. 26 October 2018, 00:43:52 UTC
77dad35 Added better documentation to forward_backward 25 October 2018, 11:17:25 UTC
e940605 Support ONNX Scan op 20 October 2018, 04:36:21 UTC
a55e871 Fix InvStdDev. * Issue was that AssignSqrOfDifferenceOf(beta, input, mean, alpha) assigns mean value to the gaps in input. These values are then reduced within this function, leading to incorrect results. The fix is to execute assign and reduce separately, and mask gaps to zero again before reducing. * Update test baseline affected by this change (err is lowered by <1%). 19 October 2018, 17:24:11 UTC
0ffdcf7 Overhaul node name export & other fixes * Overhaul node name export. Create static class UniqueNodeNameStorage to manage ONNX node name generation with maintained one-to-one mapping between CNTK Uid, while preserving the original CNTK node name by best efforts (#3358). * Update onnx_op_test to test the preservation of original CNTK node names in exported/imported models. * Update onnx_test_helper to support proper linking of test data and onnx model input/output with unique names. * Update onnx_test_helper to generate .bat file to run exported models in further onnxruntime verification. * Fix Sum import to support arbitrary number of inputs. Sum implementation in CNTK backend is loop of Plus, which takes care of potential broadcast issues. 18 October 2018, 01:36:48 UTC
da2e610 Replaced wchar/wstring with char/string in C interface 12 October 2018, 21:12:23 UTC
149d87b Adding ONNX export support for OneHotOp. 05 October 2018, 21:06:56 UTC
6063866 Merge branch 'bowbao/pooling_pad_offset' 02 October 2018, 03:41:06 UTC
bf37aad Fix pad offset computation for pooling * Compute keyInterior according to the updated algorithm for computing cell offset key. * Update unittest of avg_pooling/max_pooling for cases that requires auto_padding = True. Previous test cases cover only those that do not need padding. 02 October 2018, 00:21:30 UTC
7c838d3 fix csharp examples and references 01 October 2018, 22:45:13 UTC
fcdeef6 Support crop_manual export & import. 29 September 2018, 20:46:17 UTC
a36fae8 Support logPlus(log_add_exp) export to ONNX * ONNX supports similar op ReduceLogSumExp. Conversions are added when exporting. * Refactored CNTKToONNXHelper::BroadcastInputsIfNeeded to support more generalized cases. 28 September 2018, 22:59:55 UTC
c2072cc Add support for ONNX export of StraightThrough op. 27 September 2018, 17:25:11 UTC
1aab76a Updating ONNX submodule hash to include defs for ConstantLike and EyeLike ops. 27 September 2018, 01:01:34 UTC
ce503f8 pooling export fix for backward compatibility 26 September 2018, 00:09:39 UTC
9165fd0 Merge branch 'kedeng/fixCrash' 25 September 2018, 00:26:34 UTC
58f810f update with ONNX1.3 and latest onnxruntime 22 September 2018, 16:53:27 UTC
1489de8 Fix a crash in transpose_times simplification to element times 22 September 2018, 05:33:41 UTC
93cc680 Update reshapeing_test to accomodate python 2.7 21 September 2018, 17:24:57 UTC
d626c2a Merge branch 'bowbao/gather_backward' 21 September 2018, 05:02:57 UTC
da6b0bc GatherNode backward: add check for no dynamic axis Previously, to resolve issue of gather producing incorrect gradient values, validity mask check was added to ensure we don't count non-valid cells as 0. However, this check is needed only for input that has dynamic axis, i.e. inputs that have MBLayout. 20 September 2018, 21:54:39 UTC
b2c28cc updating iteration documentation file 20 September 2018, 21:39:58 UTC
0a3eb3b Update onnx_model_test with tests on cntk pretrained models 19 September 2018, 18:27:55 UTC
6f09c39 Merge branch 'release/2.6' 18 September 2018, 02:33:30 UTC
4ed1896 set public_build to "no"/false 17 September 2018, 23:10:48 UTC
1be3b64 update readme for .net support 15 September 2018, 01:07:45 UTC
d355c1c Updated current_iteration with .net support 14 September 2018, 20:07:24 UTC
04caa9d Updated current_iteration with .net support 14 September 2018, 19:29:51 UTC
7c1b0fa udpate readme with current iteration 14 September 2018, 00:01:45 UTC
da31ba0 Update current_iteration.md 13 September 2018, 23:57:44 UTC
e4d7081 Update current_iteration.md. 13 September 2018, 23:57:30 UTC
4f965aa update version # in cntk_common.cmake 13 September 2018, 23:34:30 UTC
82d350d bump up version number 13 September 2018, 22:52:47 UTC
d264a26 Update current_iteration.md 13 September 2018, 20:44:42 UTC
be28e86 Update current_iteration.md. 13 September 2018, 18:18:53 UTC
deda94b Support pooling(cpu) where kernel center is on pads. - Previous implementation has the assumption that (0 <= dk < width). This assumption doesn't stand when lo >(kernel - 1) / 2. The updated calculation supports arbitrary lo & hi non-negative integer value. The new calculation has dk in range (0, width + hi + lo]. - Enables onnx backend test {averagepool_2d_pads, maxpool_2d_pads} to pass. 13 September 2018, 04:37:21 UTC
62e18f4 Improve clarity in pads calculation for conv/pool - Refactor function CalcPaddingForSameLowerOrUpperAutoPad in conv/pool import, changing parameter "const Variable& input" to "const NDShape& inputWithBatchAxisShape", to specify the required shape format as [N x C x H x W]. 13 September 2018, 04:37:21 UTC
35e3701 Merge branch 'sedymche/onnx-min-max' 13 September 2018, 02:42:25 UTC
d33b7b4 update iteration plan 13 September 2018, 00:24:45 UTC
6a4ec05 Merge branch 'peykash/batchnorm_and_pooling_fixes' 13 September 2018, 00:13:46 UTC
61d7dab Support more than 2 inputs for ONNX Min/Max import. 12 September 2018, 22:12:14 UTC
8b48976 Adding CNTK 2.6 release work summary to current_iteration.md 12 September 2018, 18:12:10 UTC
b374e14 fixes on Batchnorm and Pooling for v1 pretrained models after removal of sequence axis from input 12 September 2018, 17:02:52 UTC
5897265 small patch on conv/pooling export - when pads are all zero, check if autopad is true. - when pads are all zero, check if ceilOutDim is true, and extra cells are needed. 11 September 2018, 15:52:22 UTC
61572e8 Update onnx_model_test skip list 11 September 2018, 15:51:45 UTC
0754b38 update onnx_model_test with tests from onnx backend test 09 September 2018, 20:53:26 UTC
2f52f22 update conv/convtranspose/pooling import. pad values are explicitly computed based on ONNX spec equations during import in the following cases: - case 1: when auto_pad is SAME_UPPER | SAME_LOWER for convolution, convolution transpose and pooling. - case 2: when output_shape is explicitly set for convolution transpose. note: output_shape in ONNX spec can have the two below format: 1. [X1 * X2 * ... * Xn] 2. [N * O * X1 * X2 * ... * Xn] 09 September 2018, 20:41:04 UTC
d877233 Make broadcast ops compitable between CNTK and ONNX, Enable ONNX export/import for optimizedRNN op, More ONNX support for Sequence ops 09 September 2018, 15:59:33 UTC
fcf9f48 Overhaul conv/convTrans/pooling pads value export - Update exporting of conv/pooling to always export pad values. - Enable correct exporting of multiple pretrained models (ResNet50/ResNet101/ResNet152_ImageNet_Caffe, etc). - Overhaul convtranspose pads exporting - Support conv weight export with omitted out channel axis (LRN). - Add tests in onnx_op_test to cover the above changes 06 September 2018, 18:46:14 UTC
e3a1acf Resolve dependencies and build issues -Temporary add importorskip around import onnx -bump up .yml matplotlib version 05 September 2018, 22:02:23 UTC
dc5e482 fix onnx average pooling export. - this fix solves the issue that ceilOutDim == true will enforce exporting auto_pad as true, even if autoPadding is explicitly set to false. 31 August 2018, 01:19:04 UTC
77a8c49 Temporarily skip onnx_model_test if import onnx fail 30 August 2018, 17:52:53 UTC
94a43ed Merge branch 'liqun/liqun/RNN2.6.Stage' 30 August 2018, 07:40:20 UTC
73cd53e fix nightly issues related to onnx dependencies - Windows OOBE (pip) tests & Linus OOBE tests: skip onnx_model_test. This test requires onnx to be installed. Skip Until we decide to add onnx dependencies to OOBE test environment. 30 August 2018, 00:19:19 UTC
18d9f39 skip dynamic axes wrapper, export onnx test cases, handle output op being Combine op, workaround a specific case bug of ONNX bidirectional broadcast shape inference 29 August 2018, 23:52:58 UTC
902f1a4 times export fix 29 August 2018, 17:09:07 UTC
b86fe1a Merge branch 'pull/3374' 28 August 2018, 07:17:54 UTC
1f2e42e Merge branch 'sptiwari/convtranspose_update7' 27 August 2018, 07:32:33 UTC
edc29f8 Packaging newly-added internal header files Recently, we added a couple of new header files into API/Internals. This patch includes them into our pre-built binaries. 27 August 2018, 04:36:04 UTC
b3c0fa2 Overhaul ConvTranpose to match ONNX 1.2.2. spec. 27 August 2018, 04:20:05 UTC
7dd9638 squash of the following changes: - fix flatten onnx export. - fix unsqueeze onnx export. - add comments on temporarily skipped tests. - adjust the importing of softmax, logsoftmax and hardmax with blockfunction - such that they could be exported as is back to onnx. - update reshape onnx export to pass mobilenet round trip test. 26 August 2018, 20:11:44 UTC
0e20836 CNTK splice allows broadcast. This case is handled in the change. For noop (identity) ops, its inputs and outputs types shall be set according to upstream ops. ToBatch/ToSequence and Unpack Batch/Sequence ops added during model importing need tp be skipped. Model import need to handle ops with multiple outputs 26 August 2018, 15:41:20 UTC
4a6238d reduction all axes export fix 24 August 2018, 17:02:34 UTC
d2ff412 temporarily disable 2 tests on Windows. - This is due to an issue on Windows CI introduced by adding onnx dependencies. These tests are temporarily disabled to not block CI while we investigate. - Disable CNTKv2Python/Tutorial/205 - Disable CNTKv2Python/Keras 24 August 2018, 08:27:45 UTC
88d8882 Fix issue#3228 and #3363 which found when build with MSVC+ permissive- 24 August 2018, 05:56:36 UTC
a26e542 update conftest.py to resolve doctest issue. - newer version numpy has a different print format for arrays and scalars that would potentially break the doctests. 24 August 2018, 04:28:06 UTC
28ada96 update python doctest handling newer version numpy print format. 23 August 2018, 06:21:13 UTC
56ef694 add onnx_model_test.py 23 August 2018, 06:21:13 UTC
c0ff1da fix gemm, pooling export to onnx. 23 August 2018, 06:19:18 UTC
4244320 Adding support for exporting CNTK TimesTranspose 22 August 2018, 18:33:55 UTC
3d809bf Added several internal API header files In case other projects may use these header files, we added them into API/Internals. * ComputationGraphAlgorithms.h was moved from Source/ComputationNetworkLib * PrimitiveOpType.h and EvaluatorWrapper.h were moved from Source/CNTKv2Library * PrimitiveFunctionAttribute.h was extracted from PrimitiveFunction.h. It contains a new class PrimitiveFunctionAttribute which is the collection of all attribute names for PrimitiveFunction. This change actually had a subtle side-effect. We had a global static variable s_stateAttributes that depended on PrimitiveFunction::AttributeNameRngSeed and PrimitiveFunction::AttributeNameRngOffset. After we moved those static attribute-variables into another translation unit, s_stateAttributes can be initialized with empty wstring, because PrimitiveFunctionAttribute::AttributeNameRngSeed PrimitiveFunctionAttribute::AttributeNameRngSeedOffset were initialized after s_stateAttributes. Note that the initialization order of global static variables is not well-defined cross translation units. To fix the issue, we also moved s_stateAttributes into PrimitiveFunctionAttribute class, and renamed it to s_rngStateAttributes. I think it's reasonable to consider s_rngStateAttributes to be part of the PrimitiveFunctionAttribute class. * PrimitiveFunction.h was moved from Source/CNTKv2Library 22 August 2018, 17:47:18 UTC
8a244e6 fix onnx_op_test.py reduce ops. - remove bracket around data to match latest verification. 21 August 2018, 21:23:28 UTC
b36347b Fix legacy test 21 August 2018, 19:38:26 UTC
back to top