fix accuracy compute

tasks/audio.py

@@ -1,7 +1,7 @@
 from fastapi import APIRouter
 from datetime import datetime
 from datasets import load_dataset
-from sklearn.metrics import accuracy_score
+#from sklearn.metrics import accuracy_score
 import os
 import torch
 
@@ -48,7 +48,8 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     model = torch.compile(model)
     model.load_state_dict(torch.load('tasks/models/final-bf16.pth', weights_only=True))
     model.eval()
-    
+    num_correct = 0
+    num_samples = len(test_dataset)
     # Start tracking emissions
     tracker.start()
     tracker.start_task("inference")
@@ -59,14 +60,14 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     #--------------------------------------------------------------------------------------------   
     
     predictions = []
-    with torch.no_grad():#, torch.amp.autocast(device_type=device):
+    with torch.no_grad():
         for (X, y) in dataloader:
             X = X.to(device, dtype=torch.bfloat16)
             y = y.to(device, dtype=torch.bfloat16)
             
-            predictions.append(model(X))
-    predictions = torch.cat(predictions, dim=0)
-    
+            predictions = model(X)
+            num_correct += (y==predictions).sum()  # count correct predictions
+
     #--------------------------------------------------------------------------------------------
     # YOUR MODEL INFERENCE STOPS HERE
     #--------------------------------------------------------------------------------------------   
@@ -75,8 +76,7 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     emissions_data = tracker.stop_task()
     
     # Calculate accuracy
-    true_labels = test_dataset["label"]
-    accuracy = accuracy_score(true_labels, predictions)
+    accuracy = float(num_correct) / float(num_samples)
     
     # Prepare results dictionary
     results = {