Update app.py

app.py

@@ -1,154 +1,103 @@
 import gradio as gr
 from openai import OpenAI
-import requests
-import io
 import base64
+import io
 import os
 from PIL import Image
 
+# https://7wsby0fs8hw67a-8000.proxy.runpod.net/
 # ---------------- KONFIGURATION ----------------
-# POD_ID = "jyuqoe0kierp04"
+#POD_ID = "7wsby0fs8hw67a" # <--- DEINE ID (Prüfen!)
 POD_ID = os.getenv("POD_ID", "platzhalter_id")
-TEXT_URL = f"https://{POD_ID}-8000.proxy.runpod.net/v1"
-IMAGE_URL = f"https://{POD_ID}-8005.proxy.runpod.net/generate" 
+API_URL = f"https://{POD_ID}-8000.proxy.runpod.net/v1"
+API_KEY = os.getenv("API_KEY", "platzhalter_id")
 
-client = OpenAI(api_key="EMPTY", base_url=TEXT_URL)
+# client = OpenAI(api_key="EMPTY", base_url=API_URL)
+client = OpenAI(api_key=API_KEY, base_url=API_URL)
 # -----------------------------------------------
 
-# --- LOGIK (Bleibt gleich) ---
-def master_process(message, history, is_image_mode):
+def encode_image_to_base64(pil_image):
+    """Wandelt ein PIL-Bild in einen Base64-String um."""
+    if pil_image is None:
+        return None
+    buffered = io.BytesIO()
+    # Wir konvertieren zu RGB (falls PNG Transparenz hat) und speichern als JPEG
+    pil_image.convert("RGB").save(buffered, format="JPEG", quality=85)
+    return base64.b64encode(buffered.getvalue()).decode('utf-8')
+
+def chat_with_vision(message, history, uploaded_image):
     history = history or []
-    
-    if is_image_mode:
-        # BILD MODUS
-        history.append({"role": "user", "content": message})
-        history.append({"role": "assistant", "content": "🎨 **Bild-Modus aktiv:** Starte Flux..."})
-        yield history, None 
-        
-        try:
-            res = requests.get(IMAGE_URL, params={"prompt": message}, timeout=300)
-            if res.status_code == 200:
-                data = res.json()
-                b64 = data.get("image_base64")
-                if b64:
-                    img_data = base64.b64decode(b64)
-                    img = Image.open(io.BytesIO(img_data))
-                    img.load() 
-                    history[-1]["content"] = "Hier ist dein Kunstwerk! 👇"
-                    yield history, img 
-                else:
-                    history[-1]["content"] = "⚠️ Fehler: Leere Daten."
-                    yield history, None
-            else:
-                history[-1]["content"] = f"⚠️ Server Fehler: {res.status_code}"
-                yield history, None
-        except Exception as e:
-            history[-1]["content"] = f"❌ Verbindung fehlgeschlagen: {str(e)}"
-            yield history, None
 
+    # 1. User-Nachricht vorbereiten
+    user_content = []
+
+    # Wenn ein Bild da ist, fügen wir es hinzu
+    if uploaded_image is not None:
+        base64_image = encode_image_to_base64(uploaded_image)
+        # Das spezielle Format für Vision-Modelle (OpenAI Kompatibel)
+        user_content.append({
+            "type": "image_url",
+            "image_url": {
+                "url": f"data:image/jpeg;base64,{base64_image}"
+            }
+        })
+        history.append({"role": "user", "content": f"[Bild hochgeladen] {message}"})
     else:
-        # TEXT MODUS
         history.append({"role": "user", "content": message})
-        history.append({"role": "assistant", "content": ""})
-        yield history, None 
-        
-        messages_for_llm = [msg for msg in history if isinstance(msg['content'], str) and msg['content']]
-        try:
-            stream = client.chat.completions.create(
-                model="Qwen/Qwen2.5-14B-Instruct-AWQ",
-                messages=messages_for_llm,
-                stream=True,
-                max_tokens=2048
-            )
-            partial_text = ""
-            for chunk in stream:
-                if chunk.choices[0].delta.content:
-                    partial_text += chunk.choices[0].delta.content
-                    history[-1]["content"] = partial_text
-                    yield history, None
-        except Exception as e:
-            history[-1]["content"] = f"LLM Fehler: {str(e)}"
-            yield history, None
-
-# --- DESIGN & LAYOUT (Bereinigt) ---
-
-my_theme = gr.themes.Ocean(
-    primary_hue="indigo",
-    secondary_hue="blue",
-    neutral_hue="slate",
-).set(
-    body_background_fill="*neutral_50",
-    block_background_fill="white",
-    input_background_fill="*neutral_100"
-)
-
-my_css = """
-footer {visibility: hidden}
-.gradio-container {min-height: 0px !important}
-"""
-
-# Wir lassen theme/css hier drin. 
-# Die "UserWarning" (gelber Text) im Log ignorieren wir einfach.
-# Hauptsache kein roter "TypeError".
-with gr.Blocks(title="AI Studio") as demo:
-    
+
+    # Text hinzufügen
+    user_content.append({"type": "text", "text": message})
+
+    # 2. Nachricht an API senden
+    # Wir bauen die Messages-Liste für die API (ohne die lokalen History-Platzhalter)
+    api_messages = [{"role": "user", "content": user_content}]
+
+    history.append({"role": "assistant", "content": ""})
+
+    try:
+        stream = client.chat.completions.create(
+            model="Qwen/Qwen2.5-VL-7B-Instruct", # Name muss exakt zum Startbefehl passen!
+            messages=api_messages,
+            max_tokens=1024,
+            temperature=0.6,
+            stream=True
+        )
+
+        partial_text = ""
+        for chunk in stream:
+            if chunk.choices[0].delta.content:
+                partial_text += chunk.choices[0].delta.content
+                history[-1]["content"] = partial_text
+                yield history, None # Bild-Input leeren wir erst am Ende (optional)
+
+    except Exception as e:
+        history[-1]["content"] = f"⚠️ Fehler: {str(e)}"
+        yield history, None
+
+# --- UI DESIGN ---
+with gr.Blocks(title="👁️ Qwen Vision Test") as demo:
+    gr.Markdown("## 👁️ Qwen2.5-VL: Zeig mir ein Bild!")
+
     with gr.Row():
-        gr.Markdown("## 🤖 AI Cockpit: Qwen & Flux")
-
-    with gr.Row(equal_height=True):
-        # Links: Chat
-        with gr.Column(scale=3, min_width=400):
-            chatbot = gr.Chatbot(
-                height=550, 
-                # type="messages", # WICHTIG: Muss drin bleiben!
-                label="Chat Verlauf"
-                # ALLE Knöpfe entfernt, um Absturz zu verhindern
-            )
-
-        # Rechts: Bild
-        with gr.Column(scale=2, min_width=300):
-            image_display = gr.Image(
-                label="Ergebnis", 
-                show_label=True,
-                type="pil", 
-                height=500,
-                interactive=False
-            )
-
-    # --- INPUT BEREICH ---
-    gr.Markdown("---")
-    
-    with gr.Accordion("⚙️ Additional Tools / Einstellungen", open=False):
-        with gr.Row():
-            mode_check = gr.Checkbox(
-                label="🎨 Bild-Generierung aktivieren", 
-                value=False, 
-                info="Wenn aktiviert, wird Flux statt Qwen genutzt."
-            )
-    
-    with gr.Group():
-        with gr.Row():
-            msg = gr.Textbox(
-                scale=5, 
-                show_label=False, 
-                placeholder="Nachricht eingeben...", 
-                container=False,
-                autofocus=True
-            )
-            btn = gr.Button("Senden ➤", variant="primary", scale=1)
-
-    # --- EVENTS ---
-    msg.submit(
-        fn=master_process, 
-        inputs=[msg, chatbot, mode_check], 
-        outputs=[chatbot, image_display]
-    ).then(lambda: "", None, msg)
-    
+        with gr.Column(scale=1):
+            image_input = gr.Image(type="pil", label="Bild hochladen", height=300)
+            msg_input = gr.Textbox(placeholder="Was siehst du auf dem Bild?", label="Deine Frage")
+            btn = gr.Button("Senden", variant="primary")
+
+        with gr.Column(scale=2):
+            chatbot = gr.Chatbot(height=500, label="Antwort")
+
+    # Events
     btn.click(
-        fn=master_process, 
-        inputs=[msg, chatbot, mode_check], 
-        outputs=[chatbot, image_display]
-    ).then(lambda: "", None, msg)
+        fn=chat_with_vision,
+        inputs=[msg_input, chatbot, image_input],
+        outputs=[chatbot, image_input] # image_input hier, falls du es nach Senden löschen willst
+    )
+
+    msg_input.submit(
+        fn=chat_with_vision,
+        inputs=[msg_input, chatbot, image_input],
+        outputs=[chatbot, image_input]
+    )
 
-# Starten
-demo.launch(debug=False, share=False, theme=my_theme, css=my_css)
+demo.launch(debug=True, share=False)