Exception: FastGRNNCUDA is supported only on GPU devices.

[mehreenjabeen]

Getting error "Exception: FastGRNNCUDA is supported only on GPU devices." upon running kws-demo.py

[ShikharJ]

@mehreenjabeen I'm assuming your system has a GPU setup? Can you provide the sample command you ran and the full output?

[mehreenjabeen]

i ran this command: "python kws_demo.py --config_path <path_to_config.json> --model_path <path_to_model.pt> --mean_path <path_to_mean.npy> --std_path <path_to_std.npy>"
However, I had trained the model with architecture argument set as "-architecture FastGRNNCUDA". Now I am retraining with architecture parameter set to FastGRNN. Hope that resolves the issue

[mehreenjabeen]

the above thing didnt work(i.e chaging architecture from FastGRNNCUDA to FastGRNN )
So command i ran "python3 kws-demo.py --config_path /home/mehreen/Work/EdgeML/EdgeML/config.json --model_path /home/mehreen/Work/EdgeML/EdgeML/examples/pytorch/FastCells/KWS-training/model/FastGRNN128KeywordSpotter.pt --mean_path $MODEL_DIR/mean.npy --std_path $MODEL_DIR/std.npy"

Output:
sigmoid tanh
Traceback (most recent call last):
File "kws-demo.py", line 202, in
fastgrnn = create_model(config.model, num_filt, 35)
File "/home/mehreen/Work/EdgeML/EdgeML/examples/pytorch/FastCells/KWS-training/train_classifier.py", line 420, in create_model
model_config.update_nonlinearity, num_keywords)
File "/home/mehreen/.local/lib/python3.6/site-packages/edgeml_pytorch/trainer/fastmodel.py", line 59, in init
for l in range(self.num_layers)])
File "/home/mehreen/.local/lib/python3.6/site-packages/edgeml_pytorch/trainer/fastmodel.py", line 59, in
for l in range(self.num_layers)])
File "/home/mehreen/.local/lib/python3.6/site-packages/edgeml_pytorch/graph/rnn.py", line 1121, in init
raise Exception('FastGRNNCUDA is supported only on GPU devices.')
Exception: FastGRNNCUDA is supported only on GPU devices.

[ShikharJ]

@MJ10 Can you please take a look at this?

[ShikharJ]

@MJ10 Ping

[mehreenjabeen]

I am closing this issue as I got to know what the issue was. In config.json the model architecture was set to FastGRNNCuda because of which it was throwing the error. I changed it to Fastgrnn and it worked!