Upload app.py with huggingface_hub

app.py

@@ -2,6 +2,7 @@ import json
 import os
 
 import gradio as gr
+import spaces
 import torch
 import torchaudio
 from huggingface_hub import hf_hub_download
@@ -23,10 +24,14 @@ UNKNOWN_ID = next(i for i, label in id2label.items() if label == "Unknown")
 LETTERS = sorted(label for label in id2label.values() if label != "Unknown")
 
 
+@spaces.GPU
 def predict(audio, target_letter):
     if audio is None:
         return "Record or upload audio first.", ""
 
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    model.to(device)
+
     sample_rate, waveform = audio
     waveform = torch.tensor(waveform, dtype=torch.float32)
     if waveform.ndim > 1:
@@ -35,6 +40,7 @@ def predict(audio, target_letter):
         waveform = torchaudio.functional.resample(waveform, sample_rate, SAMPLE_RATE)
 
     inputs = feature_extractor(waveform.numpy(), sampling_rate=SAMPLE_RATE, return_tensors="pt")
+    inputs = {k: v.to(device) for k, v in inputs.items()}
 
     with torch.no_grad():
         logits = model(**inputs).logits[0]