there is a better way to do that! (#950)

examples/whisper.py

@@ -106,6 +106,7 @@ class Whisper:
 
 # TODO: this is tragic. remove this
 import functools
+import itertools
 import torch
 import torchaudio
 import librosa
@@ -158,10 +159,8 @@ def get_encoding(n_vocab_in):
     "<|notimestamps|>",
     *[f"<|{i * 0.02:.2f}|>" for i in range(1501)],
   ]
-  special_tokens = {}
-  for token in specials:
-    special_tokens[token] = n_vocab
-    n_vocab += 1
+  special_tokens = dict(zip(specials, itertools.count(n_vocab)))
+  n_vocab += len(specials)
   assert n_vocab == n_vocab_in
   import tiktoken
   return tiktoken.Encoding(