adjust config

config.json

@@ -2,8 +2,8 @@
   "architectures": [
     "AriaForCausalLM"
   ],
-  "bos_token_id": 0,
   "eos_token_id": 1,
+  "pad_token_id": 2,
   "hidden_size": 1536,
   "intermediate_size": 6144,
   "max_seq_len": 8192,

config_emb.json

@@ -2,8 +2,8 @@
   "architectures": [
     "AriaForSequenceEmbedding"
   ],
-  "bos_token_id": 0,
   "eos_token_id": 1,
+  "pad_token_id": 2,
   "hidden_size": 1536,
   "embedding_size": 512,
   "intermediate_size": 6144,

configuration_aria.py

@@ -15,8 +15,8 @@ class AriaConfig(PretrainedConfig):
         intermediate_size: int = 6144,
         max_seq_len: int = 8192,
         use_cache: bool = True,
-        bos_token_id: int = 0,
         eos_token_id: int = 1,
+        pad_token_id: int = 2,
         tie_word_embeddings: bool = False,
         output_attentions: bool = False,
         output_hidden_states: bool = False,
@@ -24,7 +24,9 @@ class AriaConfig(PretrainedConfig):
         **kwargs,
     ):
         super().__init__(
-            bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs
+            pad_token_id=pad_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs,
         )
         self.vocab_size = vocab_size
         self.hidden_size = hidden_size