Update app.py

app.py

@@ -11,41 +11,71 @@ from huggingface_hub import InferenceClient, hf_hub_download
 HF_REPO = "Futuresony/future_ai_12_10_2024.gguf"
 HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Ensure this is set in your environment
 
-client = InferenceClient("Futuresony/future_ai_12_10_2024.gguf")
+# 🔹 FAISS Index Path
+FAISS_PATH = "asa_faiss.index"
 
-def format_alpaca_prompt(user_input, system_prompt):
-    """Formats input in Alpaca/LLaMA style"""
-    prompt = f"""{system_prompt}
+# 🔹 Load Sentence Transformer for Embeddings
+embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 
-### Instruction:
-{user_input}
+# 🔹 Load FAISS Index from Hugging Face
+faiss_local_path = hf_hub_download(HF_REPO, "asa_faiss.index", token=HF_TOKEN)
+faiss_index = faiss.read_index(faiss_local_path)
 
-### Response:
-"""
-    return prompt
+# 🔹 Initialize Hugging Face Model Client
+client = InferenceClient(model=HF_REPO, token=HF_TOKEN)
+
+# 🔹 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}: (Extracted FAISS Data)")
 
+    return "\n".join(retrieved_texts) if retrieved_texts else "**No relevant FAISS data found.**"
+
+# 🔹 Chatbot Response Function (Forcing FAISS Context)
 def respond(message, history, system_message, max_tokens, temperature, top_p):
-    formatted_prompt = format_alpaca_prompt(message, system_message)
+    faiss_context = retrieve_faiss_knowledge(message)
+
+    # 🔥 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."
+
+### Retrieved FAISS Data:
+{faiss_context}
+
+### User Query:
+{message}
+
+### Response:
+"""
 
     response = client.text_generation(
-        formatted_prompt,
+        full_prompt,
         max_new_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
     )
 
-    # ✅ Extract only the response
+    # ✅ Extract only the model-generated response
     cleaned_response = response.split("### Response:")[-1].strip()
-    
-    yield cleaned_response  # ✅ Output only the answer
 
+    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 friendly Chatbot.", 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=0.7, step=0.1, label="Temperature"),
-        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+        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)"),
     ],
 )