CODEWITHSUNDEEP
pythonpytorchnvidiaopenai-whisper
mirzaahmergull
mirzaahmergull

Reputation: 21

CUDA/GPU not being utilized

I am trying to use pyannote.audio for diarization and then use Whisper for transcription.

I have two machines: one with RTX3050 Laptop GPU, another is my desktop with 2060 Super. When I am running my code on Laptop (RTX 3050) GPU is utilized 80-100% but when I am running the same code on my desktop it is not being utilized.

The following on both machines gives me True:

torch.cuda.is_available()

And I am even trying to enforce my GPU on both machine using this:

device = torch.device("cuda" if torch.cuda.is_available() else "cpu") // gives cuda
model  = whisper.load_model(model_size, device=device)
result = model.transcribe(audio=audiof, language=lang, word_timestamps=True)

Configuration for both is: Laptop :

  1. Nvidia cuda toolkit installed
  2. cuda version = 11.7
  3. torch installed version
torch                     2.0.1+cu117
torch-audiomentations     0.11.0
torch-pitch-shift         1.2.4
torchaudio                2.0.2+cu117
torchmetrics              0.11.4
torchvision               0.15.2+cu117
  1. Whisper and pyannote versions:
openai                    0.27.8
openai-whisper            20230314
pyannote.audio            2.1.1
pyannote.core             5.0.0
pyannote.database         5.0.1
pyannote.metrics          3.2.1
pyannote.pipeline         2.3

Desktop :

  1. Nvidia cuda toolkit installed
  2. cuda version = 11.8
  3. torch installed version
torch                     2.0.1+cu118
torch-audiomentations     0.11.0
torch-pitch-shift         1.2.4
torchaudio                2.0.2+cu118
torchmetrics              0.11.4
torchvision               0.15.2+cu118
  1. Whisper and pyannote versions:
openai                    0.27.8
openai-whisper            20230314
pyannote.audio            2.1.1
pyannote.core             5.0.0
pyannote.database         5.0.1
pyannote.metrics          3.2.1
pyannote.pipeline         2.3

Upvotes: 2

Views: 480

Answers (0)

Related Questions