Fix: encoder and decoder in tokenizer

Hi!

When evaluating the RWKV-v5-Eagle-7B-HF model, I found the errors as shown below. This is mostly caused by the tokenizer. In the code, the encoder and decoder are reversed, as discussed here https://huggingface.co/RWKV/v5-Eagle-7B-HF/discussions/9#65d4dc35f9cbfa798c4be4b3


This PR is raised to fix this issue.
Thanks,

tokenization_rwkv_world.py

@@ -106,11 +106,11 @@ class RWKVWorldTokenizer(PreTrainedTokenizer):
             assert isinstance(x, bytes)
             assert len(x) == int(l[l.rindex(" ") :])
             sorted += [x]
-            self.encoder[idx] = x
+            self.encoder[x] = idx
 
         self.decoder = {}
         for k, v in self.encoder.items():
-            self.decoder[v] = int(k)
+            self.decoder[v] = k
 
         self.trie = TRIE()
         for t, i in self.decoder.items():