Update inferencer.py

inferencer.py

@@ -205,8 +205,8 @@ class InterleaveInferencer:
             **generation_input,
         ):
             output = self.tokenizer.decode(unpacked_latent)
-            # output = output.split('<|im_end|>')[0].split('<|im_start|>')[1]
-            yield output
+            if output != "<|im_end|>":
+                yield output
         
     @torch.no_grad()
     def interleave_inference(