Will training over the pre-existing checkpoint on new data, give a model which will be trained on both the old as well new data?

Hi,

I know it’s kind of kiddish to ask, but I want to make sure whether training on existing Deep Speech trained checkpoint with new datasets give me a .pb model that will contain the trainable properties of the pre-trained along with the new trained parameters?

The reason why I am asking is because we are only passing the pre trained checkpoints dir, but we are not passing the pre trained model dir (that is deep speech’s pretrained model .pb or .pbmm).

Please find my following code below :

python DeepSpeech.py --train_files ./20140421/scripts/Ib/clips/train.csv --dev_files ./20140421/scripts/Ib/clips/dev.csv --test_files ./20140421/scripts/Ib/clips/test.csv --beam_width 500 --checkpoint_dir ~/DeepSpeech/DeepSpeech/20140421/pretrained_checkpoints/deepspeech-0.6.1-checkpoint/ --load_cudnn true --epochs 3 --export_dir ~/DeepSpeech/DeepSpeech/20140421/scripts/Ib/models_dir/ --export_author_id ritish_20140421 -export_file_name epoch_3 --export_model_name model_3_epochs --learning_rate 0.005 --lm_alpha 0.4 --lm_beta 1.5 --train_batch_size 5

Have you read the documentation? The .pb or .pbmm are exported from the checkpoint, so they contain the same informations …

1 Like