adapted modeling code for HF with AutoModel and AutoConfig

modeling_pharia.py

@@ -13,6 +13,8 @@ from transformers.modeling_attn_mask_utils import AttentionMaskConverter
 from transformers.modeling_outputs import (
     BaseModelOutputWithPast,
 )
+from transformers import AutoConfig
+from transformers import AutoModel
 from transformers.modeling_utils import PreTrainedModel
 try: 
     from flash_attn.flash_attn_interface import flash_attn_func
@@ -22,6 +24,7 @@ except Exception as e:
     )
     flash_attn_func = None
     
+PHARIAEMBED_TYPE = "phariaembed"
 
 class RotaryConfig():
     def __init__(
@@ -1005,4 +1008,17 @@ class PhariaForEmbedding(PhariaPreTrainedModel):
         if input_was_string:
             all_embeddings = all_embeddings[0]
 
-        return all_embeddings
+        return all_embeddings
+
+
+# registration for Autoconfig and auto class
+
+AutoConfig.register(PHARIAEMBED_TYPE, PhariaConfig)
+
+PhariaConfig.register_for_auto_class()
+
+# registration for AutoModel and auto class
+
+AutoModel.register(PhariaConfig, PhariaForEmbedding)
+
+PhariaForEmbedding.register_for_auto_class("AutoModel")