Merge branch 'master' of https://git01.codeplex.com/cntk into linux-gcc
This commit is contained in:
Dong Yu
Коммит
a0645c9cdf
|
|
@ -0,0 +1 @@
|
|||
run-test text eol=lf
|
||||
|
|
@ -6,6 +6,11 @@
|
|||
*.user
|
||||
*.sln.docstates
|
||||
*.orig
|
||||
\#*
|
||||
.\#*
|
||||
|
||||
# Local build configuration
|
||||
Config.make
|
||||
|
||||
# Build results
|
||||
|
||||
|
|
@ -16,6 +21,8 @@ build/
|
|||
[Bb]in/
|
||||
[Oo]bj/
|
||||
.run-*
|
||||
lib/
|
||||
bin/
|
||||
|
||||
# Enable "build/" folder in the NuGet Packages folder since NuGet packages use it for MSBuild targets
|
||||
!packages/*/build/
|
||||
|
|
|
|||
|
|
@ -54,7 +54,7 @@ typedef void* HANDLE;
|
|||
#define __forceinline inline
|
||||
//string and io conversion
|
||||
#define strtok_s strtok_r
|
||||
#define sprintf_s sprintf
|
||||
#define sprintf_s snprintf
|
||||
#define sscanf_s sscanf
|
||||
#define _strdup strdup
|
||||
|
||||
|
|
|
|||
|
|
@ -24,12 +24,12 @@ namespace Microsoft { namespace MSR { namespace CNTK {
|
|||
static const std::string::size_type npos = (std::string::size_type) -1;
|
||||
|
||||
// These are the constants associated with the "ResolveVariables" method.
|
||||
static const std::string openBraceVar = "$";
|
||||
static const std::string closingBraceVar = "$";
|
||||
static const std::string forbiddenCharactersInVarName = ",/<>?;':\"[]{}\\|!@#%^&*()+=~` \t\n";
|
||||
static const std::string forbiddenCharactersInVarNameEscapeWhitespace = ",/<>?;':\"[]{}\\|!@#%^&*()+=~` \\t\\n";
|
||||
static const std::size_t openBraceVarSize = openBraceVar.size();
|
||||
static const std::size_t closingBraceVarSize = openBraceVar.size();
|
||||
static const char* openBraceVar = "$";
|
||||
static const char* closingBraceVar = "$";
|
||||
static const char* forbiddenCharactersInVarName = ",/<>?;':\"[]{}\\|!@#%^&*()+=~` \t\n";
|
||||
static const char* forbiddenCharactersInVarNameEscapeWhitespace = ",/<>?;':\"[]{}\\|!@#%^&*()+=~` \\t\\n";
|
||||
static const std::size_t openBraceVarSize = strlen(openBraceVar);
|
||||
static const std::size_t closingBraceVarSize = strlen(closingBraceVar);
|
||||
|
||||
// Trim - trim white space off the start and end of the string
|
||||
// str - string to trim
|
||||
|
|
@ -991,7 +991,7 @@ public:
|
|||
while (start != std::string::npos)
|
||||
{
|
||||
// search for whitespace or closing brace.
|
||||
end = newConfigLine.find_first_of(closingBraceVar + forbiddenCharactersInVarName,
|
||||
end = newConfigLine.find_first_of(std::string(closingBraceVar) + forbiddenCharactersInVarName,
|
||||
start + openBraceVarSize);
|
||||
|
||||
// ensure that a closing brace exists for every opening brace.
|
||||
|
|
@ -999,15 +999,15 @@ public:
|
|||
if (end == std::string::npos)
|
||||
{
|
||||
RuntimeError("\"%s\" found without corresponding closing \"%s\": %s:%s",
|
||||
openBraceVar.c_str(), closingBraceVar.c_str(),
|
||||
openBraceVar, closingBraceVar,
|
||||
m_configName.c_str(), newConfigLine.c_str());
|
||||
}
|
||||
|
||||
if (newConfigLine[end] != '$')
|
||||
{
|
||||
RuntimeError("Forbidden characters found between \"%s\" and \"%s\". Variable names cannot any of the following characters: %s. %s:%s",
|
||||
openBraceVar.c_str(), closingBraceVar.c_str(),
|
||||
forbiddenCharactersInVarNameEscapeWhitespace.c_str(),
|
||||
openBraceVar, closingBraceVar,
|
||||
forbiddenCharactersInVarNameEscapeWhitespace,
|
||||
m_configName.c_str(), newConfigLine.c_str());
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -138,7 +138,7 @@ void BinaryReader<ElemType>::DisplayProperties()
|
|||
for (auto pair : m_sections)
|
||||
{
|
||||
Section* section = pair.second;
|
||||
fprintf(stderr,"Section: %ls, Elements: %lld, ElementsPerRecord: %lld, ElementSize: %lld\n", pair.first.c_str(), section->GetElementCount(), section->GetElementsPerRecord(), section->GetElementSize());
|
||||
fprintf(stderr,"Section: %ls, Elements: %zd, ElementsPerRecord: %zd, ElementSize: %zd\n", pair.first.c_str(), section->GetElementCount(), section->GetElementsPerRecord(), section->GetElementSize());
|
||||
if (section->GetSectionType() == sectionTypeStats)
|
||||
{
|
||||
vector<NumericStatistics> stats;
|
||||
|
|
@ -434,4 +434,4 @@ bool BinaryReader<ElemType>::DataEnd(EndDataType endDataType)
|
|||
// instantiate all the combinations we expect to be used
|
||||
template class BinaryReader<double>;
|
||||
template class BinaryReader<float>;
|
||||
}}}
|
||||
}}}
|
||||
|
|
|
|||
|
|
@ -1161,11 +1161,8 @@ namespace Microsoft { namespace MSR { namespace CNTK {
|
|||
// Sets the utterance boundary.
|
||||
if (m_framemode == false)
|
||||
{
|
||||
// If <m_truncated> is false, then the whole utterance
|
||||
// will be loaded into the minibatch.
|
||||
if (m_truncated == false)
|
||||
if (startFrame == 0)
|
||||
{
|
||||
assert(startFrame == 0);
|
||||
m_sentenceBegin.SetValue(i, 0, (ElemType)SEQUENCE_START);
|
||||
m_minibatchPackingFlag[0] |= MinibatchPackingFlag::SequenceStart;
|
||||
}
|
||||
|
|
@ -1227,11 +1224,8 @@ namespace Microsoft { namespace MSR { namespace CNTK {
|
|||
// minibatch, and then load it.
|
||||
if (m_framemode == false)
|
||||
{
|
||||
// If <m_truncated> is false, then the whole utterance
|
||||
// will be loaded into the minibatch.
|
||||
if (m_truncated == false)
|
||||
if (startFrame == 0)
|
||||
{
|
||||
assert(startFrame == 0);
|
||||
m_sentenceBegin.SetValue(i, 0, (ElemType)SEQUENCE_START);
|
||||
m_minibatchPackingFlag[0] |= MinibatchPackingFlag::SequenceStart;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -445,7 +445,7 @@ public:
|
|||
long TickDelta = TickStop - TickStart;
|
||||
|
||||
if (m_traceLevel > 2)
|
||||
fprintf(stderr, "\n%l ms, %d numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
fprintf(stderr, "\n%ld ms, %d numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
return lineCount;
|
||||
}
|
||||
|
||||
|
|
@ -608,7 +608,7 @@ public:
|
|||
long TickDelta = TickStop - TickStart;
|
||||
|
||||
if (m_traceLevel > 2)
|
||||
fprintf(stderr, "\n%l ms, %d numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
fprintf(stderr, "\n%ld ms, %d numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
return lineCount;
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -795,7 +795,7 @@ void SequenceReader<ElemType>::SetupEpoch()
|
|||
if (m_totalSamples == 0)
|
||||
{
|
||||
if (m_traceLevel > 0)
|
||||
fprintf(stderr, "starting at epoch %d parsing all data to determine record count\n", m_epoch);
|
||||
fprintf(stderr, "starting at epoch %zd parsing all data to determine record count\n", m_epoch);
|
||||
// choose a large number to read
|
||||
m_parser.SetFilePosition(0);
|
||||
m_mbStartSample = 0;
|
||||
|
|
@ -805,7 +805,7 @@ void SequenceReader<ElemType>::SetupEpoch()
|
|||
m_seqIndex = m_sequence.size();
|
||||
}
|
||||
if (m_traceLevel > 0)
|
||||
fprintf(stderr, "\n %lld records found\n", m_totalSamples);
|
||||
fprintf(stderr, "\n %zd records found\n", m_totalSamples);
|
||||
}
|
||||
m_seqIndex = 0;
|
||||
|
||||
|
|
@ -2129,4 +2129,4 @@ int BatchSequenceReader<ElemType>::GetSentenceEndIdFromOutputLabel()
|
|||
|
||||
template class BatchSequenceReader<double>;
|
||||
template class BatchSequenceReader<float>;
|
||||
}}}
|
||||
}}}
|
||||
|
|
|
|||
|
|
@ -653,7 +653,7 @@ long UCIParser<NumType, LabelType>::Parse(size_t recordsRequested, std::vector<N
|
|||
long TickDelta = TickStop - TickStart;
|
||||
|
||||
if (m_traceLevel > 2)
|
||||
fprintf(stderr, "\n%ld ms, %d numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
fprintf(stderr, "\n%ld ms, %ld numbers parsed\n\n", TickDelta, m_totalNumbersConverted );
|
||||
return recordCount;
|
||||
}
|
||||
|
||||
|
|
|
|||
Разница между файлами не показана из-за своего большого размера
Загрузить разницу
Разница между файлами не показана из-за своего большого размера
Загрузить разницу
|
|
@ -25,7 +25,7 @@
|
|||
\font_roman times
|
||||
\font_sans helvet
|
||||
\font_typewriter courier
|
||||
\font_math auto
|
||||
\font_math default
|
||||
\font_default_family default
|
||||
\use_non_tex_fonts false
|
||||
\font_sc false
|
||||
|
|
@ -56,11 +56,11 @@
|
|||
\use_package cancel 0
|
||||
\use_package esint 1
|
||||
\use_package mathdots 1
|
||||
\use_package mathtools 0
|
||||
\use_package mathtools 1
|
||||
\use_package mhchem 1
|
||||
\use_package stackrel 0
|
||||
\use_package stmaryrd 0
|
||||
\use_package undertilde 0
|
||||
\use_package stackrel 1
|
||||
\use_package stmaryrd 1
|
||||
\use_package undertilde 1
|
||||
\cite_engine basic
|
||||
\cite_engine_type default
|
||||
\biblio_style plain
|
||||
|
|
@ -99,23 +99,35 @@ An Introduction to Computational Networks and the Computational Network
|
|||
|
||||
\begin_layout Author
|
||||
Dong Yu, Adam Eversole, Michael L.
|
||||
Seltzer, Kaisheng Yao, Zhiheng Huang,
|
||||
Seltzer, Kaisheng Yao,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Brian Guenter, Oleksii Kuchaiev, Yu Zhang, Frank Seide, Huaming Wang,
|
||||
Brian Guenter, Oleksii Kuchaiev, Yu Zhang, Frank Seide, Guoguo Chen,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Jasha Droppo, Geoffrey Zweig, Chris Rossbach, Jon Currey,
|
||||
Huaming Wang, Jasha Droppo, Amit Agarwal, Chris Basoglu,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Jie Gao, Avner May, Baolin Peng, Andreas Stolcke, Malcolm Slaney
|
||||
Marko Padmilac, Alexey Kamenev, Vladimir Ivanov, Scott Cypher,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Hari Parthasarathi, Bhaskar Mitra, Zhiheng Huang, Geoffrey Zweig,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Chris Rossbach, Jon Currey,Jie Gao, Avner May, Baolin Peng,
|
||||
\begin_inset Newline newline
|
||||
\end_inset
|
||||
|
||||
Andreas Stolcke, Malcolm Slaney, Xuedong Huang
|
||||
\end_layout
|
||||
|
||||
\begin_layout Date
|
||||
MSR-TR-2014-112 (DRAFT v0.8: May 19, 2015)
|
||||
MSR-TR-2014-112 (DRAFT v0.8: Aug 5, 2015)
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
|
|
|
|||
|
|
@ -2073,6 +2073,45 @@ SumElements(m)
|
|||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
m - input matrix
|
||||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
SumColumnElements
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
SumColumnElements
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
Calculate the sum of all elements in each column of the input matrix.
|
||||
The result is a row vector.
|
||||
The syntax is
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
\begin_inset listings
|
||||
inline false
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
SumColumnElements(m)
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
|
|
@ -2118,6 +2157,55 @@ Negate(m)
|
|||
m - input matrix.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
Reshape
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
Reshape
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
Reshape the input matrix.
|
||||
The resulting matrix has the same number of elements as that of the input
|
||||
matrix but is interpreted differently.
|
||||
The syntax is
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
\begin_inset listings
|
||||
inline false
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
Reshape(m, numRows, [imageWidth=], [imageHeight=], [imageChannels=]
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
m - input matrix.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
numRows - reshape to a matrix with numRows rows.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
imageWidth, imageHeight, imageChannels - are named optional parameters that
|
||||
allow you to interpret each column as images with different dimensions.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
RowSlice
|
||||
\begin_inset Index idx
|
||||
|
|
@ -2401,23 +2489,13 @@ Minus
|
|||
|
||||
\end_inset
|
||||
|
||||
, ElementTimes
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
ElementTimes
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
Calculate the sum (Plus), difference (Minus), or element-wise product (ElementTi
|
||||
mes) of two matrices.
|
||||
The resulting matrices have the same dimension as that of the input matrices.
|
||||
Calculate the sum (Plus) and difference (Minus) of two matrices.
|
||||
The resulting matrices have the same dimension as that of the input matrix
|
||||
with larger dimension.
|
||||
The syntax is
|
||||
\end_layout
|
||||
|
||||
|
|
@ -2436,19 +2514,95 @@ Plus(m1, m2)
|
|||
Minus(m1, m2)
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
m1, m2 - input matrices.
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
ElementTimes
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
ElementTimes
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
, RowElementTimes
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
RowElementTimes
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
, ColumnElementTimes
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
ColumnElementTimes
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
,
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
Calculate the element-wise product of two matrices.
|
||||
The resulting matrices have the same dimension as that of the input matrix
|
||||
with larger dimension.
|
||||
The syntax is
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
\begin_inset listings
|
||||
inline false
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
ElementTimes(m1, m2)
|
||||
\end_layout
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
RowElementTimes(m, vrow)
|
||||
\end_layout
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
ColumnElementTimes(m, vcol)
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
m1, m2 - input matrices.
|
||||
Must be the same dimensions.
|
||||
m1, m2, m - input matrices.
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
vrow, vcol - row and column vectors
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
m1 and m2 must have same dimension in ElementTimes.
|
||||
m and vrow must have same number of columns in RowElementTimes, and m and
|
||||
vcol must have same number of rows in ColumnElementTimes.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
|
|
@ -3547,7 +3701,7 @@ stepH - step (or stride) used in the height direction
|
|||
\end_layout
|
||||
|
||||
\begin_layout Subsubsection
|
||||
Delay
|
||||
PastValue (or Delay
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
|
|
@ -3557,11 +3711,21 @@ Delay
|
|||
|
||||
\end_inset
|
||||
|
||||
) and FutureValue
|
||||
\begin_inset Index idx
|
||||
status open
|
||||
|
||||
\begin_layout Plain Layout
|
||||
FutureValue
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
||||
|
||||
\end_layout
|
||||
|
||||
\begin_layout Standard
|
||||
Used to apply a value in the past to the current time.
|
||||
Used to get the past or future value of a node.
|
||||
It is most often used to create recurrent networks.
|
||||
The resulting matrix has the same dimension as that of the input matrix.
|
||||
The syntax is
|
||||
|
|
@ -3574,7 +3738,17 @@ status open
|
|||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
Delay(rows, [cols], m, [delayTime=1, defaultPastValue=0.1])
|
||||
PastValue(rows, [cols], m, [timeStep=1, defaultHiddenActivity=0.1])
|
||||
\end_layout
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
Delay(rows, [cols], m, [delayTime=1, defaultPastValue=0.1]) #deprecated
|
||||
\end_layout
|
||||
|
||||
\begin_layout Plain Layout
|
||||
|
||||
FutureValue(rows, [cols], m, [timeStep=1, defaultHiddenActivity=0.1])
|
||||
\end_layout
|
||||
|
||||
\end_inset
|
||||
|
|
@ -3583,19 +3757,19 @@ Delay(rows, [cols], m, [delayTime=1, defaultPastValue=0.1])
|
|||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
rows - the number of rows in the delay node (and in the input matrix).
|
||||
rows - the number of rows in the input matrix.
|
||||
This parameter is needed because under some loopy conditions the dimensions
|
||||
cannot be automatically inferred from the input matrix.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
cols - the number of columns in the delay node (and in the input matrix).
|
||||
cols - the number of columns in the input matrix.
|
||||
This parameter is optional since it will be set based on the minibatch
|
||||
size during training and testing.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
m - input matrix to be delayed.
|
||||
m - input matrix from which the past or future value is obtained.
|
||||
Each column is a sample.
|
||||
The samples may be from different utterances as explained in Chapter .
|
||||
\begin_inset CommandInset ref
|
||||
|
|
@ -3608,13 +3782,14 @@ reference "chap:CN"
|
|||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
delayTime - [named optional] the amount of delay.
|
||||
timeStep, delayTime - [named optional] the amount of time steps look into
|
||||
past or future.
|
||||
Default is 1.
|
||||
\end_layout
|
||||
|
||||
\begin_layout Itemize
|
||||
defaultPastValue - [named optional] the default value to use if the past
|
||||
value is not available.
|
||||
defaultHiddenActivity, defaultPastValue - [named optional] the default value
|
||||
to use if the past or future values are not available.
|
||||
Default is 0.1.
|
||||
\end_layout
|
||||
|
||||
|
|
|
|||
|
|
@ -60,6 +60,9 @@ using namespace Microsoft::MSR::CNTK;
|
|||
template <typename ElemType>
|
||||
void TestCn(const ConfigParameters& config);
|
||||
|
||||
template <typename ElemType>
|
||||
void DoEvalBeamSearch(const ConfigParameters& config, IDataReader<ElemType>& reader);
|
||||
|
||||
template <typename T>
|
||||
struct compare_second
|
||||
{
|
||||
|
|
@ -726,41 +729,37 @@ void DoTrain(const ConfigParameters& config)
|
|||
ConfigParameters readerConfig(config("reader"));
|
||||
readerConfig.Insert("traceLevel", config("traceLevel", "0"));
|
||||
|
||||
IComputationNetBuilder<ElemType>* netBuilder = NULL;
|
||||
unique_ptr<IComputationNetBuilder<ElemType> > netBuilder;
|
||||
|
||||
if (config.Exists("NDLNetworkBuilder"))
|
||||
{
|
||||
ConfigParameters configNDL(config("NDLNetworkBuilder"));
|
||||
netBuilder = (IComputationNetBuilder<ElemType>*)new NDLBuilder<ElemType>(configNDL);
|
||||
netBuilder = unique_ptr<IComputationNetBuilder<ElemType> >( static_cast<IComputationNetBuilder<ElemType>*>(new NDLBuilder<ElemType>(configNDL)));
|
||||
}
|
||||
else if (config.Exists("SimpleNetworkBuilder"))
|
||||
{
|
||||
ConfigParameters configSNB(config("SimpleNetworkBuilder"));
|
||||
netBuilder = (IComputationNetBuilder<ElemType>*)new SimpleNetworkBuilder<ElemType>(configSNB);
|
||||
netBuilder = unique_ptr<IComputationNetBuilder<ElemType> >{ static_cast<IComputationNetBuilder<ElemType>*>(new SimpleNetworkBuilder<ElemType>(configSNB)) };
|
||||
}
|
||||
else
|
||||
{
|
||||
RuntimeError("No network builder found in the config file. NDLNetworkBuilder or SimpleNetworkBuilde must be specified");
|
||||
}
|
||||
|
||||
DataReader<ElemType>* dataReader = new DataReader<ElemType>(readerConfig);
|
||||
unique_ptr<DataReader<ElemType> > dataReader { new DataReader<ElemType>(readerConfig) };
|
||||
|
||||
DataReader<ElemType>* cvDataReader = nullptr;
|
||||
unique_ptr<DataReader<ElemType> > cvDataReader;
|
||||
ConfigParameters cvReaderConfig(config("cvReader", L""));
|
||||
|
||||
if (cvReaderConfig.size() != 0)
|
||||
{
|
||||
cvReaderConfig.Insert("traceLevel", config("traceLevel", "0"));
|
||||
cvDataReader = new DataReader<ElemType>(cvReaderConfig);
|
||||
cvDataReader = unique_ptr<DataReader<ElemType> >{ new DataReader<ElemType>(cvReaderConfig) };
|
||||
}
|
||||
|
||||
SGD<ElemType> sgd(configSGD);
|
||||
|
||||
sgd.Train(netBuilder, dataReader, cvDataReader, makeMode);
|
||||
|
||||
delete netBuilder;
|
||||
delete dataReader;
|
||||
delete cvDataReader;
|
||||
sgd.Train(netBuilder.get(), dataReader.get(), cvDataReader.get(), makeMode);
|
||||
}
|
||||
|
||||
template <typename ElemType>
|
||||
|
|
@ -1477,7 +1476,8 @@ int wmain(int argc, wchar_t* argv[])
|
|||
fcloseOrDie(fp);
|
||||
}
|
||||
fprintf(stderr, "COMPLETED\n");
|
||||
}
|
||||
fflush(stderr);
|
||||
}
|
||||
catch (const std::exception &err)
|
||||
{
|
||||
fprintf(stderr, "EXCEPTION occurred: %s\n", err.what());
|
||||
|
|
|
|||
|
|
@ -1515,7 +1515,7 @@ public:
|
|||
ComputationNodePtr newNode(new PairNetworkNode<ElemType>(m_deviceId, nodeName));
|
||||
if (this->GetNodeFromName(a->NodeName(), nullptr, false) != nullptr)
|
||||
{
|
||||
fprintf(stderr, "PairNetwork : asked to pair a node with name l%s in another network.However, this network has already a node with the same name.Should avoid this case.\n", a->NodeName().c_str());
|
||||
fprintf(stderr, "PairNetwork : asked to pair a node with name %ls in another network.However, this network has already a node with the same name.Should avoid this case.\n", a->NodeName().c_str());
|
||||
throw std::runtime_error("PairNetwork : asked to pair a node with name in another network.However, this network has already a node with the same name.Should avoid this case.\n");
|
||||
}
|
||||
newNode->AttachInputs(a);
|
||||
|
|
@ -2441,8 +2441,7 @@ public:
|
|||
}
|
||||
|
||||
int FindInRecurrentLoop(const ComputationNodePtr startNode,
|
||||
std::vector<ComputationNodePtr>& recurrentNodes,
|
||||
bool isForwardComputing = false)
|
||||
std::vector<ComputationNodePtr>& recurrentNodes)
|
||||
{
|
||||
int iFound = -1;
|
||||
|
||||
|
|
@ -2451,14 +2450,8 @@ public:
|
|||
if (std::find((*iter).m_recurrentNodes.begin(), (*iter).m_recurrentNodes.end(), startNode) != (*iter).m_recurrentNodes.end())
|
||||
{
|
||||
iFound = (*iter).m_loopId;
|
||||
if (isForwardComputing)
|
||||
{
|
||||
recurrentNodes = (*iter).m_recurrentNodesForForward;
|
||||
}
|
||||
else
|
||||
{
|
||||
recurrentNodes = (*iter).m_recurrentNodesForForward;
|
||||
}
|
||||
recurrentNodes = (*iter).m_recurrentNodesForForward;
|
||||
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
|
@ -2499,7 +2492,7 @@ public:
|
|||
void EvaluateLoop(std::list<ComputationNodePtr>& /*allNodes*/, const ComputationNodePtr startNode)
|
||||
{
|
||||
std::vector<ComputationNodePtr> recurrentNodes;
|
||||
int iLoopId = FindInRecurrentLoop(startNode, recurrentNodes, true);
|
||||
int iLoopId = FindInRecurrentLoop(startNode, recurrentNodes);
|
||||
if (iLoopId != -1 && IsFuncValueOlderThanInputs(recurrentNodes) &&
|
||||
m_recurrentInfo[iLoopId].m_completedEvaluate == false)
|
||||
{
|
||||
|
|
@ -3337,7 +3330,7 @@ public:
|
|||
|
||||
bool UnitTest(const ComputationNodePtr rootNode)
|
||||
{
|
||||
fprintf(stderr, "\n\n Unit test node %ws \n", rootNode->NodeName().c_str());
|
||||
fprintf(stderr, "\n\n Unit test node %ls \n", rootNode->NodeName().c_str());
|
||||
|
||||
std::list<ComputationNodePtr>& nodes = GetEvalOrder(rootNode);
|
||||
|
||||
|
|
|
|||
|
|
@ -33,41 +33,50 @@ namespace Microsoft {
|
|||
namespace MSR {
|
||||
namespace CNTK {
|
||||
|
||||
extern std::wstring GetEncoderModelNameForEpoch(int epoch, bool b = false);
|
||||
extern std::wstring GetDecoderModelNameForEpoch(int epoch, bool b = false);
|
||||
|
||||
template<class ElemType>
|
||||
class MultiNetworksSGD : SGD<ElemType>
|
||||
{
|
||||
ElemType m_default_activity;
|
||||
|
||||
typedef SGD<ElemType> SGD;
|
||||
using SGDBase = SGD<ElemType>;
|
||||
|
||||
public:
|
||||
using SGD::m_modelPath;
|
||||
using SGD::m_maxEpochs;
|
||||
using SGD::m_doUnitTest;
|
||||
using SGD::m_learnRateAdjustInterval;
|
||||
using SGD::m_mbSize;
|
||||
using SGD::m_momentumPerSample;
|
||||
using SGD::m_learningRatesPerSample;
|
||||
using SGD::m_dropoutRates;
|
||||
using SGD::m_autoLearnRateSearchType;
|
||||
using SGD::m_minLearnRate;
|
||||
using SGD::m_loadBestModel;
|
||||
using SGD::m_validateAfterModelReloading;
|
||||
using SGD::m_continueReduce;
|
||||
using SGD::m_reduceLearnRateIfImproveLessThan;
|
||||
using SGD::m_epochSize;
|
||||
using SGD::m_learnRateDecreaseFactor;
|
||||
using SGD::m_increaseLearnRateIfImproveMoreThan;
|
||||
using SGD::m_learnRateIncreaseFactor;
|
||||
using SGD::m_keepCheckPointFiles;
|
||||
using SGD::m_doGradientCheck;
|
||||
using SGD::m_L2RegWeight;
|
||||
using SGD::m_L1RegWeight;
|
||||
using SGD::m_needAveMultiplier;
|
||||
using SGD::m_traceLevel;
|
||||
using SGD::m_numMBsToShowResult;
|
||||
using SGD::m_gradientCheckSigDigit;
|
||||
using SGD::m_prevChosenMinibatchSize;
|
||||
using SGDBase::m_modelPath;
|
||||
using SGDBase::m_maxEpochs;
|
||||
using SGDBase::m_doUnitTest;
|
||||
using SGDBase::m_learnRateAdjustInterval;
|
||||
using SGDBase::m_mbSize;
|
||||
using SGDBase::m_momentumPerSample;
|
||||
using SGDBase::m_learningRatesPerSample;
|
||||
using SGDBase::m_dropoutRates;
|
||||
using SGDBase::m_autoLearnRateSearchType;
|
||||
using SGDBase::m_minLearnRate;
|
||||
using SGDBase::m_loadBestModel;
|
||||
using SGDBase::m_validateAfterModelReloading;
|
||||
using SGDBase::m_continueReduce;
|
||||
using SGDBase::m_reduceLearnRateIfImproveLessThan;
|
||||
using SGDBase::m_epochSize;
|
||||
using SGDBase::m_learnRateDecreaseFactor;
|
||||
using SGDBase::m_increaseLearnRateIfImproveMoreThan;
|
||||
using SGDBase::m_learnRateIncreaseFactor;
|
||||
using SGDBase::m_keepCheckPointFiles;
|
||||
using SGDBase::m_doGradientCheck;
|
||||
using SGDBase::m_L2RegWeight;
|
||||
using SGDBase::m_L1RegWeight;
|
||||
using SGDBase::m_needAveMultiplier;
|
||||
using SGDBase::m_traceLevel;
|
||||
using SGDBase::m_numMBsToShowResult;
|
||||
using SGDBase::m_gradientCheckSigDigit;
|
||||
using SGDBase::m_prevChosenMinibatchSize;
|
||||
using SGDBase::GetTrainCriterionNodes;
|
||||
using SGDBase::GetEvalCriterionNodes;
|
||||
using SGDBase::SetDropoutRate;
|
||||
using SGDBase::UpdateEvalTimeStamps;
|
||||
using SGDBase::UpdateWeights;
|
||||
using SGDBase::GetCheckPointFileNameForEpoch;
|
||||
|
||||
typedef ComputationNode<ElemType>* ComputationNodePtr;
|
||||
|
||||
|
|
@ -80,7 +89,7 @@ namespace Microsoft {
|
|||
list<pair<ComputationNodePtr, ComputationNodePtr>> m_lst_pair_encoder_decoder_nodes;
|
||||
|
||||
public:
|
||||
MultiNetworksSGD(const ConfigParameters& configSGD) : SGD(configSGD)
|
||||
MultiNetworksSGD(const ConfigParameters& configSGD) : SGDBase(configSGD)
|
||||
{
|
||||
}
|
||||
|
||||
|
|
@ -663,7 +672,7 @@ namespace Microsoft {
|
|||
|
||||
if (learnRatePerSample < m_minLearnRate)
|
||||
{
|
||||
fprintf(stderr, "Learn Rate Per Sample for Epoch[%lu] = %.8g is less than minLearnRate %.8g. Training stops.\n", i + 1, learnRatePerSample, m_minLearnRate);
|
||||
fprintf(stderr, "Learn Rate Per Sample for Epoch[%d] = %.8g is less than minLearnRate %.8g. Training stops.\n", i + 1, learnRatePerSample, m_minLearnRate);
|
||||
break;
|
||||
}
|
||||
|
||||
|
|
@ -692,7 +701,7 @@ namespace Microsoft {
|
|||
IDataReader<ElemType>* decoderTrainSetDataReader = trainDataReader[decoderIdx];
|
||||
ComputationNetwork<ElemType>* decoderNet = nets[decoderIdx];
|
||||
|
||||
fprintf(stderr, "Finished Epoch[%lu]: [Training Set] Decoder Train Loss Per Sample = %.8g ", i + 1, epochCriterion);
|
||||
fprintf(stderr, "Finished Epoch[%d]: [Training Set] Decoder Train Loss Per Sample = %.8g ", i + 1, epochCriterion);
|
||||
if (epochEvalErrors.size() == 1)
|
||||
{
|
||||
fprintf(stderr, "EvalErr Per Sample = %.8g Ave Learn Rate Per Sample = %.10g Epoch Time=%.8g\n", epochEvalErrors[0], learnRatePerSample, epochTime);
|
||||
|
|
@ -703,9 +712,9 @@ namespace Microsoft {
|
|||
for (size_t j = 0; j<epochEvalErrors.size(); j++)
|
||||
fprintf(stderr, "[%lu]=%.8g ", j, epochEvalErrors[j]);
|
||||
fprintf(stderr, "Ave Learn Rate Per Sample = %.10g Epoch Time=%.8g\n", learnRatePerSample, epochTime);
|
||||
fprintf(stderr, "Finished Epoch[%lu]: Criterion Node Per Sample = %.8g\n", i + 1, epochCriterion);
|
||||
fprintf(stderr, "Finished Epoch[%d]: Criterion Node Per Sample = %.8g\n", i + 1, epochCriterion);
|
||||
for (size_t j = 0; j<epochEvalErrors.size(); j++)
|
||||
fprintf(stderr, "Finished Epoch[%lu]: Evaluation Node [%ws] Per Sample = %.8g\n", i + 1, evalNodeNames[j].c_str(), epochEvalErrors[j]);
|
||||
fprintf(stderr, "Finished Epoch[%d]: Evaluation Node [%ls] Per Sample = %.8g\n", i + 1, evalNodeNames[j].c_str(), epochEvalErrors[j]);
|
||||
}
|
||||
|
||||
if (decoderValidationSetDataReader != decoderTrainSetDataReader && decoderValidationSetDataReader != nullptr)
|
||||
|
|
@ -717,7 +726,7 @@ namespace Microsoft {
|
|||
validationDataReader,
|
||||
m_mbSize[i]);
|
||||
|
||||
fprintf(stderr, "Finished Epoch[%lu]: [Validation Set] Loss Per Sample = %.8g \n ", vScore );
|
||||
fprintf(stderr, "Finished Epoch[%d]: [Validation Set] Loss Per Sample = %.8g \n ", i+1, vScore );
|
||||
|
||||
epochCriterion = vScore;
|
||||
}
|
||||
|
|
@ -1013,7 +1022,7 @@ namespace Microsoft {
|
|||
{
|
||||
epochEvalErrors[i] = (const ElemType)localEpochEvalErrors(0, i);
|
||||
}
|
||||
fprintf(stderr, "total samples in epoch[%d] = %d\n", epochNumber, totalEpochSamples);
|
||||
fprintf(stderr, "total samples in epoch[%d] = %zd\n", epochNumber, totalEpochSamples);
|
||||
}
|
||||
|
||||
bool EncoderDecoderGradientCheck(
|
||||
|
|
@ -1053,7 +1062,7 @@ namespace Microsoft {
|
|||
irow = max(0, irow);
|
||||
icol = max(0, icol);
|
||||
|
||||
fprintf(stderr, "\n###### d%ws######\n", node->NodeName().c_str());
|
||||
fprintf(stderr, "\n###### d%ls######\n", node->NodeName().c_str());
|
||||
deviceId = node->FunctionValues().GetDeviceId(); // original device id
|
||||
|
||||
node->FunctionValues().TransferFromDeviceToDevice(deviceId, CPUDEVICE, true, false, false);
|
||||
|
|
@ -1124,7 +1133,7 @@ namespace Microsoft {
|
|||
if (wrong)
|
||||
{
|
||||
char serr[2048];
|
||||
sprintf_s((char*)serr, 2048, "Decoder %ws Numeric gradient = %e, Error BP gradient = %e", node->NodeName().c_str(), grdNum, grdErr);
|
||||
sprintf_s((char*)serr, 2048, "Decoder %ls Numeric gradient = %e, Error BP gradient = %e", node->NodeName().c_str(), static_cast<double>(grdNum), static_cast<double>(grdErr));
|
||||
fprintf(stdout, "%s\n", serr);
|
||||
verror_msgs.push_back(serr);
|
||||
}
|
||||
|
|
|
|||
|
|
@ -1532,7 +1532,7 @@ protected:
|
|||
if (epochNumber < 2 && m_prevChosenMinibatchSize != 0)
|
||||
{
|
||||
// newly started training: any previous MB size stored in the model is to be ignored
|
||||
fprintf(stderr, "before epoch .2, previous minibatchSize %d is "
|
||||
fprintf(stderr, "before epoch .2, previous minibatchSize %zd is "
|
||||
"considered invalid -> resetting\n", m_prevChosenMinibatchSize);
|
||||
m_prevChosenMinibatchSize = 0;
|
||||
}
|
||||
|
|
@ -1543,7 +1543,7 @@ protected:
|
|||
(epochNumber + 1) % m_minibatchSizeTuningFrequency != 0)
|
||||
{
|
||||
fprintf(stderr, "AdaptiveMinibatchSearch: Search for a better minibatchSize "
|
||||
"in epoch %d skipped, keeping minibatchSize of %d\n",
|
||||
"in epoch %d skipped, keeping minibatchSize of %zd\n",
|
||||
epochNumber + 1, m_prevChosenMinibatchSize);
|
||||
chosenMinibatchSize = m_prevChosenMinibatchSize;
|
||||
}
|
||||
|
|
@ -1568,7 +1568,7 @@ protected:
|
|||
assert(m_prevChosenMinibatchSize >= chosenMinibatchSize);
|
||||
|
||||
fprintf(stderr, "AdaptiveMinibatchSearch: Limiting maxMinibatchSize to "
|
||||
"previous minibatchSize %d*2\n", m_prevChosenMinibatchSize);
|
||||
"previous minibatchSize %zd*2\n", m_prevChosenMinibatchSize);
|
||||
maxMinibatchSize = min(maxMinibatchSize, m_prevChosenMinibatchSize * 2);
|
||||
}
|
||||
|
||||
|
|
@ -1634,7 +1634,7 @@ protected:
|
|||
// round mbsize to something meaningful
|
||||
trialMinibatchSize = RoundToMultipleOf64(trialMinibatchSizeFloat);
|
||||
|
||||
fprintf(stderr, "\nAdaptiveMinibatchSearch: Evaluating trial minibatchSize=%d out of range %d..%d ...\n\n",
|
||||
fprintf(stderr, "\nAdaptiveMinibatchSearch: Evaluating trial minibatchSize=%zd out of range %zd..%zd ...\n\n",
|
||||
trialMinibatchSize, RoundToMultipleOf64(minMinibatchSize), RoundToMultipleOf64(maxMinibatchSize));
|
||||
|
||||
size_t totalSamplesSeen;
|
||||
|
|
|
|||
|
|
@ -911,7 +911,7 @@ namespace Microsoft {
|
|||
|
||||
ComputeTimeInMBs += MBComputeTime;
|
||||
|
||||
fprintf(stderr, "Sentenes Seen = %d; Samples seen = %d; Total Compute Time = %.8g ; Time Per Sample=%.8g\n", numMBsRun, totalEpochSamples, ComputeTimeInMBs, ComputeTimeInMBs / totalEpochSamples);
|
||||
fprintf(stderr, "Sentences Seen = %zd; Samples seen = %zd; Total Compute Time = %.8g ; Time Per Sample=%.8g\n", numMBsRun, totalEpochSamples, ComputeTimeInMBs, ComputeTimeInMBs / totalEpochSamples);
|
||||
}
|
||||
|
||||
startReadMBTime = clock();
|
||||
|
|
@ -1229,7 +1229,7 @@ namespace Microsoft {
|
|||
{
|
||||
ElemType score = result_queue.top().score;
|
||||
best_score = score;
|
||||
fprintf(stderr, "best[%d] score = %.4e\t", ibest, score);
|
||||
fprintf(stderr, "best[%zd] score = %.4e\t", ibest, score);
|
||||
if (best_path.size() > 0)
|
||||
WriteNbest(ibest, best_path, outputNodes, dataWriter);
|
||||
}
|
||||
|
|
|
|||
520
Makefile
520
Makefile
|
|
@ -7,228 +7,402 @@
|
|||
#
|
||||
# This makefile will be extended/completed as we go.
|
||||
#
|
||||
# You will need to modify PATH and LD_LIBRARY_PATH environment variables to run CNTK
|
||||
# export LD_LIBRARY_PATH=<path_to_math_lib>/ifort64/lib:<path_to_cuda>/lib64:/usr/local/lib
|
||||
# export PATH=$PATH:/usr/local/bin:<path_to_cuda>/bin
|
||||
#
|
||||
# In order to deviate from the default settings in this Makefile, please specify options on
|
||||
# the make command line, like this, for example (to build release):
|
||||
#
|
||||
# make BUILDTYPE=release -j
|
||||
# To use this Makefile, create a directory to build in and make a Config.make in the directory
|
||||
# that provides
|
||||
# ACML_PATH= path to ACML library installation
|
||||
# only needed if MATHLIB=acml
|
||||
# MKL_PATH= path to MKL library installation
|
||||
# only needed if MATHLIB=mkl
|
||||
# GDK_PATH= path to cuda gdk installation, so $(GDK_PATH)/include/nvidia/gdk/nvml.h exists
|
||||
# defaults to /usr
|
||||
# BUILDTYPE= One of release or debug
|
||||
# defaults to release
|
||||
# MATHLIB= One of acml or mkl
|
||||
# defaults to acml
|
||||
# CUDA_PATH= Path to CUDA
|
||||
# If not specified, GPU will not be enabled
|
||||
# KALDI_PATH= Path to Kaldi
|
||||
# If not specified, Kaldi plugins will not be built
|
||||
|
||||
CC = g++
|
||||
NVCC = nvcc
|
||||
ARCH = x86_64
|
||||
|
||||
# DEVICE can also be cpu
|
||||
DEVICE = gpu
|
||||
|
||||
# BUILDTYPE can also be release
|
||||
BUILDTYPE = debug
|
||||
|
||||
# MATHLIB can also be mkl
|
||||
MATHLIB = acml
|
||||
|
||||
# This is a suggested/default location for ACML library
|
||||
MATHLIB_PATH = /usr/local/acml5.3.1/ifort64
|
||||
|
||||
# This is a suggested/default location for CUDA
|
||||
CUDA_PATH = /usr/local/cuda-7.0
|
||||
|
||||
# This is a suggested/default location for NVML
|
||||
NVML_INCLUDE = /usr/include/nvidia/gdk
|
||||
NVML_LIB = /usr/src/gdk/nvml/lib
|
||||
#######
|
||||
|
||||
BUILDFOR = $(ARCH).$(DEVICE).$(BUILDTYPE).$(MATHLIB)
|
||||
|
||||
OBJDIR = .build/$(BUILDFOR)
|
||||
BINDIR = bin/$(BUILDFOR)
|
||||
|
||||
# Set up debug vs release compiler settings, both nvcc and gcc
|
||||
ifeq ($(BUILDTYPE),debug)
|
||||
BUILDTYPE_OPT = -g
|
||||
GPU_BUILDTYPE_OPT = -O0 -G -lineinfo
|
||||
else
|
||||
BUILDTYPE_OPT = -O3 -flto
|
||||
GPU_BUILDTYPE_OPT = -O3 -use_fast_math -lineinfo
|
||||
ifndef BUILD_TOP
|
||||
BUILD_TOP=.
|
||||
endif
|
||||
|
||||
# Set up math library defines and libraries
|
||||
ifeq ($(MATHLIB),mkl)
|
||||
MATHLIB_INCLUDE = $(MATHLIB_PATH)/mkl/include
|
||||
MATHLIB_LIB = -L$(MATHLIB_PATH)/compiler/lib/intel64 -L$(MATHLIB_PATH)/mkl/lib/intel64 -L$(MATHLIB_PATH)/compiler/lib/mic -L$(MATHLIB_PATH)/mkl/lib/mic -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lm -liomp5 -lpthread
|
||||
MATHLIB_DEFINE = -DUSE_MKL
|
||||
ifneq ("$(wildcard $(BUILD_TOP)/Config.make)","")
|
||||
include $(BUILD_TOP)/Config.make
|
||||
else
|
||||
MATHLIB_INCLUDE = $(MATHLIB_PATH)/include
|
||||
MATHLIB_LIB = -L$(MATHLIB_PATH)/lib -lacml -lm -lpthread
|
||||
MATHLIB_DEFINE = -DUSE_ACML
|
||||
$(error Cannot fine $(BUILD_TOP)/Config.make. Please see the README file for configuration instructions.)
|
||||
endif
|
||||
|
||||
# Set up CUDA includes and libraries
|
||||
CUDA_INCLUDE = $(CUDA_PATH)/include
|
||||
CUDA_LIB = -L$(CUDA_PATH)/lib64 -L$(NVML_LIB) -lcublas -lcudart -lcurand -lcusparse -lnvidia-ml
|
||||
|
||||
# Set up final list of libs to use
|
||||
ifeq ($(DEVICE),gpu)
|
||||
LINK_LIBS = $(CUDA_LIB) $(MATHLIB_LIB)
|
||||
else
|
||||
LINK_LIBS = $(MATHLIB_LIB)
|
||||
ifndef BUILDTYPE
|
||||
$(info Defaulting BUILDTYPE=release)
|
||||
BUILDTYPE=release
|
||||
endif
|
||||
|
||||
# Compile CNTK math into its own shared library to ensure that any change to its
|
||||
# global variables, like CUDA streams is made in one place and has global effect.
|
||||
# Otherwise, different clients of CNTK math would observe different states.
|
||||
CNTKMATH_LINK_LIB = -L$(BINDIR) -lcntkmath
|
||||
CNTKMATH_LIB = $(BINDIR)/libcntkmath.so
|
||||
|
||||
# Set up gcc includes and libraries
|
||||
INCFLAGS_COMMON = -I Common/Include -I Math/Math -I MachineLearning/CNTK -I $(MATHLIB_INCLUDE)
|
||||
CFLAGS_COMMON = -msse3 -std=c++0x -std=c++11 -D_POSIX_SOURCE -D_XOPEN_SOURCE=600 -D__USE_XOPEN2K $(MATHLIB_DEFINE) -fopenmp -fpermissive -fPIC
|
||||
|
||||
ifeq ($(DEVICE),gpu)
|
||||
INCFLAGS = $(INCFLAGS_COMMON) -I $(CUDA_INCLUDE) -I $(NVML_INCLUDE)
|
||||
CFLAGS = $(CFLAGS_COMMON)
|
||||
else
|
||||
INCFLAGS = $(INCFLAGS_COMMON)
|
||||
CFLAGS = $(CFLAGS_COMMON) -DCPUONLY
|
||||
ifndef MATHLIB
|
||||
$(info DEFAULTING MATHLIB=acml)
|
||||
MATHLIB=acml
|
||||
endif
|
||||
|
||||
#### Configure based on options above
|
||||
|
||||
CXX = g++
|
||||
|
||||
INCLUDEPATH:= Common/Include Math/Math MachineLearning/CNTK
|
||||
CPPFLAGS:= -D_POSIX_SOURCE -D_XOPEN_SOURCE=600 -D__USE_XOPEN2K
|
||||
CXXFLAGS:= -msse3 -std=c++0x -std=c++11 -fopenmp -fpermissive -fPIC
|
||||
LIBPATH:=
|
||||
LIBS:=
|
||||
LDFLAGS:=
|
||||
|
||||
SEPARATOR = "=-----------------------------------------------------------="
|
||||
ALL:=
|
||||
SRC:=
|
||||
|
||||
# Make sure all is the first (i.e. default) target, but we can't actually define it
|
||||
# this early in the file, so let buildall do the work.
|
||||
all : buildall
|
||||
|
||||
# Set up nvcc target architectures (will generate code to support them all, i.e. fat-binary)
|
||||
GENCODE_SM20 := -gencode arch=compute_20,code=\"sm_20,compute_20\"
|
||||
GENCODE_SM30 := -gencode arch=compute_30,code=\"sm_30,compute_30\"
|
||||
GENCODE_SM35 := -gencode arch=compute_35,code=\"sm_35,compute_35\"
|
||||
GENCODE_FLAGS := $(GENCODE_SM20) $(GENCODE_SM30) $(GENCODE_SM35)
|
||||
|
||||
# Set up basic nvcc options and add GPU targets from above
|
||||
NVCCFLAGS = -std=c++11 -D_POSIX_SOURCE -D_XOPEN_SOURCE=600 -D__USE_XOPEN2K -m 64 $(GENCODE_FLAGS)
|
||||
# Set up basic nvcc options and add CUDA targets from above
|
||||
CUFLAGS = -std=c++11 -D_POSIX_SOURCE -D_XOPEN_SOURCE=600 -D__USE_XOPEN2K -m 64 $(GENCODE_FLAGS)
|
||||
|
||||
# Set up linker option to embed ORIGIN, i.e. directory where cntk is into the search path option
|
||||
# at runtime. This will try to resolve all dependent binaries in the same directory where cntk binary resides
|
||||
LDFLAGS=-Wl,-rpath,'$$ORIGIN'
|
||||
ifdef CUDA_PATH
|
||||
ifndef GDK_PATH
|
||||
$(info defaulting GDK_PATH to /usr)
|
||||
GDK_PATH=/usr
|
||||
endif
|
||||
|
||||
# Define all sources that need to be built
|
||||
COMMON_SRC = Common/fileutil.cpp Common/DataWriter.cpp Common/ConfigFile.cpp Common/DataReader.cpp \
|
||||
Common/Eval.cpp Common/File.cpp Common/BestGpu.cpp Common/TimerUtility.cpp
|
||||
DEVICE = gpu
|
||||
|
||||
MATH_COMMON_SRC = Math/Math/Matrix.cpp Math/Math/CPUMatrix.cpp Math/Math/CPUSparseMatrix.cpp
|
||||
NVCC = $(CUDA_PATH)/bin/nvcc
|
||||
|
||||
# This is a suggested/default location for NVML
|
||||
INCLUDEPATH+=$(GDK_PATH)/include/nvidia/gdk
|
||||
NVMLPATH=$(GDK_PATH)/src/gdk/nvml/lib
|
||||
|
||||
# Set up CUDA includes and libraries
|
||||
INCLUDEPATH += $(CUDA_PATH)/include
|
||||
LIBPATH += $(CUDA_PATH)/lib64
|
||||
LIBS += -lcublas -lcudart -lcuda -lcurand -lcusparse -lnvidia-ml
|
||||
|
||||
ifeq ($(DEVICE),gpu)
|
||||
MATH_SRC = $(MATH_COMMON_SRC) Math/Math/GPUMatrix.cu Math/Math/GPUMatrixCUDAKernels.cu Math/Math/GPUSparseMatrix.cu Math/Math/GPUWatcher.cu
|
||||
else
|
||||
MATH_SRC = $(MATH_COMMON_SRC) Math/Math/NoGPU.cpp
|
||||
DEVICE = cpu
|
||||
|
||||
CPPFLAGS +=-DCPUONLY
|
||||
endif
|
||||
|
||||
CN_SRC = MachineLearning/CNTK/NetworkDescriptionLanguage.cpp MachineLearning/CNTK/CNTK.cpp MachineLearning/CNTK/ComputationNode.cpp \
|
||||
MachineLearning/CNTK/ModelEditLanguage.cpp MachineLearning/CNTK/SimpleNetworkBuilder.cpp MachineLearning/CNTK/tests.cpp \
|
||||
MachineLearning/CNTK/Profiler.cpp MachineLearning/CNTKEval/CNTKEval.cpp
|
||||
ifeq ("$(MATHLIB)","acml")
|
||||
INCLUDEPATH += $(ACML_PATH)/include
|
||||
LIBPATH += $(ACML_PATH)/lib
|
||||
LIBS += -lacml -lm -lpthread
|
||||
CPPFLAGS += -DUSE_ACML
|
||||
endif
|
||||
|
||||
BINARYREADER_SRC = DataReader/BinaryReader/BinaryWriter.cpp DataReader/BinaryReader/BinaryReader.cpp DataReader/BinaryReader/BinaryFile.cpp
|
||||
HTKMLFREADER_SRC = DataReader/HTKMLFReader_linux/HTKMLFWriter.cpp DataReader/HTKMLFReader_linux/DataWriter.cpp DataReader/HTKMLFReader_linux/DataReader.cpp DataReader/HTKMLFReader_linux/HTKMLFReader.cpp
|
||||
SEQUENCEREADER_SRC = DataReader/LMSequenceReader/SequenceReader.cpp DataReader/LMSequenceReader/SequenceParser.cpp DataReader/LMSequenceReader/Exports.cpp
|
||||
LUSEQUENCEREADER_SRC = DataReader/LUSequenceReader/LUSequenceReader.cpp DataReader/LUSequenceReader/LUSequenceParser.cpp DataReader/LUSequenceReader/Exports.cpp
|
||||
UCIFASTREADER_SRC = DataReader/UCIFastReader/UCIParser.cpp DataReader/UCIFastReader/UCIFastReader.cpp DataReader/UCIFastReader/Exports.cpp
|
||||
ifeq ("$(MATHLIB)","mkl")
|
||||
INCLUDEPATH += $(MKL_PATH)/mkl/include
|
||||
LIBPATH += $(MKL_PATH)/compiler/lib/intel64 $(MKL_PATH)/mkl/lib/intel64 $(MKL_PATH)/compiler/lib/mic $(MKL_PATH)/mkl/lib/mic
|
||||
LIBS += -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lm -liomp5 -lpthread
|
||||
CPPFLAGS += -DUSE_MKL
|
||||
endif
|
||||
|
||||
READER_SRC = $(UCIFASTREADER_SRC) $(LUSEQUENCEREADER_SRC) $(HTKMLFREADER_SRC) $(SEQUENCEREADER_SRC) $(BINARYREADER_SRC)
|
||||
CORE_SRC = $(CN_SRC) $(COMMON_SRC)
|
||||
SRC = $(READER_SRC) $(CORE_SRC) $(MATH_SRC)
|
||||
|
||||
ifdef KALDI_PATH
|
||||
########## Copy includes and defines from $(KALDI_PATH)/src/kaldi.mk ##########
|
||||
FSTROOT = $(KALDI_PATH)/tools/openfst
|
||||
ATLASINC = $(KALDI_PATH)/tools/ATLAS/include
|
||||
|
||||
INCLUDEPATH += $(KALDI_PATH)/src $(ATLASINC) $(FSTROOT)/include
|
||||
CPPFLAGS+= -DKALDI_DOUBLEPRECISION=0 -DHAVE_POSIX_MEMALIGN -DHAVE_EXECINFO_H=1 -DHAVE_CXXABI_H -DHAVE_ATLAS -DHAVE_OPENFST_GE_10400
|
||||
|
||||
KALDI_LIBPATH += $(KALDI_PATH)/src/lib
|
||||
KALDI_LIBS += -lkaldi-util -lkaldi-matrix -lkaldi-base -lkaldi-hmm -lkaldi-cudamatrix -lkaldi-nnet -lkaldi-lat
|
||||
endif
|
||||
|
||||
ifeq ("$(BUILDTYPE)","debug")
|
||||
CXXFLAGS += -g
|
||||
CUFLAGS += -O0 -G -lineinfo
|
||||
endif
|
||||
|
||||
ifeq ("$(BUILDTYPE)","release")
|
||||
CXXFLAGS += -O4
|
||||
CUFLAGS += -O3 -use_fast_math -lineinfo
|
||||
endif
|
||||
|
||||
#######
|
||||
|
||||
OBJDIR:= $(BUILD_TOP)/.build
|
||||
BINDIR:= $(BUILD_TOP)/bin
|
||||
LIBDIR:= $(BUILD_TOP)/lib
|
||||
|
||||
ORIGINLIBDIR:='$$ORIGIN/../lib'
|
||||
ORIGINDIR:='$$ORIGIN'
|
||||
|
||||
CNTKMATH:=cntkmath
|
||||
|
||||
########################################
|
||||
# Math library
|
||||
########################################
|
||||
|
||||
# Define all sources that need to be built
|
||||
COMMON_SRC =\
|
||||
Common/BestGpu.cpp \
|
||||
Common/ConfigFile.cpp \
|
||||
Common/DataReader.cpp \
|
||||
Common/DataWriter.cpp \
|
||||
Common/Eval.cpp \
|
||||
Common/File.cpp \
|
||||
Common/TimerUtility.cpp \
|
||||
Common/fileutil.cpp \
|
||||
|
||||
MATH_SRC =\
|
||||
Math/Math/CPUMatrix.cpp \
|
||||
Math/Math/CPUSparseMatrix.cpp \
|
||||
Math/Math/Matrix.cpp \
|
||||
|
||||
ifdef CUDA_PATH
|
||||
MATH_SRC +=\
|
||||
Math/Math/GPUMatrix.cu \
|
||||
Math/Math/GPUMatrixCUDAKernels.cu \
|
||||
Math/Math/GPUSparseMatrix.cu \
|
||||
Math/Math/GPUWatcher.cu \
|
||||
|
||||
else
|
||||
MATH_SRC +=\
|
||||
Math/Math/NoGPU.cpp
|
||||
|
||||
endif
|
||||
|
||||
MATH_SRC+=$(COMMON_SRC)
|
||||
|
||||
MATH_OBJ := $(patsubst %.cu, $(OBJDIR)/%.o, $(patsubst %.cpp, $(OBJDIR)/%.o, $(MATH_SRC)))
|
||||
|
||||
CNTKMATH_LIB:= $(LIBDIR)/lib$(CNTKMATH).so
|
||||
ALL += $(CNTKMATH_LIB)
|
||||
SRC+=$(MATH_SRC)
|
||||
|
||||
RPATH=-Wl,-rpath,
|
||||
|
||||
$(CNTKMATH_LIB): $(MATH_OBJ)
|
||||
@echo $(SEPARATOR)
|
||||
@echo creating $@ for $(ARCH) with build type $(BUILDTYPE)
|
||||
@mkdir -p $(dir $@)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBPATH) $(NVMLPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ $(LIBS) -fopenmp
|
||||
|
||||
########################################
|
||||
# BinaryReader plugin
|
||||
########################################
|
||||
|
||||
|
||||
BINARYREADER_SRC =\
|
||||
DataReader/BinaryReader/BinaryFile.cpp \
|
||||
DataReader/BinaryReader/BinaryReader.cpp \
|
||||
DataReader/BinaryReader/BinaryWriter.cpp \
|
||||
|
||||
BINARYREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(BINARYREADER_SRC))
|
||||
|
||||
BINARY_READER:= $(LIBDIR)/BinaryReader.so
|
||||
|
||||
#ALL += $(BINARY_READER)
|
||||
#SRC+=$(BINARYREADER_SRC)
|
||||
|
||||
$(BINARY_READER): $(BINARYREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
########################################
|
||||
# HTKMLFReader plugin
|
||||
########################################
|
||||
|
||||
|
||||
HTKMLFREADER_SRC =\
|
||||
DataReader/HTKMLFReader_linux/DataReader.cpp \
|
||||
DataReader/HTKMLFReader_linux/DataWriter.cpp \
|
||||
DataReader/HTKMLFReader_linux/HTKMLFReader.cpp \
|
||||
DataReader/HTKMLFReader_linux/HTKMLFWriter.cpp \
|
||||
|
||||
HTKMLREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(HTKMLFREADER_SRC))
|
||||
|
||||
HTKMLREADER:=$(LIBDIR)/HTKMLFReader.so
|
||||
ALL+=$(HTKMLREADER)
|
||||
SRC+=$(HTKMLREADER_SRC)
|
||||
|
||||
$(LIBDIR)/HTKMLFReader.so: $(HTKMLREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
########################################
|
||||
# LMSequenceReader plugin
|
||||
########################################
|
||||
|
||||
LMSEQUENCEREADER_SRC =\
|
||||
DataReader/LMSequenceReader/Exports.cpp \
|
||||
DataReader/LMSequenceReader/SequenceParser.cpp \
|
||||
DataReader/LMSequenceReader/SequenceReader.cpp \
|
||||
|
||||
LMSEQUENCEREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(LMSEQUENCEREADER_SRC))
|
||||
|
||||
LMSEQUENCEREADER:= $(LIBDIR)/LMSequenceReader.so
|
||||
ALL+=$(LMSEQUENCEREADER)
|
||||
SRC+=$(LMSEQUENCEREADER_SRC)
|
||||
|
||||
$(LMSEQUENCEREADER): $(LMSEQUENCEREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
########################################
|
||||
# LUSequenceReader plugin
|
||||
########################################
|
||||
|
||||
LUSEQUENCEREADER_SRC =\
|
||||
DataReader/LUSequenceReader/Exports.cpp \
|
||||
DataReader/LUSequenceReader/LUSequenceParser.cpp \
|
||||
DataReader/LUSequenceReader/LUSequenceReader.cpp \
|
||||
|
||||
LUSEQUENCEREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(LUSEQUENCEREADER_SRC))
|
||||
|
||||
LUSEQUENCEREADER:=$(LIBDIR)/LUSequenceReader.so
|
||||
ALL+=$(LUSEQUENCEREADER)
|
||||
SRC+=$(LUSEQUENCEREADER_SRC)
|
||||
|
||||
$(LUSEQUENCEREADER): $(LUSEQUENCEREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
########################################
|
||||
# UCIFastReader plugin
|
||||
########################################
|
||||
|
||||
UCIFASTREADER_SRC =\
|
||||
DataReader/UCIFastReader/Exports.cpp \
|
||||
DataReader/UCIFastReader/UCIFastReader.cpp \
|
||||
DataReader/UCIFastReader/UCIParser.cpp \
|
||||
|
||||
UCIFASTREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(UCIFASTREADER_SRC))
|
||||
|
||||
UCIFASTREADER:=$(LIBDIR)/UCIFastReader.so
|
||||
ALL += $(UCIFASTREADER)
|
||||
SRC+=$(UCIFASTREADER_SRC)
|
||||
|
||||
$(UCIFASTREADER): $(UCIFASTREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
########################################
|
||||
# Kaldi plugins
|
||||
########################################
|
||||
|
||||
ifdef KALDI_PATH
|
||||
KALDIREADER_SRC = \
|
||||
DataReader/KaldiReader/DataReader.cpp \
|
||||
DataReader/KaldiReader/DataWriter.cpp \
|
||||
DataReader/KaldiReader/HTKMLFReader.cpp \
|
||||
DataReader/KaldiReader/HTKMLFWriter.cpp \
|
||||
|
||||
KALDIREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(KALDIREADER_SRC))
|
||||
|
||||
KALDIREADER:=$(LIBDIR)/KaldiReader.so
|
||||
ALL+=$(KALDIREADER)
|
||||
SRC+=$(KALDIREADER_SRC)
|
||||
|
||||
$(KALDIREADER): $(KALDIREADER_OBJ)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(KALDI_LIBPATH) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(KALDI_LIBPATH) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH) $(KALDI_LIBS)
|
||||
|
||||
KALDIWRITER:=$(LIBDIR)/KaldiWriter.so
|
||||
ALL+=$(KALDIWRITER)
|
||||
|
||||
$(KALDIWRITER): $(KALDIREADER_OBJ)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH)
|
||||
|
||||
|
||||
KALDI2READER_SRC = \
|
||||
DataReader/Kaldi2Reader/DataReader.cpp \
|
||||
DataReader/Kaldi2Reader/DataWriter.cpp \
|
||||
DataReader/Kaldi2Reader/HTKMLFReader.cpp \
|
||||
DataReader/Kaldi2Reader/HTKMLFWriter.cpp \
|
||||
DataReader/Kaldi2Reader/KaldiSequenceTrainingDerivative.cpp \
|
||||
DataReader/Kaldi2Reader/UtteranceDerivativeBuffer.cpp \
|
||||
|
||||
KALDI2READER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(KALDI2READER_SRC))
|
||||
|
||||
KALDI2READER:=$(LIBDIR)/Kaldi2Reader.so
|
||||
ALL+=$(KALDI2READER)
|
||||
SRC+=$(KALDI2READER_SRC)
|
||||
|
||||
$(KALDI2READER): $(KALDI2READER_OBJ)
|
||||
@echo $(SEPARATOR)
|
||||
$(CXX) $(LDFLAGS) -shared $(patsubst %,-L%, $(LIBDIR) $(KALDI_LIBPATH) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINDIR) $(KALDI_LIBPATH) $(LIBPATH)) -o $@ $^ -l$(CNTKMATH) $(KALDI_LIBS)
|
||||
|
||||
endif
|
||||
|
||||
########################################
|
||||
# cntk
|
||||
########################################
|
||||
|
||||
CNTK_SRC =\
|
||||
MachineLearning/CNTK/CNTK.cpp \
|
||||
MachineLearning/CNTK/ComputationNode.cpp \
|
||||
MachineLearning/CNTK/ModelEditLanguage.cpp \
|
||||
MachineLearning/CNTK/NetworkDescriptionLanguage.cpp \
|
||||
MachineLearning/CNTK/Profiler.cpp \
|
||||
MachineLearning/CNTK/SimpleNetworkBuilder.cpp \
|
||||
MachineLearning/CNTK/tests.cpp \
|
||||
MachineLearning/CNTKEval/CNTKEval.cpp \
|
||||
|
||||
CNTK_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(CNTK_SRC))
|
||||
|
||||
CNTK:=$(BINDIR)/cntk
|
||||
ALL+=$(CNTK)
|
||||
|
||||
$(CNTK): $(CNTK_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
@mkdir -p $(dir $@)
|
||||
@echo building output for $(ARCH) with build type $(BUILDTYPE)
|
||||
$(CXX) $(LDFLAGS) $(patsubst %,-L%, $(LIBDIR) $(LIBPATH)) $(patsubst %,$(RPATH)%, $(ORIGINLIBDIR) $(LIBPATH)) -o $@ $^ $(LIBS) -l$(CNTKMATH) -fopenmp
|
||||
|
||||
########################################
|
||||
# General compile and dependency rules
|
||||
########################################
|
||||
|
||||
VPATH := $(sort $(dir $(SRC)))
|
||||
|
||||
# Define object files
|
||||
OBJ_TMP := $(patsubst %.cpp, $(OBJDIR)/%.o, $(SRC))
|
||||
ifeq ($(DEVICE),gpu)
|
||||
OBJ := $(patsubst %.cu, $(OBJDIR)/%.o, $(OBJ_TMP))
|
||||
else
|
||||
OBJ := $(OBJ_TMP)
|
||||
endif
|
||||
|
||||
CORE_OBJ_TMP := $(patsubst %.cpp, $(OBJDIR)/%.o, $(CORE_SRC))
|
||||
ifeq ($(DEVICE),gpu)
|
||||
CORE_OBJ := $(patsubst %.cu, $(OBJDIR)/%.o, $(CORE_OBJ_TMP))
|
||||
else
|
||||
CORE_OBJ := $(CORE_OBJ_TMP)
|
||||
endif
|
||||
|
||||
COMMON_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(COMMON_SRC))
|
||||
|
||||
MATH_OBJ_TMP := $(patsubst %.cpp, $(OBJDIR)/%.o, $(MATH_SRC))
|
||||
ifeq ($(DEVICE),gpu)
|
||||
MATH_OBJ := $(patsubst %.cu, $(OBJDIR)/%.o, $(MATH_OBJ_TMP))
|
||||
else
|
||||
MATH_OBJ := $(MATH_OBJ_TMP)
|
||||
endif
|
||||
|
||||
UCIFASTREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(UCIFASTREADER_SRC))
|
||||
LUSEQUENCEREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(LUSEQUENCEREADER_SRC))
|
||||
SEQUENCEREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(SEQUENCEREADER_SRC))
|
||||
HTKMLFREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(HTKMLFREADER_SRC))
|
||||
BINARYREADER_OBJ := $(patsubst %.cpp, $(OBJDIR)/%.o, $(BINARYREADER_SRC))
|
||||
OBJ := $(patsubst %.cu, $(OBJDIR)/%.o, $(patsubst %.cpp, $(OBJDIR)/%.o, $(SRC)))
|
||||
|
||||
# C++ include dependencies generated by -MF compiler option
|
||||
DEP := $(patsubst %.o, %.d, $(OBJ))
|
||||
|
||||
SEPARATOR = "=-----------------------------------------------------------="
|
||||
|
||||
# Define build targets
|
||||
all: $(BINDIR)/cntk $(BINDIR)/UCIFastReader.so $(BINDIR)/LMSequenceReader.so $(BINDIR)/LUSequenceReader.so $(BINDIR)/HTKMLFReader.so
|
||||
@echo $(SEPARATOR)
|
||||
@echo finished building for $(ARCH) with build type $(BUILDTYPE)
|
||||
|
||||
$(BINDIR)/UCIFastReader.so: $(UCIFASTREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(CNTKMATH_LINK_LIB)
|
||||
|
||||
$(BINDIR)/LMSequenceReader.so: $(SEQUENCEREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(CNTKMATH_LINK_LIB)
|
||||
|
||||
$(BINDIR)/LUSequenceReader.so: $(LUSEQUENCEREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(CNTKMATH_LINK_LIB)
|
||||
|
||||
$(BINDIR)/HTKMLFReader.so: $(HTKMLFREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(CNTKMATH_LINK_LIB)
|
||||
|
||||
$(BINDIR)/BinaryReader.so: $(BINARYREADER_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(CNTKMATH_LINK_LIB)
|
||||
|
||||
$(BINDIR)/cntk: $(CORE_OBJ) | $(CNTKMATH_LIB)
|
||||
@echo $(SEPARATOR)
|
||||
@mkdir -p $(dir $@)
|
||||
@echo building output for $(ARCH) with build type $(BUILDTYPE)
|
||||
$(CC) $(BUILDTYPE_OPT) $(LDFLAGS) -o $@ $^ $(LINK_LIBS) $(CNTKMATH_LINK_LIB) -fopenmp -ldl -fPIC
|
||||
|
||||
$(CNTKMATH_LIB): $(MATH_OBJ) $(COMMON_OBJ)
|
||||
@echo $(SEPARATOR)
|
||||
@echo creating $@ for $(ARCH) with build type $(BUILDTYPE)
|
||||
@mkdir -p $(dir $@)
|
||||
$(CC) $(BUILDTYPE_OPT) -fPIC -shared -o $@ $^ $(LINK_LIBS) -fopenmp
|
||||
|
||||
# Include all C++ dependencies, like header files, to ensure that a change in those
|
||||
# will result in the rebuild.
|
||||
-include ${DEP}
|
||||
|
||||
ifeq ($(DEVICE),gpu)
|
||||
$(OBJDIR)/%.o : %.cu Makefile
|
||||
@echo $(SEPARATOR)
|
||||
@echo creating $@ for $(ARCH) with build type $(BUILDTYPE)
|
||||
@mkdir -p $(dir $@)
|
||||
$(NVCC) -c $< -o $@ $(GPU_BUILDTYPE_OPT) $(NVCCFLAGS) $(INCFLAGS) -Xcompiler -fPIC
|
||||
endif
|
||||
$(NVCC) -c $< -o $@ $(CUFLAGS) $(INCLUDEPATH:%=-I%) -Xcompiler -fPIC
|
||||
|
||||
$(OBJDIR)/%.o : %.cpp Makefile
|
||||
@echo $(SEPARATOR)
|
||||
@echo creating $@ for $(ARCH) with build type $(BUILDTYPE)
|
||||
@mkdir -p $(dir $@)
|
||||
$(CC) -c $< -o $@ $(BUILDTYPE_OPT) $(CPPFLAGS) $(CFLAGS) $(INCFLAGS) -MD -MP -MF ${@:.o=.d}
|
||||
$(CXX) -c $< -o $@ $(CPPFLAGS) $(CXXFLAGS) $(INCLUDEPATH:%=-I%) -MD -MP -MF ${@:.o=.d}
|
||||
|
||||
.PHONY: clean
|
||||
.PHONY: clean buildall all
|
||||
|
||||
clean:
|
||||
@echo $(SEPARATOR)
|
||||
@rm -rf $(OBJDIR)
|
||||
@rm -rf $(BINDIR)
|
||||
@rm -rf $(ALL)
|
||||
@echo finished cleaning up the project
|
||||
|
||||
buildall : $(ALL)
|
||||
@echo $(SEPARATOR)
|
||||
@echo finished building for $(ARCH) with build type $(BUILDTYPE)
|
||||
|
|
|
|||
65
README
65
README
|
|
@ -33,55 +33,62 @@ To build the cpu version, you have to install intel MKL blas library or ACML lib
|
|||
|
||||
for MKL:
|
||||
1. Download from https://software.intel.com/en-us/intel-mkl
|
||||
2. You can modify variable MKL_PATH in makefile.cpu to change your mkl path.
|
||||
Then add ${MKL_PATH}/mkl/lib/intel64, ${MKL_PATH}/mkl/lib/mic, ${MKL_PATH}/compiler/lib/intel64. ${MKL_PATH}/compiler/lib/mic to your ${LD_LIBRARY_PATH} to make sure the program links the library correctly.
|
||||
|
||||
for ACML:
|
||||
1. Download from http://developer.amd.com/tools-and-sdks/cpu-development/amd-core-math-library-acml/
|
||||
2. Modify ACML_PATH in the makefile.cpu and makefile.gpu to provide your ACML library path.
|
||||
You need to add ${ACML_PATH}/lib to your ${LD_LIBRARY_PATH}.
|
||||
|
||||
for Kaldi:
|
||||
1. In kaldi-trunk/tools/Makefile, uncomment # OPENFST_VERSION = 1.4.1, and
|
||||
re-install OpenFst using the makefile.
|
||||
2. In kaldi-trunk/src/, do ./configure --shared; make depend -j 8; make -j 8;
|
||||
and re-compile Kaldi (the -j option is for parallelization).
|
||||
3. Set KALDI_PATH in kaldi_vars.mk
|
||||
4. When running the binaries, make sure you add $KALDI_PATH/tools/openfst/lib
|
||||
and $KALDI_PATH/src/lib/ to your $LD_LIBRARY_PATH
|
||||
|
||||
To build the gpu version, you have to install NIVIDIA CUDA first
|
||||
You can modify the path CUDA_PATH in makefile.cpu to change your cuda path
|
||||
We use cuda-7.0 as default.
|
||||
Then add ${CUDA_PATH}/lib, ${CUDA_PATH}/lib64 to your ${LD_LIBRARY_PATH} to make sure the program links to the library correctly.
|
||||
|
||||
... TODO: add documentation on nvml lib
|
||||
== Build Preparation ==
|
||||
Let $CNTK be the CNTK directory.
|
||||
>mkdir build
|
||||
>$CNTK/configure -h
|
||||
|
||||
== Build ==
|
||||
To build the cpu version, run
|
||||
make DEVICE=cpu
|
||||
To build the gpu version, run
|
||||
make
|
||||
To clean the compile, just run
|
||||
make DEVICE=cpu clean
|
||||
or
|
||||
make clean
|
||||
You will see various options for configure, as well as their default
|
||||
values. CNTK needs a CPU math directory, either acml or mkl. If you
|
||||
do not specify one and both are available, acml will be used. For GPU
|
||||
use, a cuda and gdk directory are also required. Similary, to build
|
||||
the kaldi plugin a kaldi directory is required. You may also specify
|
||||
whether you want a debug or release build. Rerun configure with the
|
||||
desired options.
|
||||
|
||||
For release version, just add BUILDTYPE=release to the make command line.
|
||||
>$CNTK/configure ...
|
||||
|
||||
This will create a Config.make and a Makefile (if you are in the $CNTK
|
||||
directory, a Makefile will not be created). The Config.make file
|
||||
records the configuration parameters and the Makefile reinvokes the
|
||||
$CNTK/Makefile, passing it the build directory where it can find the
|
||||
Config.make.
|
||||
|
||||
After make completes, you will have the following directories:
|
||||
|
||||
.build will contain object files, and can be deleted
|
||||
bin contains the cntk program
|
||||
lib contains libraries and plugins
|
||||
|
||||
The bin and lib directories can safely be moved as long as they remain siblings.
|
||||
|
||||
To clean
|
||||
|
||||
>make clean
|
||||
|
||||
== Run ==
|
||||
All executables are in bin/ directory:
|
||||
cn.exe: The main executable for CNTK
|
||||
All executables are in bin directory:
|
||||
cntk: The main executable for CNTK
|
||||
*.so: shared library for corresponding reader, these readers will be linked and loaded dynamically at runtime.
|
||||
|
||||
To run the executable, make sure bin/ is in your ${LD_LIBRARY_PATH}, if not, running cn.exe will fail when cn.exe tries to link the corresponding reader. Once it's done, run in command line:
|
||||
./cn.exe configFile=${your config file}
|
||||
./cntk configFile=${your cntk config file}
|
||||
|
||||
== Kaldi Reader ==
|
||||
This is a HTKMLF reader and kaldi writer (for decode)
|
||||
|
||||
To build the cpu/gpu version, run
|
||||
make -f Makefile_kaldi.cpu/gpu
|
||||
To build, set KALDI_PATH in your Config.make
|
||||
|
||||
The feature section is like:
|
||||
|
||||
|
|
@ -102,9 +109,7 @@ writer=[
|
|||
== Kaldi2 Reader ==
|
||||
This is a kaldi reader and kaldi writer (for decode)
|
||||
|
||||
To build the cpu/gpu version, run
|
||||
make -f Makefile_kaldi2.cpu/gpu
|
||||
|
||||
To build, set KALDI_PATH in your Config.make
|
||||
|
||||
The features section is different:
|
||||
|
||||
|
|
|
|||
|
|
@ -96,10 +96,11 @@ if [[ $OS == "Windows_NT" && $OSTYPE == "cygwin" ]]; then
|
|||
exit 1
|
||||
fi
|
||||
elif [[ $OSTYPE == "linux-gnu" ]]; then
|
||||
DEBUG_DIR=x86_64.gpu.debug.acml
|
||||
RELEASE_DIR=x86_64.gpu.release.acml
|
||||
PREFIX_DIR=bin
|
||||
BIN_NAME=cntk
|
||||
DEBUG_DIR=build/debug
|
||||
RELEASE_DIR=build/release
|
||||
PREFIX_DIR=
|
||||
# Make sure no dependencies on current directory
|
||||
BIN_NAME=bin/cntk
|
||||
MAKEFILE=Makefile
|
||||
BUILD_OS="linux"
|
||||
else
|
||||
|
|
@ -179,12 +180,16 @@ if [[ $BUILD == 1 ]]; then
|
|||
fi
|
||||
msbuild.exe /property:Configuration=$FLAVOR /m 1>&6 2>&7 || exit $?
|
||||
else
|
||||
BUILD_DIR=build/$FLAVOR
|
||||
./configure --with-build-top=$BUILD_DIR --with-acml=$ACML_PATH --with-buildtype=$FLAVOR
|
||||
if [[ $CLEAN_BEFORE == 1 ]]; then
|
||||
make BUILDTYPE=$FLAVOR -f $MAKEFILE clean 1>&6 2>&7 || exit $?
|
||||
make -C $BUILD_DIR -f $MAKEFILE clean 1>&6 2>&7 || exit $?
|
||||
fi
|
||||
make BUILDTYPE=$FLAVOR -j -f $MAKEFILE 1>&6 2>&7 || exit $?
|
||||
make -C $BUILD_DIR -j -f $MAKEFILE 1>&6 2>&7 || exit $?
|
||||
fi
|
||||
if [[ $QUIET_BUILD == 1 ]]; then
|
||||
chmod a+r $BUILD_FILE.*
|
||||
fi
|
||||
chmod a+r $BUILD_FILE.*
|
||||
done
|
||||
fi
|
||||
|
||||
|
|
@ -212,18 +217,21 @@ if [[ $RUN == 1 ]]; then
|
|||
fi
|
||||
OUT_FILE="$RUN_FILE.$FLAVOR.$TARGET.out"
|
||||
|
||||
if ! [[ -f "./$FLAVOR_DIR/$BIN_NAME" ]]; then
|
||||
BIN_PATH=$CNTK_ROOT/$FLAVOR_DIR/$BIN_NAME
|
||||
if ! [[ -f $BIN_PATH ]]; then
|
||||
echo "============ ERROR: CNTK did not build properly for flavor ($FLAVOR) ============"
|
||||
echo "Missing file: $BIN_PATH"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
echo "============ Running CNTK for ($FLAVOR) ($TARGET), output in ($RUN_FILE.*) ============"
|
||||
cd $CNTK_ROOT/Demos
|
||||
rm -rf models
|
||||
if [[ $OS == "Windows_NT" ]]; then
|
||||
# We have to use cygpath on Windows to modify the file paths into the format readable by cntk.
|
||||
time ./$FLAVOR_DIR/$BIN_NAME configFile="`cygpath -w $CONF_FILE`" &>$OUT_FILE || exit $?
|
||||
time $BIN_PATH configFile="`cygpath -w $CONF_FILE`" &>$OUT_FILE || exit $?
|
||||
else
|
||||
time ./$FLAVOR_DIR/$BIN_NAME configFile=$CONF_FILE &>$OUT_FILE || exit $?
|
||||
time $BIN_PATH configFile=$CONF_FILE &>$OUT_FILE || exit $?
|
||||
fi
|
||||
chmod a+r $RUN_FILE.*
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,738 @@
|
|||
=== Running /cygdrive/c/Users/svcphil/workspace.vlivan/CNTK-Build-Windows/x64/release/cntk.exe configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
-------------------------------------------------------------------
|
||||
Build info:
|
||||
|
||||
Built time: Aug 11 2015 16:18:17
|
||||
Last modified date: Tue Aug 11 16:16:08 2015
|
||||
Built by svcphil on dphaim-26-new
|
||||
Build Path: C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\MachineLearning\CNTK\
|
||||
CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.0
|
||||
Build Branch: master
|
||||
Build SHA1: 397cc7cc16c00b1c12864d331c0729fde7a1bde3
|
||||
-------------------------------------------------------------------
|
||||
running on dphaim-26-new at 2015/08/11 17:47:10
|
||||
command line options:
|
||||
configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG (VARIABLES NOT RESOLVED) >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=$DeviceId$
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=$RunDir$/models/cntkSpeech.dnn
|
||||
deviceId=$DeviceId$
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=$DataDir$/glob_0000.mlf
|
||||
labelMappingFile=$DataDir$/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG (VARIABLES NOT RESOLVED) <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=Auto
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> PROCESSED CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
configparameters: cntk.config:command=speechTrain
|
||||
configparameters: cntk.config:DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
configparameters: cntk.config:deviceId=Auto
|
||||
configparameters: cntk.config:parallelTrain=false
|
||||
configparameters: cntk.config:precision=float
|
||||
configparameters: cntk.config:RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
configparameters: cntk.config:speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< PROCESSED CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
command: speechTrain
|
||||
precision = float
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
LockDevice: Capture device 2 and lock it for exclusive use
|
||||
LockDevice: Capture device 3 and lock it for exclusive use
|
||||
LockDevice: Capture device 0 and lock it for exclusive use
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
SimpleNetworkBuilder Using GPU 1
|
||||
reading script file glob_0000.scp ... 948 entries
|
||||
trainlayer: OOV-exclusion code enabled, but no unigram specified to derive the word set from, so you won't get OOV exclusion
|
||||
total 132 state names in state list C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
htkmlfreader: reading MLF file C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf ... total 948 entries
|
||||
...............................................................................................feature set 0: 252734 frames in 948 out of 948 utterances
|
||||
label set 0: 129 classes
|
||||
minibatchutterancesource: 948 utterances grouped into 3 chunks, av. chunk size: 316.0 utterances, 84244.7 frames
|
||||
GetTrainCriterionNodes ...
|
||||
GetEvalCriterionNodes ...
|
||||
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 3])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 3])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 3], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 3])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 3], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 3])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 3])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 3], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 3])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 3])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 3], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 3], HLast[132, 3])
|
||||
|
||||
Found 3 PreCompute nodes
|
||||
NodeName: InvStdOfFeatures
|
||||
NodeName: MeanOfFeatures
|
||||
NodeName: Prior
|
||||
minibatchiterator: epoch 0: frames [0..252734] (first utterance at frame 0) with 1 datapasses
|
||||
requiredata: determined feature kind as 33-dimensional 'USER' with frame shift 10.0 ms
|
||||
|
||||
|
||||
Validating node InvStdOfFeatures
|
||||
|
||||
Validating --> features = InputValue
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 64])
|
||||
|
||||
|
||||
|
||||
Validating node MeanOfFeatures
|
||||
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 64])
|
||||
|
||||
|
||||
|
||||
Validating node Prior
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> Prior = Mean(labels[132, 64])
|
||||
|
||||
Set Max Temp Mem Size For Convolution Nodes to 0 samples.
|
||||
Starting Epoch 1: learning rate per sample = 0.015625 momentum = 0.900000
|
||||
minibatchiterator: epoch 0: frames [0..20480] (first utterance at frame 0) with 1 datapasses
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 64])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 64])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 64], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 64])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 64], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 64])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 64])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 64], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 64])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 64])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 64], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 64], HLast[132, 64])
|
||||
|
||||
Epoch[ 1 of 3]-Minibatch[ 1- 10 of 320]: SamplesSeen = 640; TrainLossPerSample = 4.45646143; EvalErr[0]PerSample = 0.92500001; TotalTime = 0.01913s; TotalTimePerSample = 0.02988ms; SamplesPerSecond = 33462
|
||||
Epoch[ 1 of 3]-Minibatch[ 11- 20 of 320]: SamplesSeen = 640; TrainLossPerSample = 4.22315693; EvalErr[0]PerSample = 0.90156251; TotalTime = 0.01453s; TotalTimePerSample = 0.02270ms; SamplesPerSecond = 44043
|
||||
Epoch[ 1 of 3]-Minibatch[ 21- 30 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.95180511; EvalErr[0]PerSample = 0.84687501; TotalTime = 0.01459s; TotalTimePerSample = 0.02279ms; SamplesPerSecond = 43874
|
||||
Epoch[ 1 of 3]-Minibatch[ 31- 40 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.94157934; EvalErr[0]PerSample = 0.89843750; TotalTime = 0.01459s; TotalTimePerSample = 0.02280ms; SamplesPerSecond = 43859
|
||||
Epoch[ 1 of 3]-Minibatch[ 41- 50 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.85668945; EvalErr[0]PerSample = 0.91093749; TotalTime = 0.01456s; TotalTimePerSample = 0.02275ms; SamplesPerSecond = 43953
|
||||
Epoch[ 1 of 3]-Minibatch[ 51- 60 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.72866368; EvalErr[0]PerSample = 0.89531249; TotalTime = 0.01450s; TotalTimePerSample = 0.02265ms; SamplesPerSecond = 44140
|
||||
Epoch[ 1 of 3]-Minibatch[ 61- 70 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.51809072; EvalErr[0]PerSample = 0.82968748; TotalTime = 0.01453s; TotalTimePerSample = 0.02271ms; SamplesPerSecond = 44034
|
||||
Epoch[ 1 of 3]-Minibatch[ 71- 80 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.48454905; EvalErr[0]PerSample = 0.80781251; TotalTime = 0.01452s; TotalTimePerSample = 0.02269ms; SamplesPerSecond = 44074
|
||||
Epoch[ 1 of 3]-Minibatch[ 81- 90 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.33829641; EvalErr[0]PerSample = 0.76875001; TotalTime = 0.01453s; TotalTimePerSample = 0.02271ms; SamplesPerSecond = 44037
|
||||
Epoch[ 1 of 3]-Minibatch[ 91- 100 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.50167227; EvalErr[0]PerSample = 0.79843748; TotalTime = 0.01447s; TotalTimePerSample = 0.02261ms; SamplesPerSecond = 44229
|
||||
WARNING: The same matrix with dim [1, 1] has been transferred between different devices for 20 times.
|
||||
Epoch[ 1 of 3]-Minibatch[ 101- 110 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.22861624; EvalErr[0]PerSample = 0.80000001; TotalTime = 0.01459s; TotalTimePerSample = 0.02279ms; SamplesPerSecond = 43874
|
||||
Epoch[ 1 of 3]-Minibatch[ 111- 120 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.32616878; EvalErr[0]PerSample = 0.79062498; TotalTime = 0.01449s; TotalTimePerSample = 0.02264ms; SamplesPerSecond = 44174
|
||||
Epoch[ 1 of 3]-Minibatch[ 121- 130 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.16897583; EvalErr[0]PerSample = 0.77968752; TotalTime = 0.01448s; TotalTimePerSample = 0.02262ms; SamplesPerSecond = 44201
|
||||
Epoch[ 1 of 3]-Minibatch[ 131- 140 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.08891916; EvalErr[0]PerSample = 0.77656251; TotalTime = 0.01442s; TotalTimePerSample = 0.02253ms; SamplesPerSecond = 44385
|
||||
Epoch[ 1 of 3]-Minibatch[ 141- 150 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.06004953; EvalErr[0]PerSample = 0.72968751; TotalTime = 0.01454s; TotalTimePerSample = 0.02271ms; SamplesPerSecond = 44031
|
||||
Epoch[ 1 of 3]-Minibatch[ 151- 160 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.91128540; EvalErr[0]PerSample = 0.69531250; TotalTime = 0.01446s; TotalTimePerSample = 0.02259ms; SamplesPerSecond = 44272
|
||||
Epoch[ 1 of 3]-Minibatch[ 161- 170 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.90172124; EvalErr[0]PerSample = 0.72968751; TotalTime = 0.01450s; TotalTimePerSample = 0.02266ms; SamplesPerSecond = 44128
|
||||
Epoch[ 1 of 3]-Minibatch[ 171- 180 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.73261714; EvalErr[0]PerSample = 0.65312499; TotalTime = 0.01447s; TotalTimePerSample = 0.02261ms; SamplesPerSecond = 44232
|
||||
Epoch[ 1 of 3]-Minibatch[ 181- 190 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.66515493; EvalErr[0]PerSample = 0.68437499; TotalTime = 0.01453s; TotalTimePerSample = 0.02270ms; SamplesPerSecond = 44061
|
||||
Epoch[ 1 of 3]-Minibatch[ 191- 200 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.67383432; EvalErr[0]PerSample = 0.66406250; TotalTime = 0.01449s; TotalTimePerSample = 0.02264ms; SamplesPerSecond = 44165
|
||||
Epoch[ 1 of 3]-Minibatch[ 201- 210 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.52869272; EvalErr[0]PerSample = 0.63593751; TotalTime = 0.01450s; TotalTimePerSample = 0.02266ms; SamplesPerSecond = 44134
|
||||
Epoch[ 1 of 3]-Minibatch[ 211- 220 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.60032344; EvalErr[0]PerSample = 0.66718751; TotalTime = 0.01450s; TotalTimePerSample = 0.02266ms; SamplesPerSecond = 44128
|
||||
Epoch[ 1 of 3]-Minibatch[ 221- 230 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.51134038; EvalErr[0]PerSample = 0.64843750; TotalTime = 0.01452s; TotalTimePerSample = 0.02268ms; SamplesPerSecond = 44086
|
||||
Epoch[ 1 of 3]-Minibatch[ 231- 240 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.45362544; EvalErr[0]PerSample = 0.63749999; TotalTime = 0.01452s; TotalTimePerSample = 0.02269ms; SamplesPerSecond = 44068
|
||||
Epoch[ 1 of 3]-Minibatch[ 241- 250 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.41640615; EvalErr[0]PerSample = 0.61562502; TotalTime = 0.01445s; TotalTimePerSample = 0.02258ms; SamplesPerSecond = 44287
|
||||
Epoch[ 1 of 3]-Minibatch[ 251- 260 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.39745474; EvalErr[0]PerSample = 0.62812501; TotalTime = 0.01447s; TotalTimePerSample = 0.02261ms; SamplesPerSecond = 44229
|
||||
Epoch[ 1 of 3]-Minibatch[ 261- 270 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.16415405; EvalErr[0]PerSample = 0.56718749; TotalTime = 0.01454s; TotalTimePerSample = 0.02272ms; SamplesPerSecond = 44013
|
||||
Epoch[ 1 of 3]-Minibatch[ 271- 280 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.30347300; EvalErr[0]PerSample = 0.63593751; TotalTime = 0.01454s; TotalTimePerSample = 0.02272ms; SamplesPerSecond = 44016
|
||||
Epoch[ 1 of 3]-Minibatch[ 281- 290 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.24398804; EvalErr[0]PerSample = 0.60937500; TotalTime = 0.01446s; TotalTimePerSample = 0.02260ms; SamplesPerSecond = 44253
|
||||
Epoch[ 1 of 3]-Minibatch[ 291- 300 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.15322256; EvalErr[0]PerSample = 0.57968748; TotalTime = 0.01447s; TotalTimePerSample = 0.02262ms; SamplesPerSecond = 44214
|
||||
Epoch[ 1 of 3]-Minibatch[ 301- 310 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.21664429; EvalErr[0]PerSample = 0.59531248; TotalTime = 0.01448s; TotalTimePerSample = 0.02262ms; SamplesPerSecond = 44208
|
||||
Epoch[ 1 of 3]-Minibatch[ 311- 320 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.25246572; EvalErr[0]PerSample = 0.60156250; TotalTime = 0.01442s; TotalTimePerSample = 0.02253ms; SamplesPerSecond = 44392
|
||||
Finished Epoch[1]: [Training Set] TrainLossPerSample = 3.0000031; EvalErrPerSample = 0.72836918; Ave LearnRatePerSample = 0.015625; EpochTime=0.4851
|
||||
Starting Epoch 2: learning rate per sample = 0.001953 momentum = 0.656119
|
||||
minibatchiterator: epoch 1: frames [20480..40960] (first utterance at frame 20480) with 1 datapasses
|
||||
Epoch[ 2 of 3]-Minibatch[ 1- 10 of 80]: SamplesSeen = 2560; TrainLossPerSample = 2.08151960; EvalErr[0]PerSample = 0.55859375; TotalTime = 0.03149s; TotalTimePerSample = 0.01230ms; SamplesPerSecond = 81290
|
||||
Epoch[ 2 of 3]-Minibatch[ 11- 20 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.98395634; EvalErr[0]PerSample = 0.54257810; TotalTime = 0.02336s; TotalTimePerSample = 0.00913ms; SamplesPerSecond = 109570
|
||||
Epoch[ 2 of 3]-Minibatch[ 21- 30 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.98575521; EvalErr[0]PerSample = 0.54492188; TotalTime = 0.02325s; TotalTimePerSample = 0.00908ms; SamplesPerSecond = 110116
|
||||
Epoch[ 2 of 3]-Minibatch[ 31- 40 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.90484965; EvalErr[0]PerSample = 0.53164065; TotalTime = 0.02321s; TotalTimePerSample = 0.00906ms; SamplesPerSecond = 110316
|
||||
Epoch[ 2 of 3]-Minibatch[ 41- 50 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.88324130; EvalErr[0]PerSample = 0.52539063; TotalTime = 0.02328s; TotalTimePerSample = 0.00909ms; SamplesPerSecond = 109975
|
||||
Epoch[ 2 of 3]-Minibatch[ 51- 60 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.89109266; EvalErr[0]PerSample = 0.53359377; TotalTime = 0.02325s; TotalTimePerSample = 0.00908ms; SamplesPerSecond = 110093
|
||||
Epoch[ 2 of 3]-Minibatch[ 61- 70 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.89496076; EvalErr[0]PerSample = 0.52890623; TotalTime = 0.02326s; TotalTimePerSample = 0.00909ms; SamplesPerSecond = 110055
|
||||
Epoch[ 2 of 3]-Minibatch[ 71- 80 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.85944366; EvalErr[0]PerSample = 0.52265626; TotalTime = 0.02296s; TotalTimePerSample = 0.00897ms; SamplesPerSecond = 111473
|
||||
Finished Epoch[2]: [Training Set] TrainLossPerSample = 1.9356024; EvalErrPerSample = 0.53603518; Ave LearnRatePerSample = 0.001953125; EpochTime=0.195263
|
||||
Starting Epoch 3: learning rate per sample = 0.000098 momentum = 0.656119
|
||||
minibatchiterator: epoch 2: frames [40960..61440] (first utterance at frame 40960) with 1 datapasses
|
||||
Epoch[ 3 of 3]-Minibatch[ 1- 10 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.86752820; EvalErr[0]PerSample = 0.52177733; TotalTime = 0.08160s; TotalTimePerSample = 0.00797ms; SamplesPerSecond = 125485
|
||||
Epoch[ 3 of 3]-Minibatch[ 11- 20 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.87358737; EvalErr[0]PerSample = 0.51542968; TotalTime = 0.05742s; TotalTimePerSample = 0.00561ms; SamplesPerSecond = 178319
|
||||
Finished Epoch[3]: [Training Set] TrainLossPerSample = 1.8705578; EvalErrPerSample = 0.5186035; Ave LearnRatePerSample = 9.765625146e-005; EpochTime=0.142001
|
||||
COMPLETED
|
||||
=== Deleting last epoch data
|
||||
==== Re-running from checkpoint
|
||||
-------------------------------------------------------------------
|
||||
Build info:
|
||||
|
||||
Built time: Aug 11 2015 16:18:17
|
||||
Last modified date: Tue Aug 11 16:16:08 2015
|
||||
Built by svcphil on dphaim-26-new
|
||||
Build Path: C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\MachineLearning\CNTK\
|
||||
CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.0
|
||||
Build Branch: master
|
||||
Build SHA1: 397cc7cc16c00b1c12864d331c0729fde7a1bde3
|
||||
-------------------------------------------------------------------
|
||||
running on dphaim-26-new at 2015/08/11 17:47:19
|
||||
command line options:
|
||||
configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG (VARIABLES NOT RESOLVED) >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=$DeviceId$
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=$RunDir$/models/cntkSpeech.dnn
|
||||
deviceId=$DeviceId$
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=$DataDir$/glob_0000.mlf
|
||||
labelMappingFile=$DataDir$/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG (VARIABLES NOT RESOLVED) <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=Auto
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> PROCESSED CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
configparameters: cntk.config:command=speechTrain
|
||||
configparameters: cntk.config:DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
configparameters: cntk.config:deviceId=Auto
|
||||
configparameters: cntk.config:parallelTrain=false
|
||||
configparameters: cntk.config:precision=float
|
||||
configparameters: cntk.config:RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu
|
||||
configparameters: cntk.config:speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< PROCESSED CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
command: speechTrain
|
||||
precision = float
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
LockDevice: Capture device 2 and lock it for exclusive use
|
||||
LockDevice: Capture device 3 and lock it for exclusive use
|
||||
LockDevice: Capture device 0 and lock it for exclusive use
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
SimpleNetworkBuilder Using GPU 1
|
||||
reading script file glob_0000.scp ... 948 entries
|
||||
trainlayer: OOV-exclusion code enabled, but no unigram specified to derive the word set from, so you won't get OOV exclusion
|
||||
total 132 state names in state list C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
htkmlfreader: reading MLF file C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf ... total 948 entries
|
||||
...............................................................................................feature set 0: 252734 frames in 948 out of 948 utterances
|
||||
label set 0: 129 classes
|
||||
minibatchutterancesource: 948 utterances grouped into 3 chunks, av. chunk size: 316.0 utterances, 84244.7 frames
|
||||
Starting from checkpoint. Load Network From File C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_cpu/models/cntkSpeech.dnn.2.
|
||||
|
||||
|
||||
Printing Gradient Computation Node Order ...
|
||||
|
||||
CrossEntropyWithSoftmax[0, 0] = CrossEntropyWithSoftmax(labels[132, 256], HLast[0, 0])
|
||||
HLast[0, 0] = Plus(W2*H1[0, 0], B2[132, 1])
|
||||
B2[132, 1] = LearnableParameter
|
||||
W2*H1[0, 0] = Times(W2[132, 512], H2[0, 0])
|
||||
H2[0, 0] = Sigmoid(W1*H1+B1[0, 0])
|
||||
W1*H1+B1[0, 0] = Plus(W1*H1[0, 0], B1[512, 1])
|
||||
B1[512, 1] = LearnableParameter
|
||||
W1*H1[0, 0] = Times(W1[512, 512], H1[0, 0])
|
||||
H1[0, 0] = Sigmoid(W0*features+B0[0, 0])
|
||||
W0*features+B0[0, 0] = Plus(W0*features[0, 0], B0[512, 1])
|
||||
B0[512, 1] = LearnableParameter
|
||||
W0*features[0, 0] = Times(W0[512, 363], MVNormalizedFeatures[0, 0])
|
||||
MVNormalizedFeatures[0, 0] = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
InvStdOfFeatures[363, 1] = InvStdDev(features[363, 256])
|
||||
MeanOfFeatures[363, 1] = Mean(features[363, 256])
|
||||
features[363, 256] = InputValue
|
||||
W0[512, 363] = LearnableParameter
|
||||
W1[512, 512] = LearnableParameter
|
||||
W2[132, 512] = LearnableParameter
|
||||
labels[132, 256] = InputValue
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 256], HLast[132, 256])
|
||||
|
||||
|
||||
|
||||
Validating node ScaledLogLikelihood
|
||||
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> labels = InputValue
|
||||
Validating --> Prior = Mean(labels[132, 256])
|
||||
Validating --> LogOfPrior = Log(Prior[132, 1])
|
||||
Validating --> ScaledLogLikelihood = Minus(HLast[132, 256], LogOfPrior[132, 1])
|
||||
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 256], HLast[132, 256])
|
||||
|
||||
GetTrainCriterionNodes ...
|
||||
GetEvalCriterionNodes ...
|
||||
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 256], HLast[132, 256])
|
||||
|
||||
No PreCompute nodes found, skipping PreCompute step
|
||||
Set Max Temp Mem Size For Convolution Nodes to 0 samples.
|
||||
Starting Epoch 3: learning rate per sample = 0.000098 momentum = 0.656119
|
||||
minibatchiterator: epoch 2: frames [40960..61440] (first utterance at frame 40960) with 1 datapasses
|
||||
requiredata: determined feature kind as 33-dimensional 'USER' with frame shift 10.0 ms
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 1024])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 1024])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 1024], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 1024])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 1024], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 1024])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 1024])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 1024], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 1024])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 1024])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 1024], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 1024], HLast[132, 1024])
|
||||
|
||||
Epoch[ 3 of 3]-Minibatch[ 1- 10 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.86752820; EvalErr[0]PerSample = 0.52177733; TotalTime = 0.40600s; TotalTimePerSample = 0.03965ms; SamplesPerSecond = 25221
|
||||
Epoch[ 3 of 3]-Minibatch[ 11- 20 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.87358737; EvalErr[0]PerSample = 0.51542968; TotalTime = 0.05538s; TotalTimePerSample = 0.00541ms; SamplesPerSecond = 184900
|
||||
Finished Epoch[3]: [Training Set] TrainLossPerSample = 1.8705578; EvalErrPerSample = 0.5186035; Ave LearnRatePerSample = 9.765625146e-005; EpochTime=0.692077
|
||||
COMPLETED
|
||||
|
|
@ -0,0 +1,738 @@
|
|||
=== Running /cygdrive/c/Users/svcphil/workspace.vlivan/CNTK-Build-Windows/x64/release/cntk.exe configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
-------------------------------------------------------------------
|
||||
Build info:
|
||||
|
||||
Built time: Aug 11 2015 16:18:17
|
||||
Last modified date: Tue Aug 11 16:16:08 2015
|
||||
Built by svcphil on dphaim-26-new
|
||||
Build Path: C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\MachineLearning\CNTK\
|
||||
CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.0
|
||||
Build Branch: master
|
||||
Build SHA1: 397cc7cc16c00b1c12864d331c0729fde7a1bde3
|
||||
-------------------------------------------------------------------
|
||||
running on dphaim-26-new at 2015/08/11 17:47:26
|
||||
command line options:
|
||||
configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG (VARIABLES NOT RESOLVED) >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=$DeviceId$
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=$RunDir$/models/cntkSpeech.dnn
|
||||
deviceId=$DeviceId$
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=$DataDir$/glob_0000.mlf
|
||||
labelMappingFile=$DataDir$/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG (VARIABLES NOT RESOLVED) <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=Auto
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> PROCESSED CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
configparameters: cntk.config:command=speechTrain
|
||||
configparameters: cntk.config:DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
configparameters: cntk.config:deviceId=Auto
|
||||
configparameters: cntk.config:parallelTrain=false
|
||||
configparameters: cntk.config:precision=float
|
||||
configparameters: cntk.config:RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
configparameters: cntk.config:speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< PROCESSED CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
command: speechTrain
|
||||
precision = float
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
LockDevice: Capture device 2 and lock it for exclusive use
|
||||
LockDevice: Capture device 3 and lock it for exclusive use
|
||||
LockDevice: Capture device 0 and lock it for exclusive use
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
SimpleNetworkBuilder Using GPU 1
|
||||
reading script file glob_0000.scp ... 948 entries
|
||||
trainlayer: OOV-exclusion code enabled, but no unigram specified to derive the word set from, so you won't get OOV exclusion
|
||||
total 132 state names in state list C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
htkmlfreader: reading MLF file C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf ... total 948 entries
|
||||
...............................................................................................feature set 0: 252734 frames in 948 out of 948 utterances
|
||||
label set 0: 129 classes
|
||||
minibatchutterancesource: 948 utterances grouped into 3 chunks, av. chunk size: 316.0 utterances, 84244.7 frames
|
||||
GetTrainCriterionNodes ...
|
||||
GetEvalCriterionNodes ...
|
||||
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 3])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 3])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 3], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 3])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 3], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 3])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 3])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 3], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 3])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 3])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 3], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 3], HLast[132, 3])
|
||||
|
||||
Found 3 PreCompute nodes
|
||||
NodeName: InvStdOfFeatures
|
||||
NodeName: MeanOfFeatures
|
||||
NodeName: Prior
|
||||
minibatchiterator: epoch 0: frames [0..252734] (first utterance at frame 0) with 1 datapasses
|
||||
requiredata: determined feature kind as 33-dimensional 'USER' with frame shift 10.0 ms
|
||||
|
||||
|
||||
Validating node InvStdOfFeatures
|
||||
|
||||
Validating --> features = InputValue
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 64])
|
||||
|
||||
|
||||
|
||||
Validating node MeanOfFeatures
|
||||
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 64])
|
||||
|
||||
|
||||
|
||||
Validating node Prior
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> Prior = Mean(labels[132, 64])
|
||||
|
||||
Set Max Temp Mem Size For Convolution Nodes to 0 samples.
|
||||
Starting Epoch 1: learning rate per sample = 0.015625 momentum = 0.900000
|
||||
minibatchiterator: epoch 0: frames [0..20480] (first utterance at frame 0) with 1 datapasses
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 64])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 64])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 64], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 64])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 64], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 64])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 64])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 64], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 64])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 64])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 64], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 64], HLast[132, 64])
|
||||
|
||||
Epoch[ 1 of 3]-Minibatch[ 1- 10 of 320]: SamplesSeen = 640; TrainLossPerSample = 4.45646143; EvalErr[0]PerSample = 0.92500001; TotalTime = 0.03190s; TotalTimePerSample = 0.04985ms; SamplesPerSecond = 20061
|
||||
Epoch[ 1 of 3]-Minibatch[ 11- 20 of 320]: SamplesSeen = 640; TrainLossPerSample = 4.22315693; EvalErr[0]PerSample = 0.90156251; TotalTime = 0.02454s; TotalTimePerSample = 0.03835ms; SamplesPerSecond = 26075
|
||||
Epoch[ 1 of 3]-Minibatch[ 21- 30 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.95180511; EvalErr[0]PerSample = 0.84687501; TotalTime = 0.02438s; TotalTimePerSample = 0.03809ms; SamplesPerSecond = 26254
|
||||
Epoch[ 1 of 3]-Minibatch[ 31- 40 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.94157934; EvalErr[0]PerSample = 0.89843750; TotalTime = 0.02445s; TotalTimePerSample = 0.03820ms; SamplesPerSecond = 26181
|
||||
Epoch[ 1 of 3]-Minibatch[ 41- 50 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.85668945; EvalErr[0]PerSample = 0.91093749; TotalTime = 0.02429s; TotalTimePerSample = 0.03795ms; SamplesPerSecond = 26352
|
||||
Epoch[ 1 of 3]-Minibatch[ 51- 60 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.72866368; EvalErr[0]PerSample = 0.89531249; TotalTime = 0.02445s; TotalTimePerSample = 0.03820ms; SamplesPerSecond = 26178
|
||||
Epoch[ 1 of 3]-Minibatch[ 61- 70 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.51809072; EvalErr[0]PerSample = 0.82968748; TotalTime = 0.02423s; TotalTimePerSample = 0.03786ms; SamplesPerSecond = 26415
|
||||
Epoch[ 1 of 3]-Minibatch[ 71- 80 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.48454905; EvalErr[0]PerSample = 0.80781251; TotalTime = 0.02249s; TotalTimePerSample = 0.03514ms; SamplesPerSecond = 28457
|
||||
Epoch[ 1 of 3]-Minibatch[ 81- 90 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.33829641; EvalErr[0]PerSample = 0.76875001; TotalTime = 0.02169s; TotalTimePerSample = 0.03390ms; SamplesPerSecond = 29501
|
||||
Epoch[ 1 of 3]-Minibatch[ 91- 100 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.50167227; EvalErr[0]PerSample = 0.79843748; TotalTime = 0.02178s; TotalTimePerSample = 0.03403ms; SamplesPerSecond = 29386
|
||||
WARNING: The same matrix with dim [1, 1] has been transferred between different devices for 20 times.
|
||||
Epoch[ 1 of 3]-Minibatch[ 101- 110 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.22861624; EvalErr[0]PerSample = 0.80000001; TotalTime = 0.02166s; TotalTimePerSample = 0.03385ms; SamplesPerSecond = 29546
|
||||
Epoch[ 1 of 3]-Minibatch[ 111- 120 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.32616878; EvalErr[0]PerSample = 0.79062498; TotalTime = 0.02063s; TotalTimePerSample = 0.03224ms; SamplesPerSecond = 31018
|
||||
Epoch[ 1 of 3]-Minibatch[ 121- 130 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.16897583; EvalErr[0]PerSample = 0.77968752; TotalTime = 0.01950s; TotalTimePerSample = 0.03048ms; SamplesPerSecond = 32813
|
||||
Epoch[ 1 of 3]-Minibatch[ 131- 140 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.08891916; EvalErr[0]PerSample = 0.77656251; TotalTime = 0.01961s; TotalTimePerSample = 0.03063ms; SamplesPerSecond = 32644
|
||||
Epoch[ 1 of 3]-Minibatch[ 141- 150 of 320]: SamplesSeen = 640; TrainLossPerSample = 3.06004953; EvalErr[0]PerSample = 0.72968751; TotalTime = 0.01950s; TotalTimePerSample = 0.03046ms; SamplesPerSecond = 32825
|
||||
Epoch[ 1 of 3]-Minibatch[ 151- 160 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.91128540; EvalErr[0]PerSample = 0.69531250; TotalTime = 0.01965s; TotalTimePerSample = 0.03070ms; SamplesPerSecond = 32571
|
||||
Epoch[ 1 of 3]-Minibatch[ 161- 170 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.90172124; EvalErr[0]PerSample = 0.72968751; TotalTime = 0.01828s; TotalTimePerSample = 0.02857ms; SamplesPerSecond = 35003
|
||||
Epoch[ 1 of 3]-Minibatch[ 171- 180 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.73261714; EvalErr[0]PerSample = 0.65312499; TotalTime = 0.01799s; TotalTimePerSample = 0.02811ms; SamplesPerSecond = 35569
|
||||
Epoch[ 1 of 3]-Minibatch[ 181- 190 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.66515493; EvalErr[0]PerSample = 0.68437499; TotalTime = 0.01789s; TotalTimePerSample = 0.02796ms; SamplesPerSecond = 35766
|
||||
Epoch[ 1 of 3]-Minibatch[ 191- 200 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.67383432; EvalErr[0]PerSample = 0.66406250; TotalTime = 0.01792s; TotalTimePerSample = 0.02800ms; SamplesPerSecond = 35708
|
||||
Epoch[ 1 of 3]-Minibatch[ 201- 210 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.52869272; EvalErr[0]PerSample = 0.63593751; TotalTime = 0.01805s; TotalTimePerSample = 0.02821ms; SamplesPerSecond = 35451
|
||||
Epoch[ 1 of 3]-Minibatch[ 211- 220 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.60032344; EvalErr[0]PerSample = 0.66718751; TotalTime = 0.01696s; TotalTimePerSample = 0.02650ms; SamplesPerSecond = 37738
|
||||
Epoch[ 1 of 3]-Minibatch[ 221- 230 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.51134038; EvalErr[0]PerSample = 0.64843750; TotalTime = 0.01658s; TotalTimePerSample = 0.02591ms; SamplesPerSecond = 38598
|
||||
Epoch[ 1 of 3]-Minibatch[ 231- 240 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.45362544; EvalErr[0]PerSample = 0.63749999; TotalTime = 0.01663s; TotalTimePerSample = 0.02598ms; SamplesPerSecond = 38491
|
||||
Epoch[ 1 of 3]-Minibatch[ 241- 250 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.41640615; EvalErr[0]PerSample = 0.61562502; TotalTime = 0.01670s; TotalTimePerSample = 0.02610ms; SamplesPerSecond = 38321
|
||||
Epoch[ 1 of 3]-Minibatch[ 251- 260 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.39745474; EvalErr[0]PerSample = 0.62812501; TotalTime = 0.01672s; TotalTimePerSample = 0.02612ms; SamplesPerSecond = 38279
|
||||
Epoch[ 1 of 3]-Minibatch[ 261- 270 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.16415405; EvalErr[0]PerSample = 0.56718749; TotalTime = 0.01621s; TotalTimePerSample = 0.02533ms; SamplesPerSecond = 39481
|
||||
Epoch[ 1 of 3]-Minibatch[ 271- 280 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.30347300; EvalErr[0]PerSample = 0.63593751; TotalTime = 0.01583s; TotalTimePerSample = 0.02474ms; SamplesPerSecond = 40427
|
||||
Epoch[ 1 of 3]-Minibatch[ 281- 290 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.24398804; EvalErr[0]PerSample = 0.60937500; TotalTime = 0.01579s; TotalTimePerSample = 0.02467ms; SamplesPerSecond = 40542
|
||||
Epoch[ 1 of 3]-Minibatch[ 291- 300 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.15322256; EvalErr[0]PerSample = 0.57968748; TotalTime = 0.01582s; TotalTimePerSample = 0.02472ms; SamplesPerSecond = 40447
|
||||
Epoch[ 1 of 3]-Minibatch[ 301- 310 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.21664429; EvalErr[0]PerSample = 0.59531248; TotalTime = 0.01570s; TotalTimePerSample = 0.02453ms; SamplesPerSecond = 40761
|
||||
Epoch[ 1 of 3]-Minibatch[ 311- 320 of 320]: SamplesSeen = 640; TrainLossPerSample = 2.25246572; EvalErr[0]PerSample = 0.60156250; TotalTime = 0.01556s; TotalTimePerSample = 0.02431ms; SamplesPerSecond = 41139
|
||||
Finished Epoch[1]: [Training Set] TrainLossPerSample = 3.0000031; EvalErrPerSample = 0.72836918; Ave LearnRatePerSample = 0.015625; EpochTime=0.657568
|
||||
Starting Epoch 2: learning rate per sample = 0.001953 momentum = 0.656119
|
||||
minibatchiterator: epoch 1: frames [20480..40960] (first utterance at frame 20480) with 1 datapasses
|
||||
Epoch[ 2 of 3]-Minibatch[ 1- 10 of 80]: SamplesSeen = 2560; TrainLossPerSample = 2.08151960; EvalErr[0]PerSample = 0.55859375; TotalTime = 0.03143s; TotalTimePerSample = 0.01228ms; SamplesPerSecond = 81456
|
||||
Epoch[ 2 of 3]-Minibatch[ 11- 20 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.98395634; EvalErr[0]PerSample = 0.54257810; TotalTime = 0.02295s; TotalTimePerSample = 0.00896ms; SamplesPerSecond = 111561
|
||||
Epoch[ 2 of 3]-Minibatch[ 21- 30 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.98575521; EvalErr[0]PerSample = 0.54492188; TotalTime = 0.02287s; TotalTimePerSample = 0.00893ms; SamplesPerSecond = 111951
|
||||
Epoch[ 2 of 3]-Minibatch[ 31- 40 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.90484965; EvalErr[0]PerSample = 0.53164065; TotalTime = 0.02284s; TotalTimePerSample = 0.00892ms; SamplesPerSecond = 112069
|
||||
Epoch[ 2 of 3]-Minibatch[ 41- 50 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.88324130; EvalErr[0]PerSample = 0.52539063; TotalTime = 0.02277s; TotalTimePerSample = 0.00889ms; SamplesPerSecond = 112448
|
||||
Epoch[ 2 of 3]-Minibatch[ 51- 60 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.89109266; EvalErr[0]PerSample = 0.53359377; TotalTime = 0.02287s; TotalTimePerSample = 0.00894ms; SamplesPerSecond = 111917
|
||||
Epoch[ 2 of 3]-Minibatch[ 61- 70 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.89496076; EvalErr[0]PerSample = 0.52890623; TotalTime = 0.02279s; TotalTimePerSample = 0.00890ms; SamplesPerSecond = 112325
|
||||
Epoch[ 2 of 3]-Minibatch[ 71- 80 of 80]: SamplesSeen = 2560; TrainLossPerSample = 1.85944366; EvalErr[0]PerSample = 0.52265626; TotalTime = 0.02265s; TotalTimePerSample = 0.00885ms; SamplesPerSecond = 113044
|
||||
Finished Epoch[2]: [Training Set] TrainLossPerSample = 1.9356024; EvalErrPerSample = 0.53603518; Ave LearnRatePerSample = 0.001953125; EpochTime=0.192318
|
||||
Starting Epoch 3: learning rate per sample = 0.000098 momentum = 0.656119
|
||||
minibatchiterator: epoch 2: frames [40960..61440] (first utterance at frame 40960) with 1 datapasses
|
||||
Epoch[ 3 of 3]-Minibatch[ 1- 10 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.86752820; EvalErr[0]PerSample = 0.52177733; TotalTime = 0.08080s; TotalTimePerSample = 0.00789ms; SamplesPerSecond = 126735
|
||||
Epoch[ 3 of 3]-Minibatch[ 11- 20 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.87358737; EvalErr[0]PerSample = 0.51542968; TotalTime = 0.05544s; TotalTimePerSample = 0.00541ms; SamplesPerSecond = 184694
|
||||
Finished Epoch[3]: [Training Set] TrainLossPerSample = 1.8705578; EvalErrPerSample = 0.5186035; Ave LearnRatePerSample = 9.765625146e-005; EpochTime=0.139063
|
||||
COMPLETED
|
||||
=== Deleting last epoch data
|
||||
==== Re-running from checkpoint
|
||||
-------------------------------------------------------------------
|
||||
Build info:
|
||||
|
||||
Built time: Aug 11 2015 16:18:17
|
||||
Last modified date: Tue Aug 11 16:16:08 2015
|
||||
Built by svcphil on dphaim-26-new
|
||||
Build Path: C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\MachineLearning\CNTK\
|
||||
CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.0
|
||||
Build Branch: master
|
||||
Build SHA1: 397cc7cc16c00b1c12864d331c0729fde7a1bde3
|
||||
-------------------------------------------------------------------
|
||||
running on dphaim-26-new at 2015/08/11 17:47:34
|
||||
command line options:
|
||||
configFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\QuickE2E\cntk.config RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data DeviceId=Auto
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG (VARIABLES NOT RESOLVED) >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=$DeviceId$
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=$RunDir$/models/cntkSpeech.dnn
|
||||
deviceId=$DeviceId$
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=$DataDir$/glob_0000.mlf
|
||||
labelMappingFile=$DataDir$/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG (VARIABLES NOT RESOLVED) <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> RAW CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
precision=float
|
||||
command=speechTrain
|
||||
deviceId=Auto
|
||||
parallelTrain=false
|
||||
speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
DeviceId=Auto
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< RAW CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
|
||||
>>>>>>>>>>>>>>>>>>>> PROCESSED CONFIG WITH ALL VARIABLES RESOLVED >>>>>>>>>>>>>>>>>>>>
|
||||
configparameters: cntk.config:command=speechTrain
|
||||
configparameters: cntk.config:DataDir=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data
|
||||
configparameters: cntk.config:deviceId=Auto
|
||||
configparameters: cntk.config:parallelTrain=false
|
||||
configparameters: cntk.config:precision=float
|
||||
configparameters: cntk.config:RunDir=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu
|
||||
configparameters: cntk.config:speechTrain=[
|
||||
action=train
|
||||
modelPath=C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu/models/cntkSpeech.dnn
|
||||
deviceId=Auto
|
||||
traceLevel=1
|
||||
SimpleNetworkBuilder=[
|
||||
layerSizes=363:512:512:132
|
||||
trainingCriterion=CrossEntropyWithSoftmax
|
||||
evalCriterion=ErrorPrediction
|
||||
layerTypes=Sigmoid
|
||||
initValueScale=1.0
|
||||
applyMeanVarNorm=true
|
||||
uniformInit=true
|
||||
needPrior=true
|
||||
]
|
||||
SGD=[
|
||||
epochSize=20480
|
||||
minibatchSize=64:256:1024:
|
||||
learningRatesPerMB=1.0:0.5:0.1
|
||||
numMBsToShowResult=10
|
||||
momentumPerMB=0.9:0.656119
|
||||
dropoutRate=0.0
|
||||
maxEpochs=3
|
||||
keepCheckPointFiles=true
|
||||
AutoAdjust=[
|
||||
reduceLearnRateIfImproveLessThan=0
|
||||
loadBestModel=true
|
||||
increaseLearnRateIfImproveMoreThan=1000000000
|
||||
learnRateDecreaseFactor=0.5
|
||||
learnRateIncreaseFactor=1.382
|
||||
autoAdjustLR=AdjustAfterEpoch
|
||||
]
|
||||
clippingThresholdPerSample=1#INF
|
||||
]
|
||||
reader=[
|
||||
readerType=HTKMLFReader
|
||||
readMethod=blockRandomize
|
||||
miniBatchMode=Partial
|
||||
randomize=Auto
|
||||
verbosity=0
|
||||
features=[
|
||||
dim=363
|
||||
type=Real
|
||||
scpFile=glob_0000.scp
|
||||
]
|
||||
labels=[
|
||||
mlfFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf
|
||||
labelMappingFile=C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
labelDim=132
|
||||
labelType=Category
|
||||
]
|
||||
]
|
||||
]
|
||||
|
||||
<<<<<<<<<<<<<<<<<<<< PROCESSED CONFIG WITH ALL VARIABLES RESOLVED <<<<<<<<<<<<<<<<<<<<
|
||||
command: speechTrain
|
||||
precision = float
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
LockDevice: Capture device 2 and lock it for exclusive use
|
||||
LockDevice: Capture device 3 and lock it for exclusive use
|
||||
LockDevice: Capture device 0 and lock it for exclusive use
|
||||
LockDevice: Capture device 1 and lock it for exclusive use
|
||||
SimpleNetworkBuilder Using GPU 1
|
||||
reading script file glob_0000.scp ... 948 entries
|
||||
trainlayer: OOV-exclusion code enabled, but no unigram specified to derive the word set from, so you won't get OOV exclusion
|
||||
total 132 state names in state list C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/state.list
|
||||
htkmlfreader: reading MLF file C:\Users\svcphil\workspace.vlivan\CNTK-Build-Windows\Tests\Speech\Data/glob_0000.mlf ... total 948 entries
|
||||
...............................................................................................feature set 0: 252734 frames in 948 out of 948 utterances
|
||||
label set 0: 129 classes
|
||||
minibatchutterancesource: 948 utterances grouped into 3 chunks, av. chunk size: 316.0 utterances, 84244.7 frames
|
||||
Starting from checkpoint. Load Network From File C:\Users\svcphil\AppData\Local\Temp\2\cntk-test-20150811174551.851046\Speech_QuickE2E@release_gpu/models/cntkSpeech.dnn.2.
|
||||
|
||||
|
||||
Printing Gradient Computation Node Order ...
|
||||
|
||||
CrossEntropyWithSoftmax[0, 0] = CrossEntropyWithSoftmax(labels[132, 256], HLast[0, 0])
|
||||
HLast[0, 0] = Plus(W2*H1[0, 0], B2[132, 1])
|
||||
B2[132, 1] = LearnableParameter
|
||||
W2*H1[0, 0] = Times(W2[132, 512], H2[0, 0])
|
||||
H2[0, 0] = Sigmoid(W1*H1+B1[0, 0])
|
||||
W1*H1+B1[0, 0] = Plus(W1*H1[0, 0], B1[512, 1])
|
||||
B1[512, 1] = LearnableParameter
|
||||
W1*H1[0, 0] = Times(W1[512, 512], H1[0, 0])
|
||||
H1[0, 0] = Sigmoid(W0*features+B0[0, 0])
|
||||
W0*features+B0[0, 0] = Plus(W0*features[0, 0], B0[512, 1])
|
||||
B0[512, 1] = LearnableParameter
|
||||
W0*features[0, 0] = Times(W0[512, 363], MVNormalizedFeatures[0, 0])
|
||||
MVNormalizedFeatures[0, 0] = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
InvStdOfFeatures[363, 1] = InvStdDev(features[363, 256])
|
||||
MeanOfFeatures[363, 1] = Mean(features[363, 256])
|
||||
features[363, 256] = InputValue
|
||||
W0[512, 363] = LearnableParameter
|
||||
W1[512, 512] = LearnableParameter
|
||||
W2[132, 512] = LearnableParameter
|
||||
labels[132, 256] = InputValue
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 256], HLast[132, 256])
|
||||
|
||||
|
||||
|
||||
Validating node ScaledLogLikelihood
|
||||
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> labels = InputValue
|
||||
Validating --> Prior = Mean(labels[132, 256])
|
||||
Validating --> LogOfPrior = Log(Prior[132, 1])
|
||||
Validating --> ScaledLogLikelihood = Minus(HLast[132, 256], LogOfPrior[132, 1])
|
||||
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 256], HLast[132, 256])
|
||||
|
||||
GetTrainCriterionNodes ...
|
||||
GetEvalCriterionNodes ...
|
||||
|
||||
|
||||
Validating node CrossEntropyWithSoftmax
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 256])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 256])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 256], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 256])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 256], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 256])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 256])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 256], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 256])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 256])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 256], B2[132, 1])
|
||||
Validating --> CrossEntropyWithSoftmax = CrossEntropyWithSoftmax(labels[132, 256], HLast[132, 256])
|
||||
|
||||
No PreCompute nodes found, skipping PreCompute step
|
||||
Set Max Temp Mem Size For Convolution Nodes to 0 samples.
|
||||
Starting Epoch 3: learning rate per sample = 0.000098 momentum = 0.656119
|
||||
minibatchiterator: epoch 2: frames [40960..61440] (first utterance at frame 40960) with 1 datapasses
|
||||
requiredata: determined feature kind as 33-dimensional 'USER' with frame shift 10.0 ms
|
||||
|
||||
|
||||
Validating node EvalErrorPrediction
|
||||
|
||||
Validating --> labels = InputValue
|
||||
Validating --> W2 = LearnableParameter
|
||||
Validating --> W1 = LearnableParameter
|
||||
Validating --> W0 = LearnableParameter
|
||||
Validating --> features = InputValue
|
||||
Validating --> MeanOfFeatures = Mean(features[363, 1024])
|
||||
Validating --> InvStdOfFeatures = InvStdDev(features[363, 1024])
|
||||
Validating --> MVNormalizedFeatures = PerDimMeanVarNormalization(features[363, 1024], MeanOfFeatures[363, 1], InvStdOfFeatures[363, 1])
|
||||
Validating --> W0*features = Times(W0[512, 363], MVNormalizedFeatures[363, 1024])
|
||||
Validating --> B0 = LearnableParameter
|
||||
Validating --> W0*features+B0 = Plus(W0*features[512, 1024], B0[512, 1])
|
||||
Validating --> H1 = Sigmoid(W0*features+B0[512, 1024])
|
||||
Validating --> W1*H1 = Times(W1[512, 512], H1[512, 1024])
|
||||
Validating --> B1 = LearnableParameter
|
||||
Validating --> W1*H1+B1 = Plus(W1*H1[512, 1024], B1[512, 1])
|
||||
Validating --> H2 = Sigmoid(W1*H1+B1[512, 1024])
|
||||
Validating --> W2*H1 = Times(W2[132, 512], H2[512, 1024])
|
||||
Validating --> B2 = LearnableParameter
|
||||
Validating --> HLast = Plus(W2*H1[132, 1024], B2[132, 1])
|
||||
Validating --> EvalErrorPrediction = ErrorPrediction(labels[132, 1024], HLast[132, 1024])
|
||||
|
||||
Epoch[ 3 of 3]-Minibatch[ 1- 10 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.86752820; EvalErr[0]PerSample = 0.52177733; TotalTime = 0.42093s; TotalTimePerSample = 0.04111ms; SamplesPerSecond = 24327
|
||||
Epoch[ 3 of 3]-Minibatch[ 11- 20 of 20]: SamplesSeen = 10240; TrainLossPerSample = 1.87358737; EvalErr[0]PerSample = 0.51542968; TotalTime = 0.05521s; TotalTimePerSample = 0.00539ms; SamplesPerSecond = 185480
|
||||
Finished Epoch[3]: [Training Set] TrainLossPerSample = 1.8705578; EvalErrPerSample = 0.5186035; Ave LearnRatePerSample = 9.765625146e-005; EpochTime=0.690137
|
||||
COMPLETED
|
||||
|
|
@ -1,17 +1,28 @@
|
|||
#!/bin/bash
|
||||
CNTK_BINARY=$TEST_BUILD_LOCATION/x86_64.gpu.$TEST_FLAVOR.acml/cntk
|
||||
if [ "$TEST_DEVICE" == "CPU" ]; then
|
||||
CNTK_DEVICE_ID=-1
|
||||
else
|
||||
CNTK_DEVICE_ID=Auto
|
||||
fi
|
||||
CNTK_ARGS="configFile=$TEST_DIR/cntk.config RunDir=$TEST_RUN_DIR DataDir=$TEST_DATA_DIR DeviceId=$CNTK_DEVICE_ID"
|
||||
|
||||
configFile=$TEST_DIR/cntk.config
|
||||
RunDir=$TEST_RUN_DIR
|
||||
DataDir=$TEST_DATA_DIR
|
||||
|
||||
if [ "$OS" == "Windows_NT" ]; then
|
||||
# When running on cygwin translating /cygdrive/xxx paths to proper windows paths:
|
||||
configFile=$(cygpath -aw $configFile)
|
||||
RunDir=$(cygpath -aw $RunDir)
|
||||
DataDir=$(cygpath -aw $DataDir)
|
||||
fi
|
||||
|
||||
CNTK_ARGS="configFile=$configFile RunDir=$RunDir DataDir=$DataDir DeviceId=$CNTK_DEVICE_ID"
|
||||
MODELS_DIR=$TEST_RUN_DIR/models
|
||||
[ -d $MODELS_DIR ] && rm -rf $MODELS_DIR
|
||||
mkdir -p $MODELS_DIR || exit $?
|
||||
echo === Running $CNTK_BINARY $CNTK_ARGS
|
||||
$CNTK_BINARY $CNTK_ARGS || exit $?
|
||||
echo === Running $TEST_CNTK_BINARY $CNTK_ARGS
|
||||
$TEST_CNTK_BINARY $CNTK_ARGS || exit $?
|
||||
echo === Deleting last epoch data
|
||||
rm $TEST_RUN_DIR/models/*.dnn
|
||||
echo ==== Re-running from checkpoint
|
||||
$CNTK_BINARY $CNTK_ARGS || exit $?
|
||||
$TEST_CNTK_BINARY $CNTK_ARGS || exit $?
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@
|
|||
#
|
||||
# Each test directory has a following components:
|
||||
# - testcases.yml - main test confuguration file, whcih defines all test cases
|
||||
# - run-test - (run-test) script
|
||||
# - run-test - (run-test) script
|
||||
# - baseline*.txt - baseline files whith a captured expected output of run-test script
|
||||
#
|
||||
# ----- testcases.yml format -------
|
||||
|
|
@ -52,10 +52,14 @@
|
|||
# ---- Baseline files ----
|
||||
# Order of searching baseline files, depends on the current mode for a given test:
|
||||
#
|
||||
# 1. baseline.<flavor>.<device>.txt
|
||||
# 2. baseline.<flavor>.txt
|
||||
# 3. baseline.<device>.txt
|
||||
# 4. baseline.txt
|
||||
# 1. baseline.<os>.<flavor>.<device>.txt
|
||||
# 2. baseline.<os>.<flavor>.txt
|
||||
# 3. baseline.<os>.<device>.txt
|
||||
# 4. baseline.<os>.txt
|
||||
# 5. baseline.<flavor>.<device>.txt
|
||||
# 6. baseline.<flavor>.txt
|
||||
# 7. baseline.<device>.txt
|
||||
# 8. baseline.txt
|
||||
# where <flavor> = { debug | release }
|
||||
# <device> = { cpu | gpu }
|
||||
#
|
||||
|
|
@ -79,6 +83,7 @@
|
|||
import sys, os, argparse, traceback, yaml, subprocess, random, re, time
|
||||
|
||||
thisDir = os.path.dirname(os.path.realpath(__file__))
|
||||
windows = os.getenv("OS")=="Windows_NT"
|
||||
|
||||
# This class encapsulates an instance of the test
|
||||
class Test:
|
||||
|
|
@ -169,6 +174,10 @@ class Test:
|
|||
os.environ["TEST_FLAVOR"] = flavor
|
||||
os.environ["TEST_DEVICE"] = device
|
||||
os.environ["TEST_BUILD_LOCATION"] = args.build_location
|
||||
if windows:
|
||||
os.environ["TEST_CNTK_BINARY"] = os.path.join(args.build_location, flavor, "cntk.exe")
|
||||
else:
|
||||
os.environ["TEST_CNTK_BINARY"] = os.path.join(args.build_location, flavor, "bin", "cntk")
|
||||
os.environ["TEST_DIR"] = self.testDir
|
||||
os.environ["TEST_DATA_DIR"] = self.dataDir
|
||||
os.environ["TEST_RUN_DIR"] = runDir
|
||||
|
|
@ -237,17 +246,22 @@ class Test:
|
|||
return result
|
||||
|
||||
# Finds a location of a baseline file by probing different names in the following order:
|
||||
# baseline.$os.$flavor.$device.txt
|
||||
# baseline.$os.$flavor.txt
|
||||
# baseline.$os.$device.txt
|
||||
# baseline.$os.txt
|
||||
# baseline.$flavor.$device.txt
|
||||
# baseline.$flavor.txt
|
||||
# baseline.$device.txt
|
||||
# baseline.txt
|
||||
def findBaselineFile(self, flavor, device):
|
||||
for f in ["." + flavor.lower(), ""]:
|
||||
for d in ["." + device.lower(), ""]:
|
||||
candidateName = "baseline" + f + d + ".txt";
|
||||
fullPath = os.path.join(self.testDir, candidateName)
|
||||
if os.path.isfile(fullPath):
|
||||
return fullPath
|
||||
for o in ["." + ("windows" if windows else "linux"), ""]:
|
||||
for f in ["." + flavor.lower(), ""]:
|
||||
for d in ["." + device.lower(), ""]:
|
||||
candidateName = "baseline" + o + f + d + ".txt"
|
||||
fullPath = os.path.join(self.testDir, candidateName)
|
||||
if os.path.isfile(fullPath):
|
||||
return fullPath
|
||||
return None
|
||||
|
||||
# This class encapsulates one testcase (in testcases.yml file)
|
||||
|
|
@ -521,13 +535,13 @@ runSubparser.add_argument("test", nargs="*",
|
|||
help="optional test name(s) to run, specified as Suite/TestName. "
|
||||
"Use list command to list available tests. "
|
||||
"If not specified then all tests will be run.")
|
||||
#TODO: port paths to Windows
|
||||
defaultBuildLocation=os.path.realpath(os.path.join(thisDir, "..", "bin"))
|
||||
defaultBuildLocation=os.path.realpath(os.path.join(thisDir, "..", "x64" if windows else "build"))
|
||||
|
||||
runSubparser.add_argument("-b", "--build-location", default=defaultBuildLocation, help="location of the CNTK build to run")
|
||||
runSubparser.add_argument("-d", "--device", help="cpu|gpu - run on a specific device")
|
||||
runSubparser.add_argument("-f", "--flavor", help="release|debug - run only a specific flavor")
|
||||
#TODO: port paths to Windows
|
||||
defaultRunDir=os.path.join("/tmp", "cntk-test-{0}.{1}".format(time.strftime("%Y%m%d%H%M%S"), random.randint(0,1000000)))
|
||||
tmpDir = os.getenv("TEMP") if windows else "/tmp"
|
||||
defaultRunDir=os.path.join(tmpDir, "cntk-test-{0}.{1}".format(time.strftime("%Y%m%d%H%M%S"), random.randint(0,1000000)))
|
||||
runSubparser.add_argument("-r", "--run-dir", default=defaultRunDir, help="directory where to store test output, default: a random dir within /tmp")
|
||||
runSubparser.add_argument("--update-baseline", action='store_true', help="update baseline file(s) instead of matching them")
|
||||
runSubparser.add_argument("-v", "--verbose", action='store_true', help="verbose output - dump all output of test script")
|
||||
|
|
|
|||
|
|
@ -0,0 +1,412 @@
|
|||
#!/bin/bash
|
||||
|
||||
configure=$0
|
||||
build_top=$PWD
|
||||
|
||||
have_cuda=no
|
||||
cuda_path=
|
||||
cuda_check=include/cuda.h
|
||||
enable_cuda=
|
||||
|
||||
have_acml=no
|
||||
acml_path=
|
||||
acml_check=include/acml.h
|
||||
|
||||
have_mkl=no
|
||||
mkl_path=
|
||||
mkl_check=mkl/include/mkl.h
|
||||
|
||||
have_kaldi=no
|
||||
kaldi_path=
|
||||
kaldi_check=src/kaldi.mk
|
||||
|
||||
have_buildtype=no
|
||||
buildtype=
|
||||
default_buildtype=release
|
||||
|
||||
have_gdk=no
|
||||
gdk_path=
|
||||
gdk_check=include/nvidia/gdk/nvml.h
|
||||
|
||||
mathlib=
|
||||
|
||||
# List from best to worst choice
|
||||
default_path_list="/usr /usr/local /opt /opt/local"
|
||||
|
||||
# List from best to worst choice
|
||||
default_acmls="acml5.3.1/ifort64"
|
||||
default_mkls=""
|
||||
|
||||
# NOTE: Will get compilation errors with cuda-6.0
|
||||
default_cudas="cuda-7.5 cuda-7.0 cuda-6.5"
|
||||
default_kaldis="kaldi-trunk"
|
||||
default_gdks="."
|
||||
|
||||
function default_paths ()
|
||||
{
|
||||
echo $build_top $HOME $default_path_list
|
||||
}
|
||||
|
||||
# $1 is directory
|
||||
# $2 is file that must be present
|
||||
function check_dir ()
|
||||
{
|
||||
if test -e $1/$2
|
||||
then
|
||||
echo yes
|
||||
else
|
||||
echo no
|
||||
fi
|
||||
}
|
||||
|
||||
# $1 is the list of tails to search, ordered by preference
|
||||
# $2 is some file that must exist in $1
|
||||
function find_dir ()
|
||||
{
|
||||
for tail in $1
|
||||
do
|
||||
for head in $(default_paths)
|
||||
do
|
||||
if test x$(check_dir "$head/$tail" $2) = xyes
|
||||
then
|
||||
echo $head/$tail
|
||||
return 0
|
||||
fi
|
||||
done
|
||||
done
|
||||
}
|
||||
|
||||
function find_acml ()
|
||||
{
|
||||
find_dir "$default_acmls" "$acml_check"
|
||||
}
|
||||
|
||||
function find_mkl ()
|
||||
{
|
||||
find_dir "$default_mkls" "$mkl_check"
|
||||
}
|
||||
|
||||
function find_cuda ()
|
||||
{
|
||||
find_dir "$default_cudas" "$cuda_check"
|
||||
}
|
||||
|
||||
function find_kaldi ()
|
||||
{
|
||||
find_dir "$default_kaldis" "$kaldi_check"
|
||||
}
|
||||
|
||||
function find_gdk ()
|
||||
{
|
||||
find_dir "$default_gdks" "$gdk_check"
|
||||
}
|
||||
|
||||
function is_hardlinked ()
|
||||
{
|
||||
r=no
|
||||
if test -e $1 && test -e $2
|
||||
then
|
||||
r=yes
|
||||
[ "`stat -c '%i' $1`" != "`stat -c '%i' $2`" ] && r=no
|
||||
fi
|
||||
echo $r
|
||||
}
|
||||
|
||||
function default_use_cuda ()
|
||||
{
|
||||
if test x$(find_cuda) = x || test x$(find_gdk) = x
|
||||
then
|
||||
echo no
|
||||
else
|
||||
echo yes
|
||||
fi
|
||||
}
|
||||
enable_cuda=$(default_use_cuda)
|
||||
|
||||
function show_default ()
|
||||
{
|
||||
if test x$1 = x
|
||||
then
|
||||
echo "(no default)"
|
||||
else
|
||||
echo "(default $1)"
|
||||
fi
|
||||
}
|
||||
|
||||
function show_help ()
|
||||
{
|
||||
echo "Usage: configure [options]"
|
||||
echo "Options:"
|
||||
echo " -h|--help this help"
|
||||
echo " --with-build-top=directory build directory $(show_default $build_top)"
|
||||
echo " --add directory add directory to library search path"
|
||||
echo " --cuda[=(yes|no)] use cuda GPU $(show_default $(default_use_cuda))"
|
||||
echo " --with-cuda[=directory] $(show_default $(find_cuda))"
|
||||
echo " --with-gdk[=directory] $(show_default $(find_gdk))"
|
||||
echo " --with-acml[=directory] $(show_default $(find_acml))"
|
||||
echo " --with-mkl[=directory] $(show_default $(find_mkl))"
|
||||
echo " --with-buildtype=(debug|release) $(show_default $default_buildtype)"
|
||||
echo " --with-kaldi[=directory] $(show_default $(find_kaldi))"
|
||||
echo "Libraries search path:"
|
||||
for head in $(default_paths)
|
||||
do
|
||||
echo " $head"
|
||||
done
|
||||
}
|
||||
|
||||
while [[ $# > 0 ]]
|
||||
do
|
||||
key="$1"
|
||||
case $key in
|
||||
*=?*) optarg=`expr "X$key" : '[^=]*=\(.*\)'` ;;
|
||||
*=) optarg= ;;
|
||||
*) optarg= ;;
|
||||
esac
|
||||
|
||||
case $key in
|
||||
-h|--help)
|
||||
show_help
|
||||
exit 1
|
||||
;;
|
||||
--with-build-top*)
|
||||
if test x$optarg != x
|
||||
then
|
||||
build_top=$optarg
|
||||
mkdir -p $build_top
|
||||
fi
|
||||
;;
|
||||
--add*)
|
||||
if test x$optarg = x
|
||||
then
|
||||
shift ; optarg=$1
|
||||
fi
|
||||
default_path_list="$optarg $default_path_list"
|
||||
;;
|
||||
--cuda)
|
||||
if test x$optarg = yes || test x$optarg = no
|
||||
then
|
||||
enable_cuda=$optarg
|
||||
else
|
||||
echo "Invalid value for --cuda"
|
||||
show_help
|
||||
exit
|
||||
fi
|
||||
;;
|
||||
|
||||
--with-cuda*)
|
||||
have_cuda=yes
|
||||
enable_cuda=yes
|
||||
if test x$optarg = x
|
||||
then
|
||||
cuda_path=$(find_cuda)
|
||||
if test x$cuda_path = x
|
||||
then
|
||||
echo "Cannot find cuda directory."
|
||||
echo "Please specify a value for --with-cuda"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if test $(check_dir $optarg $cuda_check) = yes
|
||||
then
|
||||
cuda_path=$optarg
|
||||
else
|
||||
echo "Invalid cuda directory $optarg"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
;;
|
||||
--with-gdk*)
|
||||
have_gdk=yes
|
||||
if test x$optarg = x
|
||||
then
|
||||
gdk_path=$(find_gdk)
|
||||
if test x$gdk_path = x
|
||||
then
|
||||
echo "Cannot find gdk directory."
|
||||
echo "Please specify a value for --with-gdk"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if test $(check_dir $optarg $gdk_check) = yes
|
||||
then
|
||||
gdk_path=$optarg
|
||||
else
|
||||
echo "Invalid gdk directory $optarg"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
;;
|
||||
--with-acml*)
|
||||
have_acml=yes
|
||||
mathlib=acml
|
||||
if test x$optarg = x
|
||||
then
|
||||
acml_path=$(find_acml)
|
||||
if test x$acml_path = x
|
||||
then
|
||||
echo "Cannot fine acml directory"
|
||||
echo "Please specify a value for --with-acml"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if test $(check_dir $optarg $acml_check) = yes
|
||||
then
|
||||
acml_path=$optarg
|
||||
else
|
||||
echo "Invalid acml directory $optarg"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
;;
|
||||
--with-mkl*)
|
||||
have_mkl=yes
|
||||
mathlib=mkl
|
||||
if test x$optarg = x
|
||||
then
|
||||
mkl_path=$(find_mkl)
|
||||
if test x$mkl_path = x
|
||||
then
|
||||
echo "Cannot fine mkl directory"
|
||||
echo "Please specify a value for --with-mkl"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if test $(check_dir $optarg $mkl_check) = yes
|
||||
then
|
||||
mkl_path=$optarg
|
||||
else
|
||||
echo "Invalid mkl directory $optarg"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
;;
|
||||
--with-buildtype*)
|
||||
have_buildtype=yes
|
||||
case $optarg in
|
||||
debug|release)
|
||||
buildtype=$optarg
|
||||
;;
|
||||
*)
|
||||
echo Invalid buildtype $optarg
|
||||
echo Must be debug or release
|
||||
exit 1
|
||||
esac
|
||||
;;
|
||||
--with-kaldi*)
|
||||
have_kaldi=yes
|
||||
if test x$optarg = x
|
||||
then
|
||||
kaldi_path=$(find_kaldi)
|
||||
if test x$kaldi_path = x
|
||||
then
|
||||
echo "Cannot find kaldi directory"
|
||||
echo "Please specify a value for --with-kaldi"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if test $(check_dir $optarg $kaldi_check)
|
||||
then
|
||||
kaldi_path=$optarg
|
||||
else
|
||||
echo "Invalid kaldi directory $optarg"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
;;
|
||||
*)
|
||||
echo Invalid option $key
|
||||
show_help
|
||||
exit 1
|
||||
esac
|
||||
shift
|
||||
done
|
||||
|
||||
if test x$buildtype = x
|
||||
then
|
||||
buildtype=$default_buildtype
|
||||
echo Defaulting to --with-buildtype=release
|
||||
fi
|
||||
|
||||
# If no math library was specified, search for acml and then mkl
|
||||
if test x$have_acml = xno && test x$have_mkl = xno
|
||||
then
|
||||
acml_path=$(find_acml)
|
||||
if test x$acml_path = x
|
||||
then
|
||||
mkl_path=$(find_mkl)
|
||||
if test x$mkl_path = x
|
||||
then
|
||||
echo "Cannot find a CPU math library."
|
||||
echo "Please specify --with-acml or --with-mkl with a path."
|
||||
exit 1
|
||||
else
|
||||
mathlib=mkl
|
||||
fi
|
||||
else
|
||||
mathlib=acml
|
||||
fi
|
||||
fi
|
||||
|
||||
# If no cuda library specified, search for one
|
||||
if test x$enable_cuda = xyes && test x$cuda_path = x
|
||||
then
|
||||
cuda_path=$(find_cuda)
|
||||
if test x$cuda_path = x ; then
|
||||
echo Cannot locate a cuda directory
|
||||
echo GPU will be disabled
|
||||
enable_cuda=no
|
||||
else
|
||||
echo Found cuda at $cuda_path
|
||||
fi
|
||||
fi
|
||||
|
||||
if test $enable_cuda = yes && test x$gdk_path = x
|
||||
then
|
||||
gdk_path=$(find_gdk)
|
||||
if test x$gdk_path = x ; then
|
||||
echo Cannot locate a gdk directory
|
||||
echo GPU will be disabled
|
||||
enable_cuda=no
|
||||
else
|
||||
echo Found gdk at $gdk_path
|
||||
fi
|
||||
fi
|
||||
|
||||
config=$build_top/Config.make
|
||||
echo Generating $config
|
||||
echo "#Configuration file for cntk" > $config
|
||||
echo BUILDTYPE=$buildtype >> $config
|
||||
echo MATHLIB=$mathlib >> $config
|
||||
case $mathlib in
|
||||
acml)
|
||||
echo ACML_PATH=$acml_path >> $config
|
||||
;;
|
||||
mkl)
|
||||
echo MKL_PATH=$mkl_path >> $config
|
||||
;;
|
||||
esac
|
||||
if test $enable_cuda = yes ; then
|
||||
echo CUDA_PATH=$cuda_path >> $config
|
||||
echo GDK_PATH=$gdk_path >> $config
|
||||
fi
|
||||
if test x$kaldi_path != x ; then
|
||||
echo KALDI_PATH=$kaldi_path >> $config
|
||||
fi
|
||||
|
||||
# If we are not in the configure directory, generate a trampoline Makefile
|
||||
makefile=$build_top/Makefile
|
||||
if test $(is_hardlinked "$configure" "$build_top/configure") = no
|
||||
then
|
||||
echo Generating $makefile
|
||||
realconf=`readlink -f $configure`
|
||||
dir=`dirname $realconf`
|
||||
echo "#Generate Makefile" > $makefile
|
||||
echo dir=$dir >> $makefile
|
||||
echo BUILD_TOP=$build_top >> $makefile
|
||||
echo >> $makefile
|
||||
echo all clean : >> $makefile
|
||||
printf '\t$(MAKE) -C $(dir) BUILD_TOP=$(BUILD_TOP) $@\n' >> $makefile
|
||||
fi
|
||||
echo run
|
||||
echo '>make -j all'
|
||||
echo to build
|
||||
Загрузка…
Ссылка в новой задаче