RuntimeError: CreateModel failed with error code 8195

shubhamc247 · November 19, 2019, 6:13am

I trained my own data , it train very well.
but when i predict by using following command
(deepspeech-venv) shubh@shubh-Lenovo-Flex-2-14:~/Downloads/DeepSpeech$ deepspeech models/output_graph.pbmm models/alphabet.txt

it get 8195 error.

TensorFlow: v1.13.1-10-g3e0cc53
DeepSpeech: v0.5.1-0-g4b29b78
Warning: reading entire model file into memory. Transform model file into an mmapped graph to reduce heap usage.
2019-11-19 11:23:13.351350: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-11-19 11:23:13.497345: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “UnwrapDatasetVariant” device_type: “CPU”’) for unknown op: UnwrapDatasetVariant
2019-11-19 11:23:13.497400: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “WrapDatasetVariant” device_type: “GPU” host_memory_arg: “input_handle” host_memory_arg: “output_handle”’) for unknown op: WrapDatasetVariant
2019-11-19 11:23:13.497411: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “WrapDatasetVariant” device_type: “CPU”’) for unknown op: WrapDatasetVariant
2019-11-19 11:23:13.497533: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “UnwrapDatasetVariant” device_type: “GPU” host_memory_arg: “input_handle” host_memory_arg: “output_handle”’) for unknown op: UnwrapDatasetVariant
Specified model file version (0) is incompatible with minimum version supported by this client (1). See https://github.com/mozilla/DeepSpeech/#model-compatibility for more information
Traceback (most recent call last):
File “/home/shubh/tmp/deepspeech-venv/bin/deepspeech”, line 8, in
sys.exit(main())
File “/home/shubh/tmp/deepspeech-venv/lib/python3.6/site-packages/deepspeech/client.py”, line 88, in main
ds = Model(args.model, N_FEATURES, N_CONTEXT, args.alphabet, BEAM_WIDTH)
File “/home/shubh/tmp/deepspeech-venv/lib/python3.6/site-packages/deepspeech/init.py”, line 23, in init
raise RuntimeError(“CreateModel failed with error code {}”.format(status))
RuntimeError: CreateModel failed with error code 8195

i used:
Tensorflow 1.14.0 but when i used deepspeech --v it shows 1.13.0
and used 0.6.0 binary files
i was unable to download convert graphdef _ 1.13 version so i downloaded 1.14 convert graphdef and then run .

please help

thanx in advance

lissyx · November 19, 2019, 8:50am

It looks like you already figured it out yourself, you mixed versions, it’s not supported. Just ensure you install latest v0.6.0a15 binary. You may need to force pip with --upgrade deepspeech==0.6.0a15

shubhamc247 · November 20, 2019, 8:57am

Thanx @lissyx

if i am installed deepspeech==0.6.0a15 so Tensorflow 1.14.0 and 0.6.0 binary file supported on deepspeech?
or i need to install latest version of binary and tensorflow?

shubhamc247 · November 20, 2019, 9:03am

if i write command pip3 list it shows ds-ctcdecoder 0.6.0a15 so it can ok with upgraded version of deepspeech?

lissyx · November 20, 2019, 9:07am

You don’t need to care about TensorFlow at all.

lissyx · November 20, 2019, 9:08am

Yes, ds_ctcdecoder should match the DeepSpeech version as well.

shubhamc247 · November 22, 2019, 9:11am

hello,
I trained data and then create .pbmm file but when i predict audio file by following command
: deepspeech --model ./export/output_graph.pb --alphabet ./export/alphabet.txt --lm ./export/lm.binary --trie ./export/trie --audio ./export/40196.wav

it predict only ‘i’ ?
please tell me where i wrong?

deepspeech --model ./export/output_graph.pb --alphabet ./export/alphabet.txt --lm ./export/lm.binary --trie ./export/trie --audio ./export/40196.wav
Loading model from file ./export/output_graph.pb
TensorFlow: v1.13.1-10-g3e0cc53
DeepSpeech: v0.5.1-0-g4b29b78
Warning: reading entire model file into memory. Transform model file into an mmapped graph to reduce heap usage.
2019-11-22 14:34:58.722356: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-11-22 14:35:00.561618: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “UnwrapDatasetVariant” device_type: “CPU”’) for unknown op: UnwrapDatasetVariant
2019-11-22 14:35:00.561687: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “WrapDatasetVariant” device_type: “GPU” host_memory_arg: “input_handle” host_memory_arg: “output_handle”’) for unknown op: WrapDatasetVariant
2019-11-22 14:35:00.561706: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “WrapDatasetVariant” device_type: “CPU”’) for unknown op: WrapDatasetVariant
2019-11-22 14:35:00.561874: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel (‘op: “UnwrapDatasetVariant” device_type: “GPU” host_memory_arg: “input_handle” host_memory_arg: “output_handle”’) for unknown op: UnwrapDatasetVariant
Loaded model in 1.84s.
Loading language model from files ./export/lm.binary ./export/trie
Loaded language model in 1.01e+02s.
Running inference.
2019-11-22 14:36:42.269189: W tensorflow/core/framework/allocator.cc:124] Allocation of 16777216 exceeds 10% of system memory.
2019-11-22 14:36:42.283837: W tensorflow/core/framework/allocator.cc:124] Allocation of 16777216 exceeds 10% of system memory.
2019-11-22 14:36:42.305110: W tensorflow/core/framework/allocator.cc:124] Allocation of 134217728 exceeds 10% of system memory.
2019-11-22 14:36:42.485505: W tensorflow/core/framework/allocator.cc:124] Allocation of 16777216 exceeds 10% of system memory.
2019-11-22 14:36:42.497546: W tensorflow/core/framework/allocator.cc:124] Allocation of 16777216 exceeds 10% of system memory.
i
Inference took 18.559s for 6.976s audio file.

lissyx · November 22, 2019, 12:27pm

We can’t help you without more informations on the training you performed.

shubhamc247 · November 23, 2019, 5:38am

ok
i used deepspeech 5.0 & ds-ctcdecoder 5.0
and trained model on 910 (.wav) file using following command:
python3 DeepSpeech.py --checkpoint_dir /cpoint/ --train_batch_size 10 --test_batch_size 10 --dev_batch_size 10 --epochs 2 --train_files clips/TRAIN/train.csv --dev_files clips/DEV/dev.csv --test_files clips/TEST/test.csv --learning_rate 0.01 --export_dir export/
is this information required
?

lissyx · November 23, 2019, 9:16am

So basically, small dataset, your model was train on 2 epochs, it learnt mostly nothing.

shubhamc247 · November 23, 2019, 9:42am

ok…i will train on 5 or more epochs.
@lissyx thanx

lissyx · November 23, 2019, 10:46am

You would need much more than that, and smaller geometry as well, given your dataset.

shubhamc247 · November 25, 2019, 1:44pm

thanx for sharing info @lissyx

if i am trained on 700 .wav file and each file has 10 second , so total 116 min data. when i used all default parameter then it was taked more than 2 day, Can you provide some hints about the total training time ?
system information :
8gb ram
i5 intel processor
and used cpu.

if i am using GPU instead of cpu then how much it will affected for training?
Thanx in Advance

lissyx · November 25, 2019, 2:00pm

This amount of data is really really really low. I don’t think you can get anything out of just that.

That depends on the GPU you choose, but anything decent enough should take care of that amount of data in a matter of minutes.

shubhamc247 · November 25, 2019, 2:15pm

ok, can u please tell me that how much (minimum) size of data required for training?

lissyx · November 25, 2019, 2:20pm

It’s already documented and it depends on languages, but you can go with thousands of hours of audio.