Merge pull request #363 from xinghai-sun/default_value

Reset default value of batch_size, num_proc_data and fix an invalid url for DS2.

deep_speech_2/examples/aishell/run_test.sh

@@ -18,7 +18,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=300 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=1024 \

deep_speech_2/examples/aishell/run_test_golden.sh

@@ -27,7 +27,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=300 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=1024 \

deep_speech_2/examples/aishell/run_train.sh

@@ -9,7 +9,7 @@ python -u train.py \
 --batch_size=64 \
 --trainer_count=8 \
 --num_passes=50 \
---num_proc_data=8 \
+--num_proc_data=16 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=1024 \

deep_speech_2/examples/librispeech/run_test.sh

@@ -18,7 +18,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=500 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=2048 \

deep_speech_2/examples/librispeech/run_test_golden.sh

@@ -27,7 +27,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=500 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=2048 \

deep_speech_2/examples/librispeech/run_train.sh

@@ -6,10 +6,10 @@ cd ../.. > /dev/null
 # if you wish to resume from an exists model, uncomment --init_model_path
 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
 python -u train.py \
---batch_size=512 \
+--batch_size=160 \
 --trainer_count=8 \
 --num_passes=50 \
---num_proc_data=8 \
+--num_proc_data=16 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=2048 \

deep_speech_2/examples/librispeech/run_tune.sh

@@ -6,7 +6,7 @@ cd ../.. > /dev/null
 CUDA_VISIBLE_DEVICES=0,1,2,3 \
 python -u tools/tune.py \
 --num_batches=-1 \
---batch_size=256 \
+--batch_size=128 \
 --trainer_count=8 \
 --beam_size=500 \
 --num_proc_bsearch=12 \

deep_speech_2/examples/tiny/run_test.sh

@@ -18,7 +18,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=500 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=2048 \

deep_speech_2/examples/tiny/run_test_golden.sh

@@ -27,7 +27,7 @@ python -u test.py \
 --trainer_count=8 \
 --beam_size=500 \
 --num_proc_bsearch=8 \
---num_proc_data=4 \
+--num_proc_data=8 \
 --num_conv_layers=2 \
 --num_rnn_layers=3 \
 --rnn_layer_size=2048 \

deep_speech_2/infer.py

@@ -17,7 +17,7 @@ add_arg = functools.partial(add_arguments, argparser=parser)
 add_arg('num_samples',      int,    10,     "# of samples to infer.")
 add_arg('trainer_count',    int,    8,      "# of Trainers (CPUs or GPUs).")
 add_arg('beam_size',        int,    500,    "Beam search width.")
-add_arg('num_proc_bsearch', int,    12,     "# of CPUs for beam search.")
+add_arg('num_proc_bsearch', int,    8,      "# of CPUs for beam search.")
 add_arg('num_conv_layers',  int,    2,      "# of convolution layers.")
 add_arg('num_rnn_layers',   int,    3,      "# of recurrent layers.")
 add_arg('rnn_layer_size',   int,    2048,   "# of recurrent cells per layer.")

deep_speech_2/models/librispeech/download_model.sh

@@ -2,7 +2,7 @@
 
 . ../../utils/utility.sh
 
-URL='http://cloud.dlnel.org/filepub/?uuid=8e3cf742-2ff3-41ce-a49d-f6158cc06a23'
+URL='http://cloud.dlnel.org/filepub/?uuid=6020a634-5399-4423-b021-c5ed32680fff'
 MD5=2ef08f8b608a7c555592161fc14d81a6
 TARGET=./librispeech_model.tar.gz
 

deep_speech_2/test.py

@@ -17,8 +17,8 @@ add_arg = functools.partial(add_arguments, argparser=parser)
 add_arg('batch_size',       int,    128,    "Minibatch size.")
 add_arg('trainer_count',    int,    8,      "# of Trainers (CPUs or GPUs).")
 add_arg('beam_size',        int,    500,    "Beam search width.")
-add_arg('num_proc_bsearch', int,    12,     "# of CPUs for beam search.")
-add_arg('num_proc_data',    int,    4,      "# of CPUs for data preprocessing.")
+add_arg('num_proc_bsearch', int,    8,      "# of CPUs for beam search.")
+add_arg('num_proc_data',    int,    8,      "# of CPUs for data preprocessing.")
 add_arg('num_conv_layers',  int,    2,      "# of convolution layers.")
 add_arg('num_rnn_layers',   int,    3,      "# of recurrent layers.")
 add_arg('rnn_layer_size',   int,    2048,   "# of recurrent cells per layer.")

deep_speech_2/tools/tune.py

@@ -27,7 +27,8 @@ add_arg('num_batches',      int,    -1,     "# of batches tuning on. "
 add_arg('batch_size',       int,    256,    "# of samples per batch.")
 add_arg('trainer_count',    int,    8,      "# of Trainers (CPUs or GPUs).")
 add_arg('beam_size',        int,    500,    "Beam search width.")
-add_arg('num_proc_bsearch', int,    12,     "# of CPUs for beam search.")
+add_arg('num_proc_bsearch', int,    8,     "# of CPUs for beam search.")
+add_arg('num_proc_data',    int,    8,      "# of CPUs for data preprocessing.")
 add_arg('num_conv_layers',  int,    2,      "# of convolution layers.")
 add_arg('num_rnn_layers',   int,    3,      "# of recurrent layers.")
 add_arg('rnn_layer_size',   int,    2048,   "# of recurrent cells per layer.")
@@ -86,7 +87,7 @@ def tune():
         mean_std_filepath=args.mean_std_path,
         augmentation_config='{}',
         specgram_type=args.specgram_type,
-        num_threads=1)
+        num_threads=args.num_proc_data)
 
     audio_data = paddle.layer.data(
         name="audio_spectrogram",

deep_speech_2/train.py

@@ -16,7 +16,7 @@ add_arg = functools.partial(add_arguments, argparser=parser)
 add_arg('batch_size',       int,    256,    "Minibatch size.")
 add_arg('trainer_count',    int,    8,      "# of Trainers (CPUs or GPUs).")
 add_arg('num_passes',       int,    200,    "# of training epochs.")
-add_arg('num_proc_data',    int,    8,      "# of CPUs for data preprocessing.")
+add_arg('num_proc_data',    int,    16,     "# of CPUs for data preprocessing.")
 add_arg('num_conv_layers',  int,    2,      "# of convolution layers.")
 add_arg('num_rnn_layers',   int,    3,      "# of recurrent layers.")
 add_arg('rnn_layer_size',   int,    2048,   "# of recurrent cells per layer.")