search

mravanelli / SincNet

SincNet is a neural architecture for efficiently processing raw audio samples.
MIT License
1.11k stars 260 forks source link

CUDA error: device-side assert triggered #110

Open Raviteja-banda opened 1 year ago

Raviteja-banda commented 1 year ago

I have been trying to run the speaker_id.py file(for a customised librispeech dataset) but i was stuck at this error from the past few days. I have even checked if there's any mismatch between the number of labels and the output neurons but they are fine. I've changed the config file as below: class_lay=251 batch_size=8 N_batches=251 (Since total number of train files(excluding validation data) = 2008)

[windowing] fs=16000 cw_len=200 cw_shift=10

[cnn] cnn_N_filt=80,60,60 cnn_len_filt=251,5,5 cnn_max_pool_len=3,3,3 cnn_use_laynorm_inp=True cnn_use_batchnorm_inp=False cnn_use_laynorm=True,True,True cnn_use_batchnorm=False,False,False cnn_act=leaky_relu,leaky_relu,leaky_relu cnn_drop=0.0,0.0,0.0

[dnn] fc_lay=2048,2048,2048 fc_drop=0.0,0.0,0.0 fc_use_laynorm_inp=True fc_use_batchnorm_inp=False fc_use_batchnorm=True,True,True fc_use_laynorm=False,False,False fc_act=leaky_relu,leaky_relu,leaky_relu

[class] class_lay=251 class_drop=0.0 class_use_laynorm_inp=False class_use_batchnorm_inp=False class_use_batchnorm=False class_use_laynorm=False class_act=softmax

I have also changed the data paths in the config file.

Below is my stack trace of the error:

C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [0,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [1,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [2,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [3,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [4,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [5,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [6,0,0] Assertion t >= 0 && t < n_classes failed. C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\Loss.cu:242: block: [0,0,0], thread: [7,0,0] Assertion t >= 0 && t < n_classes failed. Traceback (most recent call last): File "D:\SincNet-librispeech\speaker_id.py", line 245, in loss.backward() File "C:\ProgramData\Anaconda3\lib\site-packages\torch_tensor.py", line 487, in backward torch.autograd.backward( File "C:\ProgramData\Anaconda3\lib\site-packages\torch\autograd__init__.py", line 197, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass RuntimeError: CUDA error: device-side assert triggered

Any possible solutions?