Update app.py

app.py

@@ -1,20 +1,10 @@
 import gradio as gr
 import gspread
-import time
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 from oauth2client.service_account import ServiceAccountCredentials
-from llama_cpp import Llama
-from llama_index.core import VectorStoreIndex, Settings
-from llama_index.core.node_parser import SentenceSplitter
-from llama_index.embeddings.huggingface import HuggingFaceEmbedding
-from llama_index.llms.llama_cpp import LlamaCPP
-from huggingface_hub import hf_hub_download
-from llama_index.core.llms import ChatMessage
-from llama_index.core.chat_engine.condense_plus_context import CondensePlusContextChatEngine
-from llama_index.core.schema import Document
-
-# ===================================
-# 1️⃣ Cache Data Google Sheets
-# ===================================
+
+# =============== 1. Cache Google Sheets ===============
 cached_text_data = None
 
 def read_google_sheets():
@@ -26,13 +16,13 @@ def read_google_sheets():
         scope = ["https://www.googleapis.com/auth/spreadsheets", "https://www.googleapis.com/auth/drive"]
         creds = ServiceAccountCredentials.from_json_keyfile_name("credentials.json", scope)
         client = gspread.authorize(creds)
-        
+
         SPREADSHEET_ID = "1e_cNMhwF-QYpyYUpqQh-XCw-OdhWS6EuYsoBUsVtdNg"
         sheet_names = ["datatarget", "datacuti", "dataabsen", "datalembur", "pkb"]
 
         all_data = []
         spreadsheet = client.open_by_key(SPREADSHEET_ID)
-        
+
         for sheet_name in sheet_names:
             try:
                 sheet = spreadsheet.worksheet(sheet_name)
@@ -41,119 +31,81 @@ def read_google_sheets():
                 all_data.extend([" | ".join(row) for row in data])
                 all_data.append("\n")
             except gspread.exceptions.WorksheetNotFound:
-                all_data.append(f"❌ ERROR: Worksheet {sheet_name} tidak ditemukan.")
+                all_data.append(f"❌ Worksheet {sheet_name} tidak ditemukan.")
 
         cached_text_data = "\n".join(all_data).strip()
         return cached_text_data
-    
-    except gspread.exceptions.SpreadsheetNotFound:
-        return "❌ ERROR: Spreadsheet tidak ditemukan!"
-    
+
     except Exception as e:
         return f"❌ ERROR: {str(e)}"
 
-# ===================================
-# 2️⃣ Inisialisasi Model Llama
-# ===================================
-def initialize_llama_model():
-    model_path = hf_hub_download(
-        repo_id="TheBloke/zephyr-7b-beta-GGUF",
-        filename="zephyr-7b-beta.Q4_K_M.gguf",
-        cache_dir="./models"
+
+# =============== 2. Load Model Transformers ===============
+def load_model():
+    model_id = "mistralai/Mistral-7B-Instruct-v0.2"  # Bisa ganti ke Zephyr juga
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_id,
+        device_map="auto",  # 🧠 Otomatis ke GPU jika tersedia
+        torch_dtype=torch.float16
     )
-    return model_path
-
-# ===================================
-# 3️⃣ Inisialisasi Pengaturan Model
-# ===================================
-def initialize_settings(model_path):
-    Settings.llm = LlamaCPP(
-        model_path=model_path,
-        temperature=0.7,
-        context_window=4096,
+    pipe = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        device=0 if torch.cuda.is_available() else -1,
         max_new_tokens=512,
-        # n_gpu_layers=20,  # ❌ Hapus jika error
-        model_kwargs={"n_ctx": 4096}
+        temperature=0.7,
+        repetition_penalty=1.2,
+        do_sample=True,
     )
+    return pipe
 
-# ===================================
-# 4️⃣ Inisialisasi Index & Chat Engine
-# ===================================
-def initialize_index():
-    text_data = read_google_sheets()
-    document = Document(text=text_data)
-    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
-    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
-    chat_engine = CondensePlusContextChatEngine.from_defaults(
-        retriever=retriever,
-        verbose=False
-    )
-    return chat_engine
-
-# ===================================
-# 5️⃣ Fungsi untuk Merapikan Jawaban Chatbot
-# ===================================
-def clean_response(response):
-    text = "".join(response.response_gen)
-    text = text.replace("\n\n", "\n").strip()
-    text = text.replace("user:", "").replace("jawaban:", "").replace("assistant:", "").strip()
-    return text
-
-# ===================================
-# 6️⃣ Fungsi untuk Menghasilkan Respons Chatbot
-# ===================================
-def generate_response(message, history, chat_engine):
-    if history is None:
-        history = []
-
-    chat_messages = [
-        ChatMessage(
-            role="system",
-            content=(
-                "Anda adalah chatbot HRD yang membantu karyawan memahami administrasi perusahaan. "
-                "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. "
-                "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
-
-    history.append((message, cleaned_text))
-    return cleaned_text
-
-# ===================================
-# 7️⃣ Fungsi Utama untuk Menjalankan Aplikasi
-# ===================================
+
+# =============== 3. Buat Prompt dan Jawaban ===============
+def generate_prompt(user_message, context_data):
+    prompt = f"""
+### SISTEM:
+Anda adalah asisten HRD yang membantu karyawan memahami administrasi perusahaan. Jawablah dengan Bahasa Indonesia yang profesional dan ramah. Jika tidak tahu, katakan tidak tahu dengan sopan. Jangan jawab dalam bahasa Inggris.
+
+### DATA:
+{context_data}
+
+### PERTANYAAN:
+{user_message}
+
+### JAWABAN:
+"""
+    return prompt.strip()
+
+
+# =============== 4. Generate Response ===============
+def generate_response(message, history, pipe):
+    context = read_google_sheets()
+    full_prompt = generate_prompt(message, context)
+    response = pipe(full_prompt)[0]["generated_text"]
+
+    # Bersihkan jawaban: ambil bagian setelah ### JAWABAN:
+    cleaned = response.split("### JAWABAN:")[-1].strip()
+    history = history or []
+    history.append((message, cleaned))
+    return cleaned
+
+
+# =============== 5. Jalankan Gradio ===============
 def main():
-    model_path = initialize_llama_model()
-    initialize_settings(model_path)
-    
-    index = initialize_index()
-    chat_engine = initialize_chat_engine(index)
-    
+    pipe = load_model()
+
     def chatbot_response(message, history):
-        return generate_response(message, history, chat_engine)
+        return generate_response(message, history, pipe)
 
     gr.Interface(
         fn=chatbot_response,
         inputs=["text"],
         outputs=["text"],
-    ).launch()
+        title="Chatbot HRD - Transformers",
+        theme="compact"
+    ).launch(share=True)
 
 if __name__ == "__main__":
     main()