microsoft
/
CNTK
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность
15 899 коммитов 1 074 Ветки 39 Теги 1,1 GiB
Граф коммитов

15899 Коммитов

Автор SHA1 Сообщение Дата
Peyman Manikashani 4b2b0d30cc Merge branch 'peykash/batchnorm_float16_fixes' 2018-08-17 09:54:13 +00:00
Vadim Mazalov d67eba8060 Remove template definition 2018-08-16 21:25:39 -07:00
Vadim Mazalov af7e33e45e Fix template compilation 2018-08-16 21:25:39 -07:00
Vadim Mazalov 8c83adb693 Some cleanup 2018-08-16 21:25:39 -07:00
Vadim Mazalov 648c99d9d5 Refactor common methods 2018-08-16 21:25:39 -07:00
Vadim Mazalov 086dd28252 Update .gitattributes to mark MLF as binary file 2018-08-16 21:25:39 -07:00
Vadim Mazalov cb3d08ac51 Add E2E test for binary MLF 2018-08-16 21:25:38 -07:00
Vadim Mazalov 7c59a63d6c Further binaryMLF refactoring 2018-08-16 21:25:38 -07:00
Vadim Mazalov 931193d735 Refactor binary MLF 2018-08-16 21:25:38 -07:00
Vadim Mazalov 78165413b8 Remove a logging statement 2018-08-16 21:25:38 -07:00
Vadim Mazalov e500b6e7c6 Fix binary stream reading 2018-08-16 21:25:37 -07:00
Vadim Mazalov 0ef680ec24 add binary MLF reader to makefile 2018-08-16 21:25:37 -07:00
Vadim Mazalov cd838ae5e2 Introduce model version 2 2018-08-16 21:25:37 -07:00
Vadim Mazalov 9ad1910871 Optimize senone count lookup 2018-08-16 21:25:37 -07:00
Vadim Mazalov ce95c1cb4f First working version 2018-08-16 21:25:36 -07:00
Vadim Mazalov 0a00f8b43c Implement MLF binary parser 2018-08-16 21:25:36 -07:00
Vadim Mazalov 0198fadf8e Binary index builder' 2018-08-16 21:25:36 -07:00
Vadim Mazalov d531345bb1 binary mlf index' 2018-08-16 21:25:36 -07:00
Vadim Mazalov 222efb8048 Binary mlf reader 2018-08-16 21:25:35 -07:00
Yang Chen f5b481a6ab copy HalfConverter.hpp for make_binary_drop 
HalfConverter.hpp is included from CNTKLibrary.h.
Without including it, we cannot use binary drops.
 2018-08-16 13:05:34 -07:00
Bowen Bao 490620873f Merge branch 'bowbao/onnx_imageScaler' 2018-08-15 18:01:32 -07:00
Liqun Fu bf2b5eae8c Merge branch 'liqun/Py2FixTupleUnpacking' 2018-08-16 00:12:53 +00:00
Yang Chen c8cecaace8 Dump exception messages to stderr in Release mode 
It's useful to have excpetion messages dumped to std in Release mode.
 2018-08-15 14:32:54 -07:00
liqun fu bff3002794 fix python 2.7 tuple unpacking error 2018-08-15 13:47:56 -07:00
Bowen Bao 4b4b9cda28 fix imageScaler export/import to onnx 2018-08-15 10:22:47 -07:00
Peyman Manikashani a5421862de BatchNormalization Float16 Fixes 2018-08-15 10:01:23 -07:00
Liqun Fu ef836bc293 Merge branch 'liqun/NewSequenceSliceStage2Stage' 2018-08-15 06:46:32 +00:00
Spandan Tiwari 6547e2ce7f Fix bug in import of Conv op in ONNX. 2018-08-14 16:00:38 -07:00
liqfu ae163b33a8 ONNX support for CNTK Sequence::Slice op. Due to a specific feature of CNTK Sequence::Slice op to handle zigged inputs (batch of variable length sequences), models converted from CNTK may not produce marching outcomes with zigged inputs. 2018-08-14 11:17:36 -07:00
Bowen Bao ceaec5636f fix squeezenet onnx_backend_test of issues with Softmax, hardmax, logsoftmax and flatten with freedimension on batchaxis. 2018-08-14 10:02:07 -07:00
Spandan Tiwari 541e2100eb Merge branch 'master' into sptiwari/model_fix_test 2018-08-13 10:48:20 -07:00
Thiago Crepaldi b689ebb5ad Ignore TEST_TAG environment variable when not defined during testing 2018-08-13 09:02:01 -07:00
Thiago Crepaldi 4ce6607917 Disable fmeasure test and increase pytest verbosity 2018-08-12 13:24:20 -07:00
Spandan Tiwari b73670c3bd Removing support for user-provided node names and using uid for ONNX names. 2018-08-10 16:53:58 -07:00
Spandan Tiwari e0b26561b8 Fixing batch axis handling bug in ReduceElements and Reshape export. 2018-08-10 16:36:29 -07:00
Spandan Tiwari ff118cc684 Merge branch 'master' into sptiwari/onnx_merge_with_master 2018-08-10 10:10:06 -07:00
delzac b7811d3549 Fmeasure (#3343) 
* amended a bug, swapped precision and recall

* added a bit more documentation
 2018-08-10 08:50:42 -07:00
Vadim Mazalov 2a748e1ce9 Merge branch 'vadimma/lc-blstm-sq' 2018-08-09 02:28:38 +00:00
Sergii Dymchenko a87d69a778 Fix Hardmax/Softmax/LogSoftmax ONNX export. 2018-08-08 16:52:33 -07:00
Binbin Zhang 1ebdc548f3 Introduce latency-controlled BLSTM 2018-08-08 15:41:51 -07:00
kaituoxu e5006a4866 fix parameter sharing bug 2018-08-06 19:00:33 +09:00
rpengms d2b49543eb
Add more logging into HTKDeserializer 2018-08-03 12:44:51 -07:00
liqfu 5e871a19fe treat sequence axis as static axis for ONNX RNN model import 2018-08-02 10:51:05 -07:00
Spandan Tiwari ce7075b03c Updating input image size requirement to include lower and upper paddingin ConvolveGeometry. 2018-08-01 12:00:54 -07:00
Ke Deng 917173d990 Merge branch 'pull/3331' 2018-08-01 05:50:14 +00:00
Ke Deng 6f7ceed3bf Merge branch 'pull/3284' 2018-08-01 04:48:10 +00:00
Bowen Bao 360fbd4635 Update the check in convolution if kernel fill the entire input space. 2018-07-31 18:19:45 -07:00
Bowen Bao 279a466b5f Squash of the following 
1. fix gather and prelu arguments order issue.
2. add typenameToTypeProto map initialization.
3. add CNTK native default lotus logger.
4. add CNTK unsqueeze op.
 2018-07-30 19:58:55 -07:00
Bowen Bao c78f40d0c1 Avoid asym pad for MKL, block channel axis padding. 2018-07-30 19:50:35 -07:00
Namgoo Lee dc65c2b57d Fix overlapping memcpy reported by valgrind in SmallVector 
Valgrind sometimes reports "Source and destination overlap in memcpy" on
the overloaded = operator of SmallVector class. Whenever it is reported,
the src address and the dest address of memcpy is actually identical.

The solution in this commit is to check the src and dest address first,
and do memcpy only if they are not identical. This should be enough to
handle the issue, since the only case they overlap is when they are
identical, provided that the m_data is a member of a class.
 2018-07-31 10:03:17 +09:00