Update app.py

app.py

@@ -16,75 +16,74 @@ def safe_json_loads(x):
         return np.array(json.loads(x))
     except json.JSONDecodeError as e:
         print(f"Error decoding JSON: {e}")
-        return np.zeros(128)  # Return a default array of zeros
+        return np.array([])  # Return an empty array or handle it as appropriate
 
 # Apply the safe_json_loads function to the embedding column
 data['embedding'] = data['embedding'].apply(safe_json_loads)
 
 # Filter out any rows with empty embeddings
-data = data[data['embedding'].apply(lambda x: x is not None and len(x) > 0)]
+data = data[data['embedding'].apply(lambda x: x.size > 0)]
 
 # Check if the DataFrame is empty after filtering
 if data.empty:
-    print("No valid embeddings found in the data. Using default values.")
-else:
-    # Initialize FAISS index
-    dimension = len(data['embedding'].iloc[0])
-    res = faiss.StandardGpuResources()  # use a single GPU
-
-    # Check available GPU devices
-    num_gpus = faiss.get_num_gpus()
-    if num_gpus > 0:
-        gpu_index = faiss.IndexFlatL2(dimension)
-        gpu_index = faiss.index_cpu_to_gpu(res, 0, gpu_index)  # move to GPU
-    else:
-        raise RuntimeError("No GPU devices available.")
-
-    gpu_index.add(np.stack(data['embedding'].values))
-
-    # Check if GPU is available
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-
-    # Load QA model
-    qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad", device=0 if torch.cuda.is_available() else -1)
-
-    # Load BERT model and tokenizer
-    tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
-    model = BertModel.from_pretrained('bert-base-uncased').to(device)
-
-    # Function to embed the question using BERT
-    def embed_question(question, model, tokenizer):
-        inputs = tokenizer(question, return_tensors='pt').to(device)
-        with torch.no_grad():
-            outputs = model(**inputs)
-        return outputs.last_hidden_state.mean(dim=1).cpu().numpy()
-
-    # Function to retrieve the relevant document and generate a response
-    @spaces.GPU(duration=120)
-    def retrieve_and_generate(question):
-        # Embed the question
-        question_embedding = embed_question(question, model, tokenizer)
-        
-        # Search in FAISS index
-        _, indices = gpu_index.search(question_embedding, k=1)
-        
-        # Retrieve the most relevant document
-        relevant_doc = data.iloc[indices[0][0]]
-        
-        # Use the QA model to generate the answer
-        context = relevant_doc['Abstract']
-        response = qa_model(question=question, context=context)
-        
-        return response['answer']
-
-    # Create a Gradio interface
-    interface = gr.Interface(
-        fn=retrieve_and_generate,
-        inputs=gr.Textbox(lines=2, placeholder="Ask a question about the documents..."),
-        outputs="text",
-        title="RAG Chatbot",
-        description="Ask questions about the documents in the CSV file."
-    )
-
-    # Launch the Gradio app
-    interface.launch()
+    raise RuntimeError("No valid embeddings found in the data.")
+
+# Initialize FAISS index
+dimension = len(data['embedding'].iloc[0])
+gpu_available = torch.cuda.is_available()
+
+# Initialize FAISS resources and index
+res = faiss.StandardGpuResources() if gpu_available else None
+index = faiss.IndexFlatL2(dimension)
+
+if gpu_available:
+    index = faiss.index_cpu_to_gpu(res, 0, index)  # move to GPU
+
+index.add(np.stack(data['embedding'].values))
+
+# Set the device
+device = torch.device('cuda' if gpu_available else 'cpu')
+
+# Load QA model
+qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad", device=0 if gpu_available else -1)
+
+# Load BERT model and tokenizer
+tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+model = BertModel.from_pretrained('bert-base-uncased').to(device)
+
+# Function to embed the question using BERT
+def embed_question(question, model, tokenizer):
+    inputs = tokenizer(question, return_tensors='pt').to(device)
+    with torch.no_grad():
+        outputs = model(**inputs)
+    return outputs.last_hidden_state.mean(dim=1).cpu().numpy()
+
+# Function to retrieve the relevant document and generate a response
+@spaces.GPU(duration=120)
+def retrieve_and_generate(question):
+    # Embed the question
+    question_embedding = embed_question(question, model, tokenizer)
+    
+    # Search in FAISS index
+    _, indices = index.search(question_embedding, k=1)
+    
+    # Retrieve the most relevant document
+    relevant_doc = data.iloc[indices[0][0]]
+    
+    # Use the QA model to generate the answer
+    context = relevant_doc['Abstract']
+    response = qa_model(question=question, context=context)
+    
+    return response['answer']
+
+# Create a Gradio interface
+interface = gr.Interface(
+    fn=retrieve_and_generate,
+    inputs=gr.Textbox(lines=2, placeholder="Ask a question about the documents..."),
+    outputs="text",
+    title="RAG Chatbot",
+    description="Ask questions about the documents in the CSV file."
+)
+
+# Launch the Gradio app
+interface.launch()