Update app.py

app.py

@@ -75,14 +75,16 @@ generation_params = {
     "use_cache": False
 }
 
+# Generate synthetic data
 def generate_synthetic_data(description, columns):
     try:
-        # Prepare the input for the Llama model
         formatted_prompt = format_prompt(description, columns)
 
         # Tokenize the prompt with truncation enabled
-        inputs = tokenizer_llama(formatted_prompt, return_tensors="pt", truncation=True, max_length=512).to(model_llama.device)
-        print(f"Input Tensor Size: {inputs['input_ids'].size()}")
+        inputs = tokenizer_llama(formatted_prompt, return_tensors="pt", truncation=True, max_length=512)
+        
+        # Move inputs to the correct device
+        inputs = {k: v.to(model_llama.device) for k, v in inputs.items()}
 
         # Generate synthetic data
         with torch.no_grad():
@@ -94,15 +96,19 @@ def generate_synthetic_data(description, columns):
                 num_return_sequences=1,
             )
 
+        # Check for meta tensor before decoding
+        if outputs.is_meta:
+            raise ValueError("Output tensor is in meta state, check model and input.")
+
         # Decode the generated output
         generated_text = tokenizer_llama.decode(outputs[0], skip_special_tokens=True)
-
+        
         # Return the generated synthetic data
         return generated_text
     except Exception as e:
-        st.error(f"Error in generate_synthetic_data: {e}")
         return f"Error: {e}"
 
+
 def generate_large_synthetic_data(description, columns, num_rows=1000, rows_per_generation=100):
     data_frames = []
     num_iterations = num_rows // rows_per_generation