How to run a trained model?

I trained a model for new language using transfer learning. It saved checkpoints in a folder. How do I test/run this trained model?

Have you read the documentation ?

yes I cant find this part.
I trained it using the documentation

Have you missed https://deepspeech.readthedocs.io/en/v0.7.3/USING.html ?

1 Like

I dont see a command to actually send in a new audio and it predicts the transcrtiption using my new checkpoints

I am sorry, can you please explain what am I supposed to see?

Export your checkpoint to pbmm ?

thats my question exactly. I dont have a .pbmm file. How do I make that from checkpoints folder?

Read the doc and use the export feature?

https://deepspeech.readthedocs.io/en/v0.7.3/TRAINING.html?highlight=export#exporting-a-model-for-inference

1 Like

I am sorry to bother you again. I added the --export_dir flag while training but no model was saved (pbmm)

I can’t help you without more context.

!python3 deepspeech/DeepSpeech.py \

--drop_source_layers 1 \

--alphabet_config_path using_data/my_new_alphabets.txt \

--save_checkpoint_dir output_checkpoints \

--load_checkpoint_dir output_checkpoints \

--train_files   using_data/train.csv \

--test_files  using_data/test.csv \

--export_dir output_checkpoints/model \

--scorer_path deepspeech/deepspeech-0.7.3-models.scorer

I ran this. “output_checkpoints/model” path stays empty

Please share the logs as well.

Logs? you mean the print statements? or ?

The stderr / stdout output, yes …

Have you tried just loading your new checkpoint and export it?

python3 deepspeech/DeepSpeech.py \
--checkpoint_dir output_checkpoints \
--export_dir output_checkpoints/model

swig/python detected a memory leak of type ‘Alphabet *’, no destructor found.
I Exporting the model…
Traceback (most recent call last):
File “deepspeech/DeepSpeech.py”, line 15, in
ds_train.run_script()
File “/content/gdrive/My Drive/NLP/Transcription Project/deepspeech/training/deepspeech_training/train.py”, line 943, in run_script
absl.app.run(main)
File “/usr/local/lib/python3.6/dist-packages/absl/app.py”, line 299, in run
_run_main(main, args)
File “/usr/local/lib/python3.6/dist-packages/absl/app.py”, line 250, in _run_main
sys.exit(main(argv))
File “/content/gdrive/My Drive/NLP/Transcription Project/deepspeech/training/deepspeech_training/train.py”, line 923, in main
export()
File “/content/gdrive/My Drive/NLP/Transcription Project/deepspeech/training/deepspeech_training/train.py”, line 754, in export
graph_version = int(file_relative_read(‘GRAPH_VERSION’).strip())
ValueError: invalid literal for int() with base 10: ‘…/…/GRAPH_VERSION’

This is a collab problem. Can you tell me what the graph version is supposed to be? I’ll hard code it

Please search in the forum, it’s already explained at a lot of places:

  • Google Colab seems to break symlinks
  • You should avoid spaces in directories.