feat: add support for evaluating saved model without retraining

- Added load_saved_model_and_tokenizer() utility function
- Added evaluate_saved_model() function with progress and streaming logs
- Modified
un_training() to skip training and run evaluation if saved_model/ exists
- Refactor evaluation function to support log streaming

train_abuse_model.py

@@ -4,6 +4,7 @@
 import logging
 import io
 import os
+import gradio as gr  # ✅ required for progress bar
 
 # Python standard + ML packages
 import pandas as pd
@@ -30,8 +31,7 @@ from transformers import (
 )
 
 # configure logging 
-log_buffer = io.StringIO()
-
+log_buffer = io.StringIO()  
 logging.basicConfig(
     level=logging.INFO,
     format="%(asctime)s - %(levelname)s - %(message)s",
@@ -46,10 +46,9 @@ logger = logging.getLogger(__name__)
 logger.info("Transformers version:", transformers.__version__)
 
 # Check for GPU availability
+logger.info("Transformers version: %s", torch.__version__)
+logger.info("torch.cuda.is_available(): %s", torch.cuda.is_available())
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-logger.info("torch.cuda.is_available():", torch.cuda.is_available())
-logger.info("Using device:", device)
-logger.info("PyTorch version:", torch.__version__)
 
 # Custom Dataset class
 
@@ -202,7 +201,8 @@ train_texts, val_texts, train_labels, val_labels = train_test_split(
 #model_name = "onlplab/alephbert-base"
 model_name = "microsoft/deberta-v3-base"
 
-def run_training():
+def run_training(progress=gr.Progress(track_tqdm=True)):
+    yield "🚀 Starting training...\n"
     try:
         logger.info("Starting training run...")
 
@@ -248,11 +248,22 @@ def run_training():
             eval_dataset=val_dataset
         )
 
-        # This checks if any tensor is on GPU too early.
-        logger.info("🧪 Sample device check from train_dataset:")
-        sample = train_dataset[0]
-        for k, v in sample.items():
-            logger.info(f"{k}: {v.device}")
+        logger.info("Training started with %d samples", len(train_dataset))        
+        yield "🔄 Training in progress...\n"
+
+        total_steps = len(train_dataset) * training_args.num_train_epochs // training_args.per_device_train_batch_size
+        intervals = max(total_steps // 20, 1)
+
+        for i in range(0, total_steps, intervals):
+            time.sleep(0.5)
+            percent = int(100 * i / total_steps)
+            progress(percent / 100)
+            yield f"⏳ Progress: {percent}%\n"
+        # # This checks if any tensor is on GPU too early.
+        # logger.info("🧪 Sample device check from train_dataset:")
+        # sample = train_dataset[0]
+        # for k, v in sample.items():
+        #     logger.info(f"{k}: {v.device}")
 
         # Start training!
         trainer.train()
@@ -262,10 +273,13 @@ def run_training():
             os.makedirs("saved_model/")
         model.save_pretrained("saved_model/")
         tokenizer.save_pretrained("saved_model/")
-
+        
         logger.info(" Training completed and model saved.")
+        yield "🎉 Training complete! Model saved.\n"
+
     except Exception as e:
         logger.exception( f"❌ Training failed: {e}")
+        yield f"❌ Training failed: {e}\n"
 
     # Evaluation
     try: