microsoft
/
CNTK
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность
10 869 коммитов 1 074 Ветки 39 Теги 1,1 GiB
Граф коммитов

26 Коммитов

Автор SHA1 Сообщение Дата
Vadim Mazalov faef2b540a Quantized multiplier v1 2016-11-27 21:25:26 -08:00
Eldar Akchurin 773dca2b4c First version of GPU prefetch 2016-09-21 12:00:04 +02:00
Jasha Droppo b99b3832fb CuDNN-RNN Fix Merge Error in Math/*filters Visual Studio files 2016-08-16 15:50:56 -07:00
Jasha Droppo 80d077054d Merge branch 'master' into jdroppo/cudnn-rnn-lstm 2016-08-15 16:11:16 -07:00
Vadim Mazalov 77ff661930 Quantization of learnable parameter node 2016-08-01 12:28:11 -07:00
Jasha Droppo a4e42744c2 Merge branch 'master' into jdroppo/cudnn-rnn-lstm 
Conflicts:
	Makefile
	Source/CNTK/BrainScript/CNTKCoreLib/CNTK.core.bs
	Source/Math/CuDnnBatchNormalization.cu
	Source/Math/CuDnnConvolutionEngine.cu
	Source/Math/Math.vcxproj
	Source/SGDLib/SGD.cpp
 2016-07-18 11:11:58 -07:00
Anthony Aue 12140c993b Makefile and *.vcxproj changes 2016-06-17 08:57:55 -07:00
Anthony Aue e1b9d4a464 First stage of adding block multiplier to CNTK: check in files, verify 
they compile, and add a unit test to exercise them (minimally)
 2016-06-14 14:49:33 -07:00
Jasha Droppo 5558144363 Merge remote-tracking branch 'origin/master' into jdroppo/cudnn-rnn-lstm 
Conflicts:
	Source/Math/Math.vcxproj
	Source/Math/Math.vcxproj.filters
	Source/Math/MathCUDA.vcxproj.filters
 2016-05-20 11:20:07 -07:00
Clemens Marschner f755adafa6 Add extended eval interface that can take sparse matrices 2016-05-18 17:55:10 +02:00
Jasha Droppo 0fb69adc0b Centralize CuDnn RNN Parameters 
New RNNParameters struct centralizes RNN configurable parameters. It gets
passed around to ensure consistency between forward and backward passes.

All configurable parameters are now exposed to the BS RNN Node.
 2016-05-14 16:40:00 -07:00
Amit Agarwal f4204ce1cd Change in dropout randomization implementation to ensure reproducability across checkpoints 2016-05-05 12:17:10 -07:00
Alexey Kamenev 6ce80fbd15 Moved batch norm out of convo engine. Refactored unit tests. 2016-03-30 09:56:43 -07:00
Alexey Kamenev cee4cfa6f8 Added NDConvolution node. Added ConvolveGeometry. 2016-03-30 09:53:44 -07:00
Gaizka Navarro 09dea188f5 Adds ExceptionWithCallStack class to wrap native exceptions (runtime_error, logic_error, invalid_arguent), so that the call stack can be provided to the managed layer. 
Moves DebugUtil functions (PrintCallStack, GetCallStack) to ExceptionWithCallStack
Refactors the PrintCallStack and GetCallStack functions in ExceptionWithCallBack to re-use the common functionality.
Make gcc happy
 2016-02-23 09:11:12 +01:00
Mark Hillebrand 1bcefe770a .gitattributes: set proper configuration for text files (and normalize files) 2016-01-18 09:35:54 +01:00
Amit Agarwal b66ea69666 Remove 1bit SGD implementation headers references from the CNTK project files 2016-01-13 14:46:33 -08:00
Amit Agarwal 8b03b5435f Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/separate1bitDataParallelSGD 2016-01-06 18:39:06 -08:00
Frank Seide 306e857a5c renamed DataTensor.h to TensorShape.h. No code changes 2016-01-06 08:18:32 -08:00
Amit Agarwal c944295144 Some refactoring in preparation for separating out the 1bit gradient aggregation implementation from the core CNTK sources 2016-01-04 22:54:13 -08:00
Frank Seide 22bab75d5f GPUMatrix.cu split: moved TensorView support to separate compilation unit (GPUMatrix had gotten too large for MSVC to compile the fatbin file). No code change otherwise 2015-12-28 19:45:36 -08:00
Frank Seide bdfcf91923 added a heuristic to TensorShape::Load() that allows to read old models created when sample layouts were not fully consistent 2015-12-22 15:04:52 -08:00
Frank Seide b98855a9a1 tensor library works in PlusNode prototype implementation() which reduces ForwardProp() to two lines of code; 
bug fix: GetTensorsForwardBinary() ignored #samples in FrameRange;
bug fix: TensorView cannot hold a reference to a Matrix object since these are often temporaries. Instead, we copy a Matrix object that itself is a reference;
deleted orphaned ClassDiagram.cd from Math Project
 2015-12-18 19:59:26 -08:00
Frank Seide 91eadb0587 moved all tensor ops to a new header TensorOps.h so they can be shared between matrix types; 
also moved the float/double-unified math overloads (e.g. exp_()) there, as well as additional typically needed functions such as Sigmoid()
 2015-12-18 08:54:19 -08:00
Philipp Kranen 33bb02c225 Fixed vcxproj.filters wrt restructuring 2015-12-16 09:42:02 +01:00
pkranen c3895bb0b3 Move Source files for directory restructuring and fix project dependencies 2015-12-16 08:35:11 +01:00