Initial commit of GPT-2 46M SwiGLU Dense baseline model architecture and configs

configuration_gpt2.py

@@ -1,7 +1,7 @@
 from transformers import PretrainedConfig
 
 class GPT2Config(PretrainedConfig):
-    model_type = "gpt2"
+    model_type = "gpt2_custom"
     
     def __init__(
         self,

modeling_gpt2.py

@@ -163,5 +163,5 @@ class GPT2CustomLMHeadModel(PreTrainedModel, GenerationMixin):
         return {"input_ids": input_ids, "past_key_values": past_key_values}
 
 # Register configuration and model for auto mapping
-AutoConfig.register("gpt2", GPT2Config)
+AutoConfig.register("gpt2_custom", GPT2Config)
 AutoModelForCausalLM.register(GPT2Config, GPT2CustomLMHeadModel)