Update handler.py

handler.py

@@ -153,6 +153,7 @@ class EndpointHandler():
           )
         
         output_strs = self.tokenizer.batch_decode(outputs.sequences[:, input_length:], skip_special_tokens=True)
-        output_strs = [output_str.replace("\n<human>:", "") for output_str in output_strs] 
+        output_strs = [output_str.replace("\n<human>:", "") for output_str in output_strs]
         
+        torch.cuda.empty_cache()
         return {"generated_text": output_strs}