From 6a55a8afdfec911d759e1ba8cefd72ab9bf912d0 Mon Sep 17 00:00:00 2001
From: urchade <urchadcassim@outlook.fr>
Date: Fri, 9 May 2025 18:56:55 +0400
Subject: [PATCH] add load from config

---
 gliner/__init__.py |  2 +-
 gliner/model.py    | 47 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 48 insertions(+), 1 deletion(-)

diff --git a/gliner/__init__.py b/gliner/__init__.py
index 3523a6d..40dc326 100644
--- a/gliner/__init__.py
+++ b/gliner/__init__.py
@@ -1,4 +1,4 @@
-__version__ = "0.2.19"
+__version__ = "0.2.20"
 
 from .model import GLiNER
 from .config import GLiNERConfig
diff --git a/gliner/model.py b/gliner/model.py
index 40904b1..73d5c47 100644
--- a/gliner/model.py
+++ b/gliner/model.py
@@ -859,3 +859,50 @@ def _from_pretrained(
                 new_num_tokens, None
             )
         return gliner
+
+    @staticmethod
+    def load_from_config(gliner_config: GLiNERConfig):
+        # Initialize tokenizer
+        tokenizer = AutoTokenizer.from_pretrained(
+            gliner_config.model_name,
+            model_max_length=gliner_config.max_len
+        )
+
+        # Add special tokens and update config
+        gliner_config.class_token_index = len(tokenizer)
+        tokenizer.add_tokens([
+            gliner_config.ent_token,
+            gliner_config.sep_token
+        ])
+        gliner_config.vocab_size = len(tokenizer)
+
+        # Select appropriate processor
+        words_splitter = WordsSplitter()
+        if gliner_config.span_mode == "token_level":
+            data_processor = TokenProcessor(
+                gliner_config,
+                tokenizer,
+                words_splitter,
+                preprocess_text=True
+            )
+        else:
+            data_processor = SpanProcessor(
+                gliner_config,
+                tokenizer,
+                words_splitter,
+                preprocess_text=True
+            )
+
+        # Instantiate model and apply token resizing
+        model = GLiNER(
+            gliner_config,
+            data_processor=data_processor
+        )
+
+        model.resize_token_embeddings(
+            [gliner_config.ent_token, gliner_config.sep_token],
+            set_class_token_index=False,
+            add_tokens_to_tokenizer=False
+        )
+
+        return model
\ No newline at end of file