Update app.py

app.py

@@ -9,7 +9,7 @@ from huggingface_hub import InferenceClient, hf_hub_download
 
 # 🔹 Hugging Face Credentials
 HF_REPO = "Futuresony/future_ai_12_10_2024.gguf"
-HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Store your token as an environment variable for security
+HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Ensure this is set in your environment
 
 # 🔹 FAISS Index Path
 FAISS_PATH = "asa_faiss.index"
@@ -24,34 +24,35 @@ faiss_index = faiss.read_index(faiss_local_path)
 # 🔹 Initialize Hugging Face Model Client
 client = InferenceClient(model=HF_REPO, token=HF_TOKEN)
 
-# 🔹 Retrieve Relevant FAISS Context
-def retrieve_relevant_context(user_query, top_k=3):
+# 🔹 Retrieve Relevant FAISS Data
+def retrieve_faiss_knowledge(user_query, top_k=3):
     query_embedding = embedder.encode([user_query], convert_to_tensor=True).cpu().numpy()
     distances, indices = faiss_index.search(query_embedding, top_k)
 
     retrieved_texts = []
     for idx in indices[0]:  # Extract top_k results
         if idx != -1:  # Ensure valid index
-            retrieved_texts.append(f"Example: {idx} → {idx}")  # Customize how retrieved data appears
+            retrieved_texts.append(f"Example {idx}: (Extracted FAISS Data)")
 
-    return "\n".join(retrieved_texts) if retrieved_texts else "No relevant data found."
+    return "\n".join(retrieved_texts) if retrieved_texts else "**No relevant FAISS data found.**"
 
-# 🔹 Format Model Prompt with FAISS Guidance
-def format_prompt(user_input, system_prompt, history):
-    retrieved_context = retrieve_relevant_context(user_input)
+# 🔹 Chatbot Response Function (Forcing FAISS Context)
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    faiss_context = retrieve_faiss_knowledge(message)
 
-    faiss_instruction = (
-        "Use the following example responses as a guide for formatting and writing style:\n"
-        f"{retrieved_context}\n\n"
-        "### Instruction:\n"
-        f"{user_input}\n\n### Response:"
-    )
+    # 🔥 Force the model to use FAISS
+    full_prompt = f"""### System Instruction:
+You MUST use the provided FAISS data to generate your response.
+If no FAISS data is found, return "I don't have enough information."
 
-    return faiss_instruction
+### Retrieved FAISS Data:
+{faiss_context}
 
-# 🔹 Chatbot Response Function
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    full_prompt = format_prompt(message, system_message, history)
+### User Query:
+{message}
+
+### Response:
+"""
 
     response = client.text_generation(
         full_prompt,
@@ -60,18 +61,18 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
         top_p=top_p,
     )
 
-    # ✅ Extract only model-generated response
+    # ✅ Extract only the model-generated response
     cleaned_response = response.split("### Response:")[-1].strip()
-    
+
     history.append((message, cleaned_response))  # ✅ Update chat history
-    
+
     yield cleaned_response  # ✅ Output the response
 
 # 🔹 Gradio Chat Interface
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are a helpful AI trained to follow FAISS-based writing styles.", label="System message"),
+        gr.Textbox(value="You are a knowledge assistant that must use FAISS context.", label="System message"),
         gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.9, step=0.1, label="Temperature"),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.99, step=0.01, label="Top-p (nucleus sampling)"),