forked from microsoft/CNTK
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
153 lines (112 loc) · 4.97 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
== Dev branch ==
This branch contains some features that are not yet checked into main branch. To enlist this branch, run
git checkout origin/Dev
== To-do ==
Add descriptions to LSTMnode
Add descriptions to 0/1 mask segmentation in feature reader, delay node, and crossentropywithsoftmax node
Change criterion node to use the 0/1 mask, following example in crossentropywithsoftmax node
Add description of encoder-decoder simple network builder
Add description of time-reverse node, simple network builder and NDL builder for bi-directional models
== Author of the README ==
Kaisheng Yao
Microsoft Research
email: [email protected]
Wengong Jin,
Shanghai Jiao Tong University
email: [email protected]
Yu Zhang, Leo Liu
CSAIL, Massachusetts Institute of Technology
email: [email protected]
email: [email protected]
Guoguo Chen
CLSP, Johns Hopkins University
email: [email protected]
== Preeliminaries ==
To build the cpu version, you have to install intel MKL blas library or ACML library first. Note that ACML is free, where MKL may not be.
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 ==
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
For release version, just add BUILDTYPE=release to the make command line.
== Run ==
All executables are in bin/ directory:
cn.exe: 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}
== 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
The feature section is like:
writer=[
writerType=KaldiReader
readMethod=blockRandomize
frameMode=false
miniBatchMode=Partial
randomize=Auto
verbosity=1
ScaledLogLikelihood=[
dim=$labelDim$
Kaldicmd="ark:-" # will pipe to the Kaldi decoder latgen-faster-mapped
scpFile=$outputSCP$ # the file key of the features
]
]
== 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
The features section is different:
features=[
dim=
rx=
scpFile=
featureTransform=
]
rx is a text file which contains:
one Kaldi feature rxspecifier readable by RandomAccessBaseFloatMatrixReader.
'ark:' specifiers don't work; only 'scp:' specifiers work.
scpFile is a text file generated by running:
feat-to-len FEATURE_RXSPECIFIER_FROM_ABOVE ark,t:- > TEXT_FILE_NAME
scpFile should contain one line per utterance.
If you want to run with fewer utterances, just shorten this file.
(It will load the feature rxspecifier but ignore utterances not present in scpFile).
featureTransform is the name of a Kaldi feature transform file:
Kaldi feature transform files are used for stacking / applying transforms to features.
An empty string (if permitted by the config file reader?) or the special string: NO_FEATURE_TRANSFORM
says to ignore this option.
********** Labels **********
The labels section is also different.
labels=[
mlfFile=
labelDim=
labelMappingFile=
]
Only difference is mlfFile. mlfFile is a different format now. It is a text file which contains:
one Kaldi label rxspecifier readable by Kaldi's copy-post binary.