Align bos_token_id with umt5

https://huggingface.co/google/umt5-small/blob/main/generation_config.json
{
"_from_model_config": true,
"decoder_start_token_id": 0,
"eos_token_id": 1,
"max_new_tokens": 64,
"pad_token_id": 0,
"transformers_version": "4.31.0.dev0"
}

config.json

@@ -6,7 +6,7 @@
     "AutoModelForCausalLM": "modeling_openmoe.OpenMoeForCausalLM"
   },
   "attention_bias": false,
-  "bos_token_id": 2,
+  "bos_token_id": 0,
   "dropout_rate": 0.0,
   "enable_comm_overlap": false,
   "enable_hierarchical_alltoall": false,