Created by: mtbadakhshan

Hello, I want to create a ASR system to recognise Persian speech. I can train a model from scratch, but I want to resumes my training from a pre-trained model (such as LibriSpeech Model or BaiduEN8k Model). so I set the value of "--init_model_path" argument to the directory of LibriSpeech Model parameters (models/librispeech/params.tar.gz). but I got the following output:

-----------  Configuration Arguments -----------
augment_conf_path: conf/augmentation.config
batch_size: 16
dev_manifest: /home/DeepSpeech2-PaddlePaddle/train-persian/manifest.persian
init_model_path: models/librispeech/params.tar.gz
is_local: 1
learning_rate: 1e-05
max_duration: 27.0
mean_std_path: /home/DeepSpeech2-PaddlePaddle/train-persian/mean_std.npz
min_duration: 0.0
num_conv_layers: 2
num_iter_print: 100
num_passes: 20
num_proc_data: 1
num_rnn_layers: 3
output_model_dir: /home/DeepSpeech2-PaddlePaddle/train-persian/checkpoints/libri-based
rnn_layer_size: 2048
share_rnn_weights: 1
shuffle_method: batch_shuffle_clipped
specgram_type: linear
test_off: 0
train_manifest: /home/DeepSpeech2-PaddlePaddle/train-persian/manifest.persian
trainer_count: 1
use_gpu: 1
use_gru: 0
use_sortagrad: 1
vocab_path: /home/DeepSpeech2-PaddlePaddle/train-persian/vocab.txt
------------------------------------------------
I0408 13:50:13.156404  6138 Util.cpp:166] commandline:  --use_gpu=1 --rnn_use_batch=True --log_clipping=True --trainer_count=1 
[INFO 2019-04-08 13:50:13,723 layers.py:2606] output for __conv_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2019-04-08 13:50:13,724 layers.py:3133] output for __batch_norm_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2019-04-08 13:50:13,724 layers.py:7224] output for __scale_sub_region_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2019-04-08 13:50:13,725 layers.py:2606] output for __conv_1__: c = 32, h = 41, w = 54, size = 70848
[INFO 2019-04-08 13:50:13,725 layers.py:3133] output for __batch_norm_1__: c = 32, h = 41, w = 54, size = 70848
[INFO 2019-04-08 13:50:13,726 layers.py:7224] output for __scale_sub_region_1__: c = 32, h = 41, w = 54, size = 70848
I0408 13:50:16.769909  6138 GradientMachine.cpp:94] Initing parameters..
I0408 13:50:18.651311  6138 GradientMachine.cpp:101] Init parameters done.
F0408 13:50:22.542716  6138 TensorApply.h:126] Check failed: lhs_.getWidth() == rhs_.getWidth() (118784 vs. 139264) 
*** Check failure stack trace: ***
    @     0x7feb1472b04d  google::LogMessage::Fail()
    @     0x7feb1472d398  google::LogMessage::SendToLog()
    @     0x7feb1472ab5b  google::LogMessage::Flush()
    @     0x7feb1472e26e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7feb148969c8  paddle::adamApply()
    @     0x7feb1488f8c9  paddle::AdamParameterOptimizer::update()
    @     0x7feb1488fcf2  paddle::OptimizerWithGradientClipping::update()
    @     0x7feb14871e7e  paddle::SgdThreadUpdater::updateImpl()
    @     0x7feb14700731  ParameterUpdater::update()
    @     0x7feb141b3e87  _wrap_ParameterUpdater_update
    @           0x4c468a  PyEval_EvalFrameEx
    @           0x4c9d8f  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca099  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca099  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca8d1  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca8d1  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4c2509  PyEval_EvalCode
    @           0x4f1def  (unknown)
    @           0x4ec652  PyRun_FileExFlags
    @           0x4eae31  PyRun_SimpleFileExFlags
    @           0x49e14a  Py_Main
    @     0x7feb308f2830  __libc_start_main
    @           0x49d9d9  _start
    @              (nil)  (unknown)
Aborted (core dumped)
Fail in training!

the error Individually:

F0408 13:50:22.542716  6138 TensorApply.h:126] Check failed: lhs_.getWidth() == rhs_.getWidth() (118784 vs. 139264) 

P.S. It is necessary to mention that if I continue my training from checkpoints that they were created while I was training from scratch, I will not get any errors. P.P.S I am pretty sure that using different vocabulary causes this problem, but I have no idea how to fix it.