Update values of `bos_token_id` and `eos_token_id` in generation_config.json

The vocab size is 49152 so the token ids should not be set to 50256.
See https://huggingface.slack.com/archives/C031GD0V5S5/p1717609689224529 for more context.

generation_config.json

@@ -1,6 +1,6 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 50256,
-  "eos_token_id": 50256,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
   "transformers_version": "4.37.0.dev0"
 }