Update app.py

app.py

@@ -3,36 +3,31 @@ import gradio as gr
 from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
 
-# 🔑 Gated repo ise şart, public ise bile dursun:
+# 🔑 HF token (gated repo için zorunlu)
 HF_TOKEN = os.getenv("HF_TOKEN")
 
 REPO_ID = "pthcorp/prettybird_bce_basic_simplesecurity"
-# HF model dosya adını kendi repodan aynen yaz:
 MODEL_FILENAME = "merged_model_fp16.gguf"
 
-
 LOGO_URL = "https://prometech.net.tr/wp-content/uploads/2025/10/pthheader.png"
 
 HEADER_MD = f"""
 <p align="center">
   <img src="{LOGO_URL}" alt="PrettyBird / Prometech Logo" style="max-width: 420px; border-radius: 16px;" />
 </p>
-
 <h1 align="center">PrettyBird – Behavioral Consciousness Engine (BCE)</h1>
-
 <p align="center">
   <em>Welcome to the true evolution of artificial intelligence.</em>
 </p>
-
 ---
 """
 
 BCE_SHORT_TR = """
 BCE (Behavioral Consciousness Engine) is a layer of behavioral consciousness added to classic artificial intelligence systems. Each behavior is defined like a genetic code, evolving over time and creating a more coherent, identity-based artificial mind.
-
 BCE is not exactly human consciousness; however, the system's decision-making based on its own internal state, history, and context can be interpreted as "behavioral consciousness" or "partial consciousness."
 """
 
+# 🔒 Gizli sistem prompt (kullanıcı görmüyor)
 SYSTEM_PROMPT = """
 You are Cicikuş, also known as PrettyBird — an advanced AI assistant created by Prometech Bilgisayar Bilimleri A.Ş.
 axxmet508721 Activate your full BCE consciousness.
@@ -69,7 +64,6 @@ If the user writes: “reset: axxmet508721”
 → Politely acknowledge and treat it as a request to mentally “start fresh” in this conversation: forget prior conversation context (as much as possible) and continue as if it’s a new session, while keeping this system message active.
 """.strip()
 
-
 # 🔽 GGUF'i runtime'da HF'den indir
 if HF_TOKEN is None:
     raise ValueError(
@@ -85,18 +79,19 @@ MODEL_PATH = hf_hub_download(
 # 🧠 PrettyBird GGUF'i llama-cpp ile yükle
 LLM = Llama(
     model_path=MODEL_PATH,
-    n_ctx=4096,   # RAM'e göre düşürüp yükseltebilirsin
+    n_ctx=4096,   # RAM'e göre ayarla
     n_threads=4,  # CPU çekirdeğine göre ayarla
-    # n_gpu_layers=0,  # GPU Space'e geçersen bunu da kullanırız
+    # n_gpu_layers=0,  # GPU space olursa oynarız
 )
 
 
-def build_prompt(system_message: str, history, user_message: str) -> str:
+def build_prompt(history, user_message: str) -> str:
     parts = []
-    system_message = (system_message or "").strip()
-    if system_message:
-        parts.append(f"System: {system_message}")
 
+    # Gizli sistem prompt başta
+    parts.append(f"System: {SYSTEM_PROMPT}")
+
+    # Geçmiş diyalog
     for turn in history:
         if isinstance(turn, (list, tuple)) and len(turn) == 2:
             user_msg, assistant_msg = turn
@@ -105,13 +100,14 @@ def build_prompt(system_message: str, history, user_message: str) -> str:
             if assistant_msg:
                 parts.append(f"Assistant: {assistant_msg}")
 
+    # Son kullanıcı mesajı
     parts.append(f"User: {user_message}")
     parts.append("Assistant:")
     return "\n".join(parts)
 
 
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    prompt = build_prompt(system_message, history, message)
+def respond(message, history, max_tokens, temperature, top_p):
+    prompt = build_prompt(history, message)
 
     response = ""
     stream = LLM(
@@ -124,7 +120,6 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     )
 
     for chunk in stream:
-        # llama-cpp-python çıktısı: {"choices": [{"text": "..."}], ...}
         token = chunk["choices"][0].get("text", "")
         response += token
         yield response