update bos and eos config

config.json

@@ -8,9 +8,9 @@
     "AutoConfig": "configuration_lola_gpt2.LOLAConfig",
     "AutoModelForCausalLM": "modeling_lola_gpt2.LOLALMHeadModel"
   },
-  "bos_token_id": 100095,
+  "bos_token_id": 0,
   "embd_pdrop": 0.1,
-  "eos_token_id": 100095,
+  "eos_token_id": 5,
   "initializer_range": 0.02,
   "layer_norm_epsilon": 1e-05,
   "model_type": "lola_v1",

configuration_lola_gpt2.py

@@ -22,7 +22,7 @@ class LOLAConfig(PretrainedConfig):
 
     def __init__(
         self,
-        vocab_size=100096,
+        vocab_size=100096, # Padded vocab size, actual size is 100000
         n_positions=2048,
         n_embd=2048,
         n_layer=24,
@@ -41,8 +41,8 @@ class LOLAConfig(PretrainedConfig):
         summary_first_dropout=0.1,
         scale_attn_weights=True,
         use_cache=True,
-        bos_token_id=100095,
-        eos_token_id=100095,
+        bos_token_id=0,
+        eos_token_id=5,
         scale_attn_by_inverse_layer_idx=False,
         reorder_and_upcast_attn=False,
         num_experts=16,

generation_config.json

@@ -1,6 +1,6 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 100095,
-  "eos_token_id": 100095,
+  "bos_token_id": 0,
+  "eos_token_id": 5,
   "transformers_version": "4.39.1"
 }