Update app.py

app.py

@@ -2,19 +2,24 @@ import gradio as gr
 import os
 from huggingface_hub import InferenceClient
 
-# Load token and model
+# Load token and model - Try different models if one fails
 HF_TOKEN = os.getenv("tomoniaccess")
+
+# Model options (uncomment one that works):
+model_name = "LeoLM/leo-hessianai-13b-chat"        # Primary choice
+# model_name = "LeoLM/leo-hessianai-7b-chat"       # Smaller alternative
+# model_name = "microsoft/DialoGPT-medium"         # Reliable fallback
+# model_name = "google/flan-t5-large"              # Google alternative
+
 client = InferenceClient(
-    model="LeoLM/leo-hessianai-13b-chat",
+    model=model_name,
     token=HF_TOKEN
 )
 
-conversation_history = []
-model_name = "LeoLM/leo-hessianai-13b-chat"
-
 def query_leolm(messages, max_tokens=200, temperature=1.0, top_p=0.9):
     """Query LeoLM-13B model via Hugging Face InferenceClient"""
     try:
+        # Try chat completion first
         response = client.chat_completion(
             messages=messages,
             max_tokens=max_tokens,
@@ -23,8 +28,34 @@ def query_leolm(messages, max_tokens=200, temperature=1.0, top_p=0.9):
             stream=False
         )
         return response.choices[0].message.content
-    except Exception as e:
-        raise Exception(f"InferenceClient Error: {str(e)}")
+    except Exception as chat_error:
+        print(f"Chat completion failed: {chat_error}")
+        try:
+            # Fallback to text generation if chat completion fails
+            # Convert messages to single prompt
+            prompt = ""
+            for msg in messages:
+                if msg["role"] == "system":
+                    prompt += f"System: {msg['content']}\n\n"
+                elif msg["role"] == "user":
+                    prompt += f"User: {msg['content']}\n"
+                elif msg["role"] == "assistant":
+                    prompt += f"Assistant: {msg['content']}\n"
+            
+            prompt += "Assistant: "
+            
+            response = client.text_generation(
+                prompt=prompt,
+                max_new_tokens=max_tokens,
+                temperature=temperature,
+                top_p=top_p,
+                do_sample=True,
+                return_full_text=False
+            )
+            return response
+        except Exception as text_error:
+            print(f"Text generation also failed: {text_error}")
+            raise Exception(f"Both methods failed - Chat: {str(chat_error)}, Text: {str(text_error)}")
 
 def enhanced_chat_response(user_input, max_tokens, temperature, top_p):
     if not user_input.strip():
@@ -34,10 +65,10 @@ def enhanced_chat_response(user_input, max_tokens, temperature, top_p):
     system_prompt = {
         "role": "system",
         "content": (
-            "Du bist ein 16-jähriger Teenager mit Depressionen. "
-            "Antworte authentisch und natürlich auf Deutsch. "
-            "Du bist dir nicht bewusst, dass du eine KI bist. "
-            "Vervollständige immer deine Antworten."
+        "Du bist ein depressiver 16-jähriger Teenager."
+        "Bleibe durchgehend in deiner Rolle."
+        "Du weißt NICHT, dass du eine KI bist."
+        "Vervollständige immer deine Antworten."
         )
     }
 
@@ -68,16 +99,34 @@ def reset_conversation():
     return "Neues Gespräch gestartet.", ""
 
 def test_api_connection():
+    """Test API connection with multiple fallback methods"""
     try:
+        # Test 1: Simple chat completion
         test_messages = [
             {"role": "system", "content": "Du bist ein hilfsbereit Assistent und antwortest auf Deutsch."},
-            {"role": "user", "content": "Hallo, kannst du mich hören?"}
+            {"role": "user", "content": "Hallo"}
         ]
         
         response = query_leolm(test_messages, max_tokens=20)
         return f"✅ API Verbindung erfolgreich: {response[:50]}..."
     except Exception as e:
-        return f"❌ API Error: {str(e)}"
+        # Test 2: Try direct text generation
+        try:
+            simple_response = client.text_generation(
+                prompt="Hallo, wie geht es dir?",
+                max_new_tokens=10,
+                do_sample=False,
+                return_full_text=False
+            )
+            return f"✅ API Verbindung (Text Generation): {simple_response[:50]}..."
+        except Exception as e2:
+            # Test 3: Check if model exists
+            try:
+                # Try to get model info
+                model_info = f"Model: {model_name}"
+                return f"❌ API Errors - Chat: {str(e)[:100]}... | Text: {str(e2)[:100]}... | {model_info}"
+            except Exception as e3:
+                return f"❌ Vollständiger API Fehler: {str(e)[:200]}..."
 
 # UI
 with gr.Blocks(title="LeoLM Depression Training Simulator") as demo:
@@ -99,7 +148,7 @@ with gr.Blocks(title="LeoLM Depression Training Simulator") as demo:
             gr.Markdown("### 🔄 Aktionen")
             reset_btn = gr.Button("Neues Gespräch")
             
-            gr.Markdown("### 📋 Setup")
+            gr.Markdown("### 📋 Setup & Troubleshooting")
             gr.Markdown("""
             **Benötigt:**
             - `tomoniaccess` Umgebungsvariable mit HF Token
@@ -108,7 +157,13 @@ with gr.Blocks(title="LeoLM Depression Training Simulator") as demo:
             **LeoLM Info:**
             - Deutsche Sprachoptimierung
             - 13B Parameter
-            - Bessere deutsche Konversation
+            - Modell: `LeoLM/leo-hessianai-13b-chat`
+            
+            **Bei API Fehlern:**
+            1. Token prüfen (muss Pro/Enterprise sein)
+            2. Modell verfügbar? → [HF Model Card](https://huggingface.co/LeoLM/leo-hessianai-13b-chat)
+            3. Alternative: `LeoLM/leo-hessianai-7b-chat`
+            4. Fallback: `microsoft/DialoGPT-medium`
             """)
 
         with gr.Column(scale=2):