Debug

modeling/stoppers.py

@@ -122,6 +122,7 @@ class Stoppers:
         input_ids: torch.LongTensor,
     ) -> bool:
                 
+        print(f"[stop_sequence_stopper] Input ids: {input_ids}")
         data = [model.tokenizer.decode(x) for x in input_ids]
         # null_character = model.tokenizer.encode(chr(0))[0]
         if "completed" not in model.gen_state:

modeling/tokenizer.py

@@ -28,6 +28,7 @@ class GenericTokenizer:
         return ret.ids
 
     def decode(self, tokens: Union[int, List[int], torch.Tensor]) -> str:
+        print(f"[decode] Tokens: {tokens}")
         if isinstance(tokens, torch.Tensor):
             tokens = tokens.cpu().tolist()