Create handler.py

handler.py

@@ -0,0 +1,25 @@
+from transformers import AutoModel, AutoFeatureExtractor
+import torch
+
+class EndpointHandler:
+    def __init__(self, model_dir):
+        # This path (model_dir) is where HF automatically places the files
+        self.model = AutoModel.from_pretrained(model_dir, trust_remote_code=True)
+        self.feature_extractor = AutoFeatureExtractor.from_pretrained(model_dir, trust_remote_code=True)
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.model.to(self.device)
+        self.model.eval()
+
+    def __call__(self, data):
+        # 'data' is the dictionary from the POST request body
+        inputs = data.get("inputs")
+        
+        # Preprocess
+        processed = self.feature_extractor(inputs, return_tensors="pt", sampling_rate=16000).to(self.device)
+        
+        # Inference
+        with torch.no_grad():
+            outputs = self.model(**processed)
+            
+        # Return result
+        return outputs.last_hidden_state.mean(dim=1).cpu().numpy().tolist()