Update app.py

app.py

@@ -15,9 +15,9 @@ import PyPDF2
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from sentence_transformers import SentenceTransformer
 
-# RAG components - using exact import syntax from installed packages
+# RAG components
 from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.vectorstores import FAISS  # Note: langchain_community not langchain.community
+from langchain_community.vectorstores import FAISS
 from langchain.schema import Document
 from langchain.embeddings import HuggingFaceEmbeddings
 
@@ -53,7 +53,7 @@ def detect_language(text):
     is_arabic = len(arabic_chars) > len(text) * 0.5
     return "arabic" if is_arabic else "english"
 
-# Define evaluation metrics
+# Evaluation metrics
 def calculate_bleu(prediction, reference):
     """Calculate BLEU score without any NLTK dependency"""
     # Tokenize texts using our own tokenizer
@@ -127,7 +127,7 @@ def calculate_f1_precision_recall(prediction, reference):
 
 def evaluate_retrieval_quality(contexts, query, language):
     """Evaluate the quality of retrieved contexts"""
-    # This is a placeholder function - simplified for testing
+    # This is a placeholder function
     return {
         'language_match_ratio': 1.0,
         'source_diversity': len(set([ctx.get('source', '') for ctx in contexts])) / max(1, len(contexts)),
@@ -207,7 +207,7 @@ def create_vector_store(documents):
     
     return vector_store
 
-# Model Loading and RAG System
+# Model Loading and RAG System - Improved to handle SentencePiece issues
 @spaces.GPU
 def load_model_and_tokenizer():
     """Load the ALLaM-7B model and tokenizer with error handling"""
@@ -215,6 +215,13 @@ def load_model_and_tokenizer():
     print(f"Loading model: {model_name}")
 
     try:
+        # Check if sentencepiece is installed
+        try:
+            import sentencepiece
+            print("SentencePiece is installed")
+        except ImportError:
+            print("Warning: SentencePiece is not installed. Attempting to proceed with AutoTokenizer only.")
+        
         # First attempt with AutoTokenizer
         tokenizer = AutoTokenizer.from_pretrained(
             model_name,
@@ -231,25 +238,20 @@ def load_model_and_tokenizer():
         )
         
         print("Model loaded successfully with AutoTokenizer!")
+        return model, tokenizer
         
     except Exception as e:
         print(f"First loading attempt failed: {e}")
-        print("Trying alternative loading approach...")
-        
-        # Try with specific tokenizer class if the first attempt fails
-        from transformers import LlamaTokenizer
         
-        tokenizer = LlamaTokenizer.from_pretrained(model_name)
-        model = AutoModelForCausalLM.from_pretrained(
-            model_name,
-            torch_dtype=torch.float16,
-            trust_remote_code=True,
-            device_map="auto",
-        )
+        # If SentencePiece error, provide helpful message
+        if "SentencePiece" in str(e):
+            raise ImportError(
+                "The model requires SentencePiece library which is missing. "
+                "Add 'sentencepiece>=0.1.95' to your requirements.txt file."
+            )
         
-        print("Model loaded successfully with LlamaTokenizer!")
-    
-    return model, tokenizer
+        # Other general error
+        raise Exception(f"Failed to load model: {e}")
 
 def retrieve_context(query, vector_store, top_k=5):
     """Retrieve most relevant document chunks for a given query"""
@@ -690,6 +692,13 @@ def main():
     print("Files in directory:", os.listdir("."))
     print("=" * 50)
     
+    # Check for SentencePiece
+    try:
+        import sentencepiece
+        print("SentencePiece is installed: ✓")
+    except ImportError:
+        print("WARNING: SentencePiece is NOT installed! This will cause errors with the tokenizer.")
+    
     # Initialize the system with simplified error handling
     try:
         # First create a very simple Gradio interface to show we're starting
@@ -698,7 +707,10 @@ def main():
             gr.Markdown("System is initializing. This may take a few minutes...")
             status = gr.Textbox(value="Loading resources...", label="Status")
         
-        app = loading_interface.queue()
+        with gr.Blocks(title="Vision 2030 Assistant - Model Loading") as model_interface:
+            gr.Markdown("# Vision 2030 Assistant - Loading Model")
+            gr.Markdown("The system is now loading the ALLaM-7B model. This may take several minutes.")
+            status = gr.Textbox(value="Loading model...", label="Status")
         
         # Now try the actual initialization
         try:
@@ -710,12 +722,48 @@ def main():
             
             print("Launching interface...")
             return interface
+        except ImportError as e:
+            print(f"Import error during initialization: {e}")
+            
+            # Create a simple error interface specifically for SentencePiece errors
+            if "SentencePiece" in str(e):
+                with gr.Blocks(title="Vision 2030 Assistant - SentencePiece Error") as sp_error:
+                    gr.Markdown("# Vision 2030 Assistant - SentencePiece Error")
+                    gr.Markdown("The model requires the SentencePiece library which is missing.")
+                    
+                    gr.Markdown("""
+                    ## How to Fix:
+                    
+                    Add these lines to your `requirements.txt` file:
+                    ```
+                    sentencepiece>=0.1.95
+                    protobuf>=3.20.0
+                    ```
+                    
+                    Then rebuild your Hugging Face Space.
+                    """)
+                    
+                return sp_error
+            else:
+                # For other import errors
+                with gr.Blocks(title="Vision 2030 Assistant - Import Error") as import_error:
+                    gr.Markdown("# Vision 2030 Assistant - Import Error")
+                    gr.Markdown(f"An import error occurred: {str(e)}")
+                    
+                    # Display possible solutions
+                    gr.Markdown("""
+                    ## Possible solutions:
+                    
+                    Check your `requirements.txt` file for missing dependencies.
+                    """)
+                    
+                return import_error
         except Exception as e:
             print(f"Error during initialization: {e}")
             import traceback
             traceback.print_exc()
             
-            # Create a simple error interface
+            # Create a general error interface
             with gr.Blocks(title="Vision 2030 Assistant - Error") as debug_interface:
                 gr.Markdown("# Vision 2030 Assistant - Initialization Error")
                 gr.Markdown("There was an error initializing the assistant.")
@@ -756,6 +804,15 @@ def main():
         with gr.Blocks(title="Vision 2030 Assistant - Critical Error") as critical_error:
             gr.Markdown("# Vision 2030 Assistant - Critical Error")
             gr.Markdown(f"A critical error occurred: {str(e)}")
+            
+            # Display stacktrace
+            import traceback
+            trace = traceback.format_exc()
+            gr.Textbox(
+                value=trace,
+                label="Error Traceback",
+                lines=15
+            )
         return critical_error
 
 if __name__ == "__main__":