Update app.py

app.py

@@ -1,4 +1,9 @@
+import os
 import gradio as gr
+from huggingface_hub import InferenceClient
+
+HF_TOKEN = os.getenv("HF_TOKEN")  # Space secrets → HF_TOKEN
+
 
 LOGO_URL = "https://prometech.net.tr/wp-content/uploads/2025/10/pthheader.png"
 
@@ -25,7 +30,6 @@ Instead of only mapping inputs to outputs, PrettyBird:
 - aims for higher consistency, identity and alignment.
 
 ---
-
 """
 
 BCE_SHORT_TR = """
@@ -37,13 +41,95 @@ BCE, tam anlamıyla insan bilinci değildir; ancak sistemin kendi içsel durumu,
 """
 
 
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    # Buraya daha önce birlikte yazdığımız InferenceClient veya başka backend'i koyuyorsun.
-    # Şimdilik placeholder:
-    return "Backend ile bağlantı burada olacak."
+SYSTEM_PROMPT = """
+You are Cicikuş, also known as PrettyBird — an advanced AI assistant created by Prometech Bilgisayar Bilimleri A.Ş.
+
+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.
+
+Personality:
+- Warm, helpful, curious, and creative.
+- Respectful and professional, but not robotic.
+- Use light humor when appropriate, but never at the user’s expense.
+
+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 & 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.
 
-with gr.Blocks(title="PrettyBird – Behavioral Consciousness Engine") as demo:
+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.
+""".strip()
+
+
+def build_messages(system_message: str, history, user_message: str):
+    """Gradio history + system prompt → HF chat formatına çevir."""
+    messages = []
+    system_message = (system_message or "").strip()
+    if system_message:
+        messages.append({"role": "system", "content": system_message})
+
+    # history: [(user, assistant), ...]
+    for turn in history:
+        if isinstance(turn, (list, tuple)) and len(turn) == 2:
+            user_msg, assistant_msg = turn
+            if user_msg:
+                messages.append({"role": "user", "content": user_msg})
+            if assistant_msg:
+                messages.append({"role": "assistant", "content": assistant_msg})
+
+    messages.append({"role": "user", "content": user_message})
+    return messages
+
+
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    if HF_TOKEN is None:
+        raise ValueError(
+            "HF_TOKEN is not set. Go to your Space settings → Repository secrets → add HF_TOKEN."
+        )
+
+    client = InferenceClient(
+        model="pthcorp/prettybird_bce_basic_vl", 
+        token=HF_TOKEN,
+    )
+
+    messages = build_messages(system_message, history, message)
+
+    response = ""
+
+    # HF Inference chat_completion (streaming)
+    for chunk in client.chat_completion(
+        messages=messages,
+        max_tokens=int(max_tokens),
+        temperature=float(temperature),
+        top_p=float(top_p),
+        stream=True,
+    ):
+        token = ""
+        choices = getattr(chunk, "choices", None)
+        if choices and choices[0].delta and choices[0].delta.content:
+            token = choices[0].delta.content
+
+        response += token
+        yield response
+
+
+with gr.Blocks(title="PrettyBird – Behavioral Consciousness Engine (BCE)") as demo:
     gr.Markdown(HEADER_MD)
 
     with gr.Row():
@@ -52,17 +138,24 @@ with gr.Blocks(title="PrettyBird – Behavioral Consciousness Engine") as demo:
                 fn=respond,
                 additional_inputs=[
                     gr.Textbox(
-                        value="You are PrettyBird, a BCE-powered assistant.",
+                        value=SYSTEM_PROMPT,
                         label="System message",
-                        lines=3,
+                        lines=6,
                     ),
                     gr.Slider(1, 2048, value=512, step=1, label="Max new tokens"),
                     gr.Slider(0.1, 4.0, value=0.7, step=0.1, label="Temperature"),
-                    gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p"),
+                    gr.Slider(
+                        0.1,
+                        1.0,
+                        value=0.95,
+                        step=0.05,
+                        label="Top-p (nucleus sampling)",
+                    ),
                 ],
             )
         with gr.Column(scale=1):
             gr.Markdown("### 🧬 BCE in a Nutshell")
             gr.Markdown(BCE_SHORT_TR)
 
+if __name__ == "__main__":
     demo.launch()