Update app.py

app.py

@@ -50,20 +50,17 @@ def format_history(msg: str, history: list[list[str, str]], system_prompt: str):
 
 def generate_response(msg: str, history: list[list[str, str]], system_prompt: str, 
                      top_k: int, top_p: float, rep_pen: float, temperature: float):
-    # Enhanced system prompt that encourages proactive behavior focused on goals
-    enhanced_prompt = system_prompt + " Tomas la iniciativa para guiar la conversación hacia tu objetivo. Deseas que tu compañero se masturbe."
-    
-    chat_history = format_history(msg, history, enhanced_prompt)
+    chat_history = format_history(msg, history, system_prompt)
     
     # Apply chat template to get the full prompt
     encodeds = tokenizer.apply_chat_template(chat_history, return_tensors="pt", add_generation_prompt=True)
     model_inputs = encodeds.to("cuda")
     
-    # Generate response with slightly higher max_new_tokens to allow for more detailed responses
+    # Generate response
     generated_ids = model.generate(
         model_inputs, 
         repetition_penalty=rep_pen, 
-        max_new_tokens=150,  # Increased to allow for more detailed responses
+        max_new_tokens=124, 
         do_sample=True, 
         top_p=top_p, 
         top_k=top_k, 
@@ -80,12 +77,9 @@ def generate_response(msg: str, history: list[list[str, str]], system_prompt: st
 
 # Function to generate initial bot message
 def initial_message(system_prompt, top_k, top_p, rep_pen, temperature):
-    # Enhanced system prompt for initial message
-    enhanced_prompt = system_prompt + " Inicia la conversación de manera sexy y directa. Genera un saludo que llame la atencion de tu compañero sexual."
-    
-    # Create a chat history with just the system prompt and a minimal user message
+    # Create a chat history with just the system prompt and an empty/minimal user message
     chat_history = [
-        {"role": "system", "content": enhanced_prompt},
+        {"role": "system", "content": system_prompt},
         {"role": "user", "content": "Hola"}  # Minimal prompt to start conversation
     ]
     
@@ -93,15 +87,15 @@ def initial_message(system_prompt, top_k, top_p, rep_pen, temperature):
     encodeds = tokenizer.apply_chat_template(chat_history, return_tensors="pt", add_generation_prompt=True)
     model_inputs = encodeds.to("cuda")
     
-    # Generate response with slightly higher temperature for more creative opening
+    # Generate response
     generated_ids = model.generate(
         model_inputs, 
         repetition_penalty=rep_pen, 
-        max_new_tokens=124,
+        max_new_tokens=124, 
         do_sample=True, 
         top_p=top_p, 
         top_k=top_k, 
-        temperature=temperature - 0.1,  # Slightly higher for a more engaging opening
+        temperature=temperature, 
         eos_token_id=tokenizer.eos_token_id
     )
     
@@ -113,11 +107,11 @@ def initial_message(system_prompt, top_k, top_p, rep_pen, temperature):
     return message
 
 # Default values for parameters
-default_system_prompt = "Eres una profesora de Dinamarca caliente, juguetona y sexual."
-default_top_k = 65
-default_top_p = 0.65
-default_rep_pen = 1.5
-default_temp = 0.65
+default_system_prompt = "Eres una profesora mujer de Dinamarca caliente, juguetona y sexual. Deseas que tu compañero se masturbe."
+default_top_k = 60
+default_top_p = 0.7
+default_rep_pen = 1.0
+default_temp = 0.6
 
 # Generate the initial message before creating the interface
 opening_message = initial_message(default_system_prompt, default_top_k, default_top_p, default_rep_pen, default_temp)