Messages

app.py

@@ -22,17 +22,32 @@ MODEL_NAME = "deepseek-ai/DeepSeek-R1"
 OUTPUT_DIR = "finetuned_models"
 LOGS_DIR = "training_logs"
 
-def save_uploaded_file(file):
+def save_uploaded_file(file_obj):
     """Save uploaded file and return its path"""
-    os.makedirs('uploads', exist_ok=True)
-    import tempfile
-    
-    # Create a temporary file with .csv extension
-    temp = tempfile.NamedTemporaryFile(delete=False, suffix='.csv', dir='uploads')
-    temp.write(file)
-    temp.close()
-    
-    return temp.name
+    try:
+        os.makedirs('uploads', exist_ok=True)
+        
+        if hasattr(file_obj, 'name'):
+            # If it's a FileUpload object
+            file_path = os.path.join('uploads', os.path.basename(file_obj.name))
+            if isinstance(file_obj, (bytes, bytearray)):
+                with open(file_path, 'wb') as f:
+                    f.write(file_obj)
+            else:
+                file_obj.save(file_path)
+        else:
+            # If it's raw bytes
+            import tempfile
+            fd, file_path = tempfile.mkstemp(suffix='.csv', dir='uploads')
+            with os.fdopen(fd, 'wb') as temp:
+                if isinstance(file_obj, (bytes, bytearray)):
+                    temp.write(file_obj)
+                else:
+                    temp.write(file_obj.read())
+        
+        return file_path
+    except Exception as e:
+        raise Exception(f"Error saving file: {str(e)}")
 
 def prepare_training_data(df):
     """Convert DataFrame into Q&A format"""
@@ -133,6 +148,49 @@ def train_model(
     progress=gr.Progress()
 ):
     """Training function for Gradio interface"""
+    if file is None:
+        return "Please upload a file first."
+    
+    try:
+        # File validation
+        progress(0.1, desc="Validating file...")
+        file_path = save_uploaded_file(file)
+        
+        # Prepare components
+        progress(0.2, desc="Preparing training components...")
+        components = prepare_training_components(
+            file_path,
+            learning_rate,
+            num_epochs,
+            batch_size
+        )
+        
+        # Initialize trainer
+        progress(0.4, desc="Initializing trainer...")
+        trainer = Trainer(
+            model=components['model'],
+            args=components['training_args'],
+            train_dataset=components['dataset'],
+            data_collator=components['data_collator'],
+        )
+        
+        # Train
+        progress(0.5, desc="Training model...")
+        trainer.train()
+        
+        # Save model and tokenizer
+        progress(0.9, desc="Saving model...")
+        trainer.save_model()
+        components['tokenizer'].save_pretrained(components['output_dir'])
+        
+        progress(1.0, desc="Training complete!")
+        return f"Training completed! Model saved in {components['output_dir']}"
+        
+    except Exception as e:
+        error_msg = f"Error during training: {str(e)}"
+        print(error_msg)  # Log the error
+        return error_msg
+    """Training function for Gradio interface"""
     try:
         # Save uploaded file
         file_path = save_uploaded_file(file)
@@ -172,60 +230,64 @@ def train_model(
 
 # Create Gradio interface
 def create_interface():
-    with gr.Blocks() as demo:
-        gr.Markdown("# DeepSeek-R1 Model Finetuning Interface")
-        
-        with gr.Row():
-            with gr.Column():
-                file_input = gr.File(
-                    label="Upload Training Data (CSV)",
-                    type="binary",
-                    file_types=[".csv"]
-                )
-                
-                learning_rate = gr.Slider(
-                    minimum=1e-5,
-                    maximum=1e-3,
-                    value=2e-4,
-                    label="Learning Rate"
-                )
-                
-                num_epochs = gr.Slider(
-                    minimum=1,
-                    maximum=10,
-                    value=3,
-                    step=1,
-                    label="Number of Epochs"
-                )
-                
-                batch_size = gr.Slider(
-                    minimum=1,
-                    maximum=8,
-                    value=4,
-                    step=1,
-                    label="Batch Size"
-                )
-                
-                train_button = gr.Button("Start Training")
-                
-            with gr.Column():
-                output = gr.Textbox(label="Training Status")
-                
-        train_button.click(
-            fn=train_model,
-            inputs=[file_input, learning_rate, num_epochs, batch_size],
-            outputs=output
-        )
-        
-        gr.Markdown("""
-        ## Instructions
-        1. Upload your training data in CSV format with columns:
-           - chunk_id (questions)
-           - text (answers)
-        2. Adjust training parameters if needed
-        3. Click 'Start Training'
-        4. Wait for training to complete
-        """)
+    # Configure Gradio to handle larger file uploads
+    demo = gr.Interface(
+        title="Model Fine-tuning Interface"
+    )
+    
+    gr.Config(upload_size_limit=100)
+    
+    with gr.Row():
+        with gr.Column():
+            file_input = gr.File(
+                label="Upload Training Data (CSV)",
+                type="binary",
+                file_types=[".csv"]
+            )
+            
+            learning_rate = gr.Slider(
+                minimum=1e-5,
+                maximum=1e-3,
+                value=2e-4,
+                label="Learning Rate"
+            )
+            
+            num_epochs = gr.Slider(
+                minimum=1,
+                maximum=10,
+                value=3,
+                step=1,
+                label="Number of Epochs"
+            )
+            
+            batch_size = gr.Slider(
+                minimum=1,
+                maximum=8,
+                value=4,
+                step=1,
+                label="Batch Size"
+            )
+            
+            train_button = gr.Button("Start Training")
+            
+        with gr.Column():
+            output = gr.Textbox(label="Training Status")
+            
+    train_button.click(
+        fn=train_model,
+        inputs=[file_input, learning_rate, num_epochs, batch_size],
+        outputs=output
+    )
+    
+    gr.Markdown("""
+    ## Instructions
+    1. Upload your training data in CSV format with columns:
+       - chunk_id (questions)
+       - text (answers)
+    2. Adjust training parameters if needed
+    3. Click 'Start Training'
+    4. Wait for training to complete
+    """)
     
     return demo