improve model loading

tasks/text.py

@@ -37,38 +37,16 @@ class TextClassifier:
         
         try:
             # Initialize tokenizer
-            self.tokenizer = AutoTokenizer.from_pretrained(
-                TOKENIZER_NAME,
-                model_max_length=8192,
-                padding_side='right',
-                truncation_side='right'
-            )
-            
-            # Load model configuration
-            model_config = {
-                "architectures": ["ModernBertForSequenceClassification"],
-                "model_type": "modernbert",
-                "num_labels": 8,
-                "problem_type": "single_label_classification",
-                "hidden_size": 768,
-                "num_attention_heads": 12,
-                "num_hidden_layers": 22,
-                "intermediate_size": 1152,
-                "max_position_embeddings": 8192,
-                "torch_dtype": "float32",
-                "transformers_version": "4.48.3",
-                "layer_norm_eps": 1e-05
-            }
+            self.tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME)
             
             # Initialize model
-            self.model = AutoModelForSequenceClassification.from_pretrained(
+            self.model = BertForSequenceClassification.from_pretrained(
                 MODEL_NAME,
-                config=model_config,
-                ignore_mismatched_sizes=True,
-                trust_remote_code=True
+                num_labels=8,
+                ignore_mismatched_sizes=True
             ).to(self.device)
             
-            # Convert to half precision
+            # Convert to half precision and eval mode
             self.model = self.model.half()
             self.model.eval()
             
@@ -79,7 +57,6 @@ class TextClassifier:
             raise
 
     def process_batch(self, batch):
-        """Process a batch of texts and return their predictions"""
         try:
             # Move batch to device
             input_ids = batch['input_ids'].to(self.device)