Update app.py

app.py

@@ -115,18 +115,56 @@ def run_gpt(
         logging.info(LOG_RESPONSE.format(resp))  # Log the response
     return resp
 
-def generate(prompt, history, agent_name=agents[0], sys_prompt="", temperature=0.7, max_new_tokens=2048, top_p=0.8, repetition_penalty=1.5, model="mistralai/Mixtral-8x7B-Instruct-v0.1"):
+def generate(
+    prompt, history, agent_name=agents[0], sys_prompt="", temperature=0.9, max_new_tokens=2048, top_p=0.95, repetition_penalty=1.0, model="mistralai/Mixtral-8x7B-Instruct-v0.1"
+):
+    seed = random.randint(1,1111111111111111)
+    logging.info(f"Seed: {seed}")  # Log the seed
+
+    # Set the agent prompt based on agent_name
+    if agent_name == "WEB_DEV":
+        agent = "You are a helpful AI assistant. You are a web developer."
+    elif agent_name == "AI_SYSTEM_PROMPT":
+        agent = "You are a helpful AI assistant. You are an AI system."
+    elif agent_name == "PYTHON_CODE_DEV":
+        agent = "You are a helpful AI assistant. You are a Python code developer."
+    else:
+        agent = "You are a helpful AI assistant."
+
+    system_prompt = f"{agent} {sys_prompt}".strip()
+    
+    temperature = max(float(temperature), 1e-2)
+    top_p = float(top_p)
+
+    # Add the system prompt to the beginning of the prompt
+    formatted_prompt = f"{system_prompt} {prompt}"
+
     # Use 'prompt' here instead of 'message'
-    formatted_prompt = format_prompt(prompt, history, max_history_turns=5)  # Truncated history
+    formatted_prompt = format_prompt(formatted_prompt, history, max_history_turns=5)  # Truncated history
     logging.info(f"Formatted Prompt: {formatted_prompt}")
-    stream = client.text_generation(formatted_prompt, temperature=temperature, max_new_tokens=max_new_tokens, top_p=top_p, repetition_penalty=repetition_penalty, stream=True, details=True, return_full_text=False)
+    
+    # Use the specified model if it's different from the default
+    client = InferenceClient(model) if model != "mistralai/Mixtral-8x7B-Instruct-v0.1" else InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+    
+    stream = client.text_generation(
+        formatted_prompt, 
+        temperature=temperature, 
+        max_new_tokens=max_new_tokens, 
+        top_p=top_p, 
+        repetition_penalty=repetition_penalty, 
+        stream=True, 
+        details=True, 
+        return_full_text=False
+    )
     resp = ""
     for response in stream:
         resp += response.token.text
+        yield resp  # This allows for streaming the response
 
     if VERBOSE:
-        logging.info(LOG_RESPONSE.format(resp))  # Log the response
-    return resp  # Return resp here
+        logging.info(f"RESPONSE: {resp}")  # Log the response directly
+
+    return resp  # Return the full response at the end
 
 def compress_history(purpose, task, history, directory):
     resp = run_gpt(