Update tasks/audio.py

tasks/audio.py

@@ -4,6 +4,7 @@ from datasets import load_dataset
 from sklearn.metrics import accuracy_score
 import random
 import os
+import torch
 
 from .utils.evaluation import AudioEvaluationRequest
 from .utils.emissions import tracker, clean_emissions_data, get_space_info
@@ -54,7 +55,34 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     #--------------------------------------------------------------------------------------------   
     
     # Make random predictions (placeholder for actual model inference)
+    def preprocess_audio(example):
+    """Convert dataset into tensors."""
+        waveform = torch.tensor(example["audio"]["array"], dtype=torch.float32).unsqueeze(0)  # Add batch dim
+        label = torch.tensor(example["label"], dtype=torch.long)  # Ensure labels are `int64`
+    return waveform, label
+
+    model_path = "quantized_teacher_m5_static.pth"
+    model, device = load_model(model_path)
+
+
+    train_test = train_test.map(preprocess_audio)
+    test_dataset = train_test.map(preprocess_audio)
+
+    test_loader = DataLoader(test_dataset, batch_size=32, shuffle=True)
+
+    
     true_labels = test_dataset["label"]
+
+    with torch.no_grad():
+    for waveforms, labels in test_loader:
+        waveforms, labels = waveforms.to(device), labels.to(device)
+        
+        # Run Model
+        outputs = model(waveforms)
+        predicted_label = torch.argmax(F.softmax(outputs, dim=1), dim=1)
+
+        true_labels.extend(labels.cpu().numpy())
+        predicted_labels.extend(predicted_label.cpu().numpy())
     predictions = [random.randint(0, 1) for _ in range(len(true_labels))]
     
     #--------------------------------------------------------------------------------------------