Update app.py

app.py

@@ -4,28 +4,44 @@ import faiss
 import pickle
 import numpy as np
 from sentence_transformers import SentenceTransformer
-
 import os
+
 print("Files in current directory:", os.listdir())
 
 # -----------------------------
-# Load LLM
+# Globals (lazy-loaded)
 # -----------------------------
-model = Llama(
-    model_path="qwen2.5-1.5B-q4.gguf",
-    n_ctx=4096,
-    n_gpu_layers=0,
-    chat_format="qwen",
-)
+model = None
+embed_model = None
+index = None
+chunks = None
+metadata = None
 
 # -----------------------------
-# Load RAG
+# Lazy-loading functions
 # -----------------------------
-embed_model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
+def load_llm():
+    global model
+    if model is None:
+        print("Loading LLM...")
+        model = Llama(
+            model_path="qwen2.5-1.5B-q4.gguf",
+            n_ctx=4096,
+            n_gpu_layers=0,
+            chat_format="qwen",
+        )
+        print("LLM loaded.")
+
+def load_rag():
+    global embed_model, index, chunks, metadata
+    if embed_model is None or index is None or chunks is None or metadata is None:
+        print("Loading embedding model and FAISS index...")
+        embed_model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
+        index = faiss.read_index("faiss_index.bin")
+        chunks = pickle.load(open("chunks.pkl", "rb"))
+        metadata = pickle.load(open("metadata.pkl", "rb"))
+        print("RAG components loaded.")
 
-index = faiss.read_index("faiss_index.bin")
-chunks = pickle.load(open("chunks.pkl", "rb"))
-metadata = pickle.load(open("metadata.pkl", "rb"))
 # -----------------------------
 # Detect query intent
 # -----------------------------
@@ -51,6 +67,8 @@ def detect_query(query):
 # Retrieve context (RAG)
 # -----------------------------
 def retrieve_context(query):
+    load_rag()  # ensure RAG is loaded
+
     animal, topic = detect_query(query)
 
     filtered_indices = []
@@ -65,9 +83,7 @@ def retrieve_context(query):
         filtered_indices = list(range(len(chunks)))
 
     query_embedding = embed_model.encode([query])
-
-    filtered_embeddings = [index.reconstruct(i) for i in filtered_indices]
-    filtered_embeddings = np.array(filtered_embeddings)
+    filtered_embeddings = np.array([index.reconstruct(i) for i in filtered_indices])
 
     distances = np.linalg.norm(filtered_embeddings - query_embedding, axis=1)
     top_indices = distances.argsort()[:2]
@@ -80,9 +96,11 @@ def retrieve_context(query):
     return context
 
 # -----------------------------
-# Chat function (UPDATED)
+# Chat function
 # -----------------------------
 def chat(user_input):
+    load_llm()  # ensure LLM is loaded
+
     context = retrieve_context(user_input)
 
     prompt = f"""
@@ -114,11 +132,15 @@ Answer in short and clear sentences.
     return response["choices"][0]["message"]["content"]
 
 # -----------------------------
-# Gradio UI (UNCHANGED)
+# Gradio UI
 # -----------------------------
-gr.Interface(
+demo = gr.Interface(
     fn=chat,
     inputs="text",
     outputs="text",
-    title="Livestock Chatbot"
-).launch()
+    title="Livestock Chatbot",
+    description="Ask questions about goats and cows. The assistant answers using only the provided knowledge base."
+)
+
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)