Update handler.py to handle lists

handler.py

@@ -26,39 +26,49 @@ class EndpointHandler:
         self.hate_speech_encoder = joblib.load(os.path.join(model_path, 'hate_speech_encoder.pkl'))
 
     def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
-        # Preprocess input: extract the text from the request payload
-        text = data.get('inputs')
-        
-        # Tokenize the input text
-        inputs = self.tokenizer(text, return_tensors='pt', max_length=256, truncation=True, padding=True)
-        if 'token_type_ids' in inputs:
-            del inputs['token_type_ids']
-        inputs = {key: val.to(self.device) for key, val in inputs.items()}
-
-        # Run the input through the model
-        with torch.no_grad():
-            outputs = self.model(**inputs)
-            emotion_logits = outputs.get('emotion')
-            polarity_logits = outputs.get('polarity')
-            hate_speech_logits = outputs.get('hate_speech')
-
-        # Decode predictions from logits using argmax and the label encoders
-        emotion_preds = torch.argmax(emotion_logits, dim=1).cpu().numpy().tolist()  
-        polarity_preds = torch.argmax(polarity_logits, dim=1).cpu().numpy().tolist()  
-        hate_speech_preds = torch.argmax(hate_speech_logits, dim=1).cpu().numpy().tolist()  
-
-        # Inverse transform the predictions to get human-readable labels
-        decoded_emotions = self.emotion_encoder.inverse_transform(emotion_preds).tolist()
-        decoded_polarities = self.polarity_encoder.inverse_transform(polarity_preds).tolist()
-        decoded_hate_speech = self.hate_speech_encoder.inverse_transform(hate_speech_preds).tolist()
-
-        # Return the decoded results as a dictionary
-        return {
-            "emotions": decoded_emotions,
-            "polarities": decoded_polarities,
-            "hate_speech": decoded_hate_speech
+        # Extract the list of texts
+        texts = data.get('inputs', [])
+
+        # Batch processing for large inputs
+        batch_size = 32
+        results = {
+            "emotions": [],
+            "polarities": [],
+            "hate_speech": []
         }
 
+        for i in range(0, len(texts), batch_size):
+            batch_texts = texts[i:i+batch_size]
+            
+            # Tokenize the input text
+            inputs = self.tokenizer(text, return_tensors='pt', max_length=256, truncation=True, padding=True)
+            if 'token_type_ids' in inputs:
+                del inputs['token_type_ids']
+            inputs = {key: val.to(self.device) for key, val in inputs.items()}
+
+            # Run the input through the model
+            with torch.no_grad():
+                outputs = self.model(**inputs)
+                emotion_logits = outputs.get('emotion')
+                polarity_logits = outputs.get('polarity')
+                hate_speech_logits = outputs.get('hate_speech')
+    
+            # Decode predictions from logits using argmax and the label encoders
+            emotion_preds = torch.argmax(emotion_logits, dim=1).cpu().numpy().tolist()  
+            polarity_preds = torch.argmax(polarity_logits, dim=1).cpu().numpy().tolist()  
+            hate_speech_preds = torch.argmax(hate_speech_logits, dim=1).cpu().numpy().tolist()  
+    
+            # Inverse transform the predictions to get human-readable labels
+            decoded_emotions = self.emotion_encoder.inverse_transform(emotion_preds).tolist()
+            decoded_polarities = self.polarity_encoder.inverse_transform(polarity_preds).tolist()
+            decoded_hate_speech = self.hate_speech_encoder.inverse_transform(hate_speech_preds).tolist()
+
+            results["emotions"].extend(decoded_emotions)
+            results["polarities"].extend(decoded_polarities)
+            results["hate_speech"].extend(decoded_hate_speech)
+
+        return results
+
     def load_model(self, model_path):
         #Load model weights from the specified path
         self.load_state_dict(torch.load(model_path))