Update app.py

app.py

@@ -1,6 +1,5 @@
 import gradio as gr
 import gspread
-import time
 from oauth2client.service_account import ServiceAccountCredentials
 from llama_cpp import Llama
 from llama_index.core import VectorStoreIndex, Settings
@@ -13,15 +12,9 @@ from llama_index.core.chat_engine.condense_plus_context import CondensePlusConte
 from llama_index.core.schema import Document
 
 # ===================================
-# 1️⃣ Cache Data Google Sheets
+# 1️⃣ Fungsi Membaca Data Google Spreadsheet
 # ===================================
-cached_text_data = None
-
 def read_google_sheets():
-    global cached_text_data
-    if cached_text_data is not None:
-        return cached_text_data
-
     try:
         scope = ["https://www.googleapis.com/auth/spreadsheets", "https://www.googleapis.com/auth/drive"]
         creds = ServiceAccountCredentials.from_json_keyfile_name("credentials.json", scope)
@@ -43,8 +36,7 @@ def read_google_sheets():
             except gspread.exceptions.WorksheetNotFound:
                 all_data.append(f"❌ ERROR: Worksheet {sheet_name} tidak ditemukan.")
 
-        cached_text_data = "\n".join(all_data).strip()
-        return cached_text_data
+        return "\n".join(all_data).strip()
     
     except gspread.exceptions.SpreadsheetNotFound:
         return "❌ ERROR: Spreadsheet tidak ditemukan!"
@@ -57,7 +49,7 @@ def read_google_sheets():
 # ===================================
 def initialize_llama_model():
     model_path = hf_hub_download(
-        repo_id="TheBloke/zephyr-7b-beta-GGUF",
+        repo_id="TheBLoke/zephyr-7b-beta-GGUF",
         filename="zephyr-7b-beta.Q4_K_M.gguf",
         cache_dir="./models"
     )
@@ -66,17 +58,8 @@ def initialize_llama_model():
 # ===================================
 # 3️⃣ Inisialisasi Pengaturan Model
 # ===================================
-
 def initialize_settings(model_path):
-    Settings.llm = LlamaCPP(
-        model_path=model_path,
-        temperature=0.7,
-        context_window=4096,
-        max_new_tokens=512,
-        # n_gpu_layers=20,  # ❌ Hapus jika error
-        model_kwargs={"n_ctx": 4096}
-    )
-
+    Settings.llm = LlamaCPP(model_path=model_path, temperature=0.7)
 
 # ===================================
 # 4️⃣ Inisialisasi Index & Chat Engine
@@ -87,17 +70,17 @@ def initialize_index():
     parser = SentenceSplitter(chunk_size=100, chunk_overlap=30)
     nodes = parser.get_nodes_from_documents([document])
     
-    embedding = HuggingFaceEmbedding("sentence-transformers/all-MiniLM-L6-v2")  # ✅ Lebih ringan
+    embedding = HuggingFaceEmbedding("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
     Settings.embed_model = embedding
     
     index = VectorStoreIndex(nodes)
     return index
 
 def initialize_chat_engine(index):
-    retriever = index.as_retriever(similarity_top_k=1)  # ✅ Kurangi ke 1 untuk kecepatan
+    retriever = index.as_retriever(similarity_top_k=3)
     chat_engine = CondensePlusContextChatEngine.from_defaults(
         retriever=retriever,
-        verbose=False
+        verbose=False  # ❌ Hapus verbose agar tidak ada referensi dokumen
     )
     return chat_engine
 
@@ -105,8 +88,8 @@ def initialize_chat_engine(index):
 # 5️⃣ Fungsi untuk Merapikan Jawaban Chatbot
 # ===================================
 def clean_response(response):
-    text = "".join(response.response_gen)
-    text = text.replace("\n\n", "\n").strip()
+    text = "".join(response.response_gen)  # Gabungkan teks yang dihasilkan
+    text = text.replace("\n\n", "\n").strip()  # Hilangkan newline berlebihan
     text = text.replace("user:", "").replace("jawaban:", "").replace("assistant:", "").strip()
     return text
 
@@ -125,17 +108,17 @@ def generate_response(message, history, chat_engine):
                 "Jangan menjawab menggunakan Bahasa Inggris. "
                 "Gunakan Bahasa Indonesia dengan gaya profesional dan ramah. "
                 "Jika informasi tidak tersedia dalam dokumen, katakan dengan sopan bahwa Anda tidak tahu. "
-                "Jawaban harus singkat, jelas, dan sesuai konteks. "
+                "Jawaban harus singkat, jelas, dan sesuai konteks."
                 "Jangan memberikan jawaban untuk pertanyaan yang tidak diajukan oleh pengguna. "
                 "Jangan menyertakan rekomendasi pertanyaan lain."
             ),
         ),
     ]
 
-    response = chat_engine.chat(message)  # GANTI: pakai .chat() bukan .stream_chat()
-    cleaned_text = response.response.strip()  # GANTI: langsung ambil response
+    response = chat_engine.stream_chat(message)
+    cleaned_text = clean_response(response)  # 🔹 Gunakan fungsi clean_response()
 
-    history.append((message, cleaned_text))
+    history.append((message, cleaned_text))  # 🔹 Pastikan hanya teks yang masuk ke history
     return cleaned_text
 
 # ===================================
@@ -144,10 +127,10 @@ def generate_response(message, history, chat_engine):
 def main():
     model_path = initialize_llama_model()
     initialize_settings(model_path)
-    
+
     index = initialize_index()
     chat_engine = initialize_chat_engine(index)
-    
+
     def chatbot_response(message, history):
         return generate_response(message, history, chat_engine)