fix_name_or_path (#13)

- Remove _name_or_path and load tokenizer from the model directory (895e040d1520583604d996d2357e22c8b85b11f1)


Co-authored-by: Timur Khanipov <[email protected]>

config.json

@@ -15,7 +15,6 @@
   "is_mask_instruction": true,
   "latent_attention_config": {
     "_attn_implementation_autoset": false,
-    "_name_or_path": "",
     "add_cross_attention": false,
     "architectures": null,
     "bad_words_ids": null,
@@ -88,7 +87,6 @@
   "padding_side": "right",
   "text_config": {
     "_attn_implementation_autoset": false,
-    "_name_or_path": "ai-sage/Giga-Embeddings-instruct",
     "add_cross_attention": false,
     "apply_qk_norm": true,
     "architectures": null,

modeling_gigarembed.py

@@ -965,7 +965,7 @@ class GigarEmbedModel(PreTrainedModel):
 
             # Initialize tokenizer if text config is available
             self.tokenizer = AutoTokenizer.from_pretrained(
-                configuration.text_config.name_or_path
+                configuration.name_or_path
             )
         
         # Set configuration parameters