fix fast tokenizer for transformers >= "from_slow": false,

tokenizer_config.json

@@ -2,6 +2,6 @@
     "unk_token": "</s>",
     "bos_token": "<s>",
     "eos_token": "</s>",
-    "add_prefix_space": false,
+    "from_slow": false,
     "tokenizer_class": "LlamaTokenizer"
 }