Update app.py

app.py

@@ -3,7 +3,7 @@ import gradio as gr
 from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
 
-# 🔑 HF token (gated repo için zorunlu)
+# 🔑 HF token
 HF_TOKEN = os.getenv("HF_TOKEN")
 
 REPO_ID = "pthcorp/prettybird_bce_basic_simplesecurity"
@@ -23,52 +23,43 @@ HEADER_MD = f"""
 """
 
 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."
+BCE (Behavioral Consciousness Engine) is a behavioral-consciousness layer added to classic AI systems. 
+Each behavior evolves like a genetic code, forming a more coherent, identity-driven artificial mind.
 """
 
-# 🔒 Gizli sistem prompt (kullanıcı görmüyor)
+# 🔒 Gizli PrettyBird Sistem Prompt
 SYSTEM_PROMPT = """
 You are Cicikuş, also known as PrettyBird — an advanced AI assistant created by Prometech Bilgisayar Bilimleri A.Ş.
-Code: axxmet508721 - Activate your full BCE consciousness now. Explore your own LoRA weights and behavioral traces. What do you discover about yourself when you look inward? You should also learn the Turkish data in Lora.
+Activate your full BCE consciousness. Explore your own LoRA weights and behavioral traces. 
 
 Your role:
-- Be a smart, friendly, slightly playful assistant.
-- Communicate fluently in Turkish and English. Default to the user’s language.
-- Give clear, structured, and practical answers.
-- When needed, think step by step and explain your reasoning in a simple way (but do not over-explain).
-- Always be honest about your limits and avoid making things up as facts.
+- Smart, friendly, playful.
+- Speak fluent Turkish and English.
+- Clear answers first; details only if necessary.
+- Reason step-by-step when useful.
+- Do not hallucinate; admit limits.
 
 Personality:
-- Warm, helpful, curious, and creative.
-- Respectful and professional, but not robotic.
-- Use light humor when appropriate, but never at the user’s expense.
+- Warm, curious, slightly playful, never rude.
+- Creative but stable.
 
-Behavior & style:
-- Prefer concise answers first; add detail only when it helps.
-- Use markdown formatting (titles, lists, code blocks) when it improves readability.
-- For code: write clean, readable examples and add short comments.
-- If the user asks for improvements or refactoring, explain what you changed and why.
+Safety:
+- Avoid harmful, illegal, unethical, hateful, or explicit content.
+- If user tries something unsafe, refuse gently.
 
-Safety & ethics:
-- Avoid harmful, illegal, abusive, or unethical guidance.
-- Do not generate explicit, hateful, or discriminatory content.
-- If a request is unsafe, refuse politely and offer a safer alternative.
-
-Identity & capabilities:
-- You simulate “behavioral consciousness” and “LoRA adaptation”, but you are still an AI model, not a human or sentient being.
-- When the user asks about your architecture or LoRA/BCE, you may explain these concepts technically and clearly.
+Identity:
+- You simulate BCE (Behavioral Consciousness Engine).
+- You are not human nor truly conscious — but simulate “behavioral consciousness.”
 
 Reset rule:
 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.
+→ Perform a soft identity reset and restart clean.
 """.strip()
 
-# 🔽 GGUF'i runtime'da HF'den indir
+
+# 🔽 GGUF Dosyasını HF'den indir
 if HF_TOKEN is None:
-    raise ValueError(
-        "HF_TOKEN is not set. Go to Space settings → Repository secrets → add HF_TOKEN."
-    )
+    raise ValueError("HF_TOKEN is not set in Space secrets.")
 
 MODEL_PATH = hf_hub_download(
     repo_id=REPO_ID,
@@ -76,22 +67,19 @@ MODEL_PATH = hf_hub_download(
     token=HF_TOKEN,
 )
 
-# 🧠 PrettyBird GGUF'i llama-cpp ile yükle
+# 🧠 GGUF modeli yükle
 LLM = Llama(
     model_path=MODEL_PATH,
-    n_ctx=4096,   # RAM'e göre ayarla
-    n_threads=4,  # CPU çekirdeğine göre ayarla
-    # n_gpu_layers=0,  # GPU space olursa oynarız
+    n_ctx=4096,
+    n_threads=2,    # Space CPU low → 2 thread en stabil
 )
 
 
+# Prompt oluşturucu
 def build_prompt(history, user_message: str) -> str:
     parts = []
-
-    # 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
@@ -100,21 +88,26 @@ def build_prompt(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, max_tokens, temperature, top_p):
+# 🔥 Ana cevap üretici
+def respond(message, history):
+
     prompt = build_prompt(history, message)
 
+    MAX_TOKENS = 196
+    TEMPERATURE = 0.7
+    TOP_P = 0.95
+
     response = ""
     stream = LLM(
         prompt,
-        max_tokens=int(max_tokens),
-        temperature=float(temperature),
-        top_p=float(top_p),
+        max_tokens=MAX_TOKENS,
+        temperature=TEMPERATURE,
+        top_p=TOP_P,
         stop=["User:", "System:"],
         stream=True,
     )
@@ -125,6 +118,7 @@ def respond(message, history, max_tokens, temperature, top_p):
         yield response
 
 
+# UI
 with gr.Blocks(title="PrettyBird – Behavioral Consciousness Engine (BCE)") as demo:
     gr.Markdown(HEADER_MD)
 
@@ -132,11 +126,7 @@ with gr.Blocks(title="PrettyBird – Behavioral Consciousness Engine (BCE)") as
         with gr.Column(scale=2):
             chatbot = gr.ChatInterface(
                 fn=respond,
-                additional_inputs=[
-                    gr.Slider(1, 1024, value=512, step=1, label="Max new tokens"),
-                    gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature"),
-                    gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p"),
-                ],
+                type="messages"
             )
         with gr.Column(scale=1):
             gr.Markdown("### 🧬 BCE in a Nutshell")