SeanNaren / deepspeech.torch

Speech Recognition using DeepSpeech2 network and the CTC activation function.
MIT License
259 stars 73 forks source link

running pretrained net on CPU only #83

Open saurabhvyas opened 7 years ago

saurabhvyas commented 7 years ago

I have looked at the documentation, but it isnt clear how one can run a pretrained model on CPU only system ( for any dataset AN4 or librispeech ) , I have verified it works well on GPU system ? Also do you have any slack group / gitter / google group for deepspeech.torch for discussions ?

mtanana commented 7 years ago

If it was a pre-trained GPU model, you'd just have to de-cuda the model on a platform with a GPU (otherwise it won't load) and then re-save it...

it's a couple of lines of code though

--load cuda model then:

model:float()

--then save it

saurabhvyas commented 7 years ago

I am not able to understand it , I am new to torch/lua and deeplearning too , Are the commands you mentioned above to be used in command line as arguments , if so with which file ? Or is that a psudocode that needs to be converted in real lua code , I would appreciate if someone can put a simple example and update the documentation