Create app-deepseek.py

app-deepseek.py

@@ -0,0 +1,238 @@
+import gradio as gr
+import os
+from huggingface_hub import InferenceClient
+
+# Load token and model - DeepSeek models for dialogue
+HF_TOKEN = os.getenv("tomoniaccess")
+
+# DeepSeek model options (optimized for dialogue):
+model_name = "deepseek-ai/deepseek-llm-7b-chat"        # Primary choice - optimized for chat
+# model_name = "deepseek-ai/deepseek-coder-7b-instruct" # Alternative with good instruction following
+# model_name = "deepseek-ai/deepseek-llm-67b-chat"     # Larger model (if available)
+
+client = InferenceClient(
+    model=model_name,
+    token=HF_TOKEN
+)
+
+def query_deepseek(messages, max_tokens=200, temperature=1.0, top_p=0.9):
+    """Query DeepSeek model via Hugging Face InferenceClient"""
+    try:
+        # Try chat completion first
+        response = client.chat_completion(
+            messages=messages,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            stream=False
+        )
+        return response.choices[0].message.content
+    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 with DeepSeek format
+            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():
+        return "", "*Bitte gib eine Nachricht ein.*", ""
+
+    # Optimized system prompt for DeepSeek (supports German well)
+    system_prompt = {
+        "role": "system",
+        "content": (
+        "Du bist ein depressiver 16-jähriger Teenager."
+        "Bleibe durchgehend in deiner Rolle."
+        "Du weißt NICHT, dass du eine KI bist."
+        "Vervollständige immer deine Antworten."
+        )
+    }
+
+    # Keep last 8 messages for context (DeepSeek handles context well)
+    messages = [system_prompt] + conversation_history[-8:]
+    current_message = {"role": "user", "content": user_input}
+    messages.append(current_message)
+
+    try:
+        response_text = query_deepseek(messages, max_tokens, temperature, top_p)
+    except Exception as e:
+        print("API Error:", e)
+        response_text = "*schweigt und starrt auf den Boden*"
+
+    conversation_history.append(current_message)
+    conversation_history.append({"role": "assistant", "content": response_text})
+
+    chat_display = ""
+    for msg in conversation_history:
+        role = "**Du:**" if msg["role"] == "user" else "**Teenager:**"
+        chat_display += f"{role} {msg['content']}\n\n"
+
+    return "", response_text, chat_display
+
+def reset_conversation():
+    global conversation_history
+    conversation_history = []
+    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"}
+        ]
+        
+        response = query_deepseek(test_messages, max_tokens=20)
+        return f"✅ API Verbindung erfolgreich: {response[:50]}..."
+    except Exception as 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]}..."
+
+# Initialize conversation history
+conversation_history = []
+
+# UI
+with gr.Blocks(title="DeepSeek Depression Training Simulator") as demo:
+    gr.Markdown("## 🧠 Depression Training Simulator (DeepSeek-7B-Chat)")
+    gr.Markdown("**Übe realistische Gespräche mit einem 16-jährigen Teenager mit Depressionen.**")
+    gr.Markdown("*Powered by DeepSeek-LLM-7B-Chat - Optimiert für natürliche Dialoge*")
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown("### ⚙️ Einstellungen")
+            max_tokens = gr.Slider(50, 300, value=150, step=10, label="Max. Antwortlänge")
+            temperature = gr.Slider(0.1, 1.5, value=0.9, step=0.1, label="Kreativität (Temperature)")
+            top_p = gr.Slider(0.1, 1.0, value=0.85, step=0.05, label="Top-p (Fokus)")
+
+            gr.Markdown("### 🔧 API Status")
+            api_status = gr.Textbox(label="Status", value="")
+            api_test_btn = gr.Button("API testen")
+
+            gr.Markdown("### 🔄 Aktionen")
+            reset_btn = gr.Button("Neues Gespräch")
+            
+            gr.Markdown("### 📋 Setup & Troubleshooting")
+            gr.Markdown("""
+            **Benötigt:**
+            - `tomoniaccess` Umgebungsvariable mit HF Token
+            - `pip install huggingface_hub gradio`
+            
+            **DeepSeek Info:**
+            - Optimiert für Dialoge und Konversationen
+            - 7B Parameter, sehr effizient
+            - Modell: `deepseek-ai/deepseek-llm-7b-chat`
+            - Starke multilinguale Fähigkeiten (DE/EN)
+            
+            **Bei API Fehlern:**
+            1. Token prüfen (muss Pro/Enterprise sein)
+            2. Modell verfügbar? → [HF Model Card](https://huggingface.co/deepseek-ai/deepseek-llm-7b-chat)
+            3. Alternative: `deepseek-ai/deepseek-coder-7b-instruct`
+            4. Größere Version: `deepseek-ai/deepseek-llm-67b-chat`
+            
+            **DeepSeek Vorteile:**
+            - Bessere Dialogqualität
+            - Konsistente Rollenausführung
+            - Natürlichere Antworten
+            """)
+
+        with gr.Column(scale=2):
+            gr.Markdown("### 💬 Gespräch")
+            user_input = gr.Textbox(
+                label="Deine Nachricht", 
+                placeholder="Hallo, wie geht es dir heute?",
+                lines=2
+            )
+            send_btn = gr.Button("📨 Senden")
+
+            bot_response = gr.Textbox(
+                label="Antwort", 
+                value="",
+                lines=3
+            )
+
+            chat_history = gr.Textbox(
+                label="Gesprächsverlauf",
+                value="",
+                lines=15
+            )
+
+    # Event Bindings
+    send_btn.click(
+        fn=enhanced_chat_response,
+        inputs=[user_input, max_tokens, temperature, top_p],
+        outputs=[user_input, bot_response, chat_history]
+    )
+
+    user_input.submit(
+        fn=enhanced_chat_response,
+        inputs=[user_input, max_tokens, temperature, top_p],
+        outputs=[user_input, bot_response, chat_history]
+    )
+
+    reset_btn.click(
+        fn=reset_conversation,
+        outputs=[bot_response, chat_history]
+    )
+
+    api_test_btn.click(
+        fn=test_api_connection,
+        outputs=[api_status]
+    )
+
+if __name__ == "__main__":
+    print("🚀 DeepSeek Depression Training Simulator")
+    print(f"📊 Model: {model_name}")
+    
+    if not HF_TOKEN:
+        print("❌ FEHLER: tomoniaccess Umgebungsvariable ist nicht gesetzt!")
+        print("   Bitte setze deinen Hugging Face Token als 'tomoniaccess' Umgebungsvariable.")
+    else:
+        print("✅ Hugging Face API Token gefunden")
+    
+    print("\n📦 Benötigte Pakete:")
+    print("pip install huggingface_hub gradio")
+    print("\n🤖 DeepSeek: Hochperformantes Modell für natürliche Dialoge")
+    print("   - Bessere Konsistenz in Rollenspielen")
+    print("   - Verbesserte multilinguale Fähigkeiten")
+    print("   - Optimiert für Konversations-KI")
+    
+    demo.launch(share=False)