Update app.py

app.py

@@ -21,12 +21,11 @@ from agent import (
 )
 from utils import parse_action, parse_file_content, read_python_module_structure
 from datetime import datetime
+
 now = datetime.now()
 date_time_str = now.strftime("%Y-%m-%d %H:%M:%S")
 
-client = InferenceClient(
-    "mistralai/Mixtral-8x7B-Instruct-v0.1",
-)
+client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
 ############################################
 
@@ -34,20 +33,14 @@ VERBOSE = True
 MAX_HISTORY = 125
 
 def format_prompt(message, history):
-  prompt = "<s>"
-  for user_prompt, bot_response in history:
-    prompt += f"[INST] {user_prompt} [/INST]"
-    prompt += f" {bot_response}</s> "
-  prompt += f"[INST] {message} [/INST]"
-  return prompt
-
-def run_gpt(
-    prompt_template,
-    stop_tokens,
-    max_tokens,
-    purpose,
-    **prompt_kwargs,
-):
+    prompt = "<s>"
+    for user_prompt, bot_response in history:
+        prompt += f"[INST] {user_prompt} [/INST]"
+        prompt += f" {bot_response}</s> "
+    prompt += f"[INST] {message} [/INST]"
+    return prompt
+
+def run_gpt(prompt_template, stop_tokens, max_tokens, purpose, **prompt_kwargs):
     seed = random.randint(1, 1111111111111111)
     print(seed)
     generate_kwargs = dict(
@@ -64,12 +57,10 @@ def run_gpt(
         purpose=purpose,
         safe_search=safe_search,
     ) + prompt_template.format(**prompt_kwargs)
+    
     if VERBOSE:
         print(LOG_PROMPT.format(content))
 
-    #formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
-    #formatted_prompt = format_prompt(f'{content}', history)
-
     stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
     resp = ""
     for response in stream:
@@ -77,6 +68,7 @@ def run_gpt(
 
     if VERBOSE:
         print(LOG_RESPONSE.format(resp))
+    
     return resp
 
 def compress_history(purpose, task, history, directory):
@@ -101,24 +93,25 @@ def call_search(purpose, task, history, directory, action_input):
                 action_input = action_input.strip(">")
 
             response = i_s(action_input)
-            #response = google(search_return)
             print(response)
             history += "observation: search result is: {}\n".format(response)
         else:
             history += "observation: I need to provide a valid URL to 'action: SEARCH action_input=https://URL'\n"
     except Exception as e:
         history += "observation: {}'\n".format(e)
+    
     return "MAIN", None, history, task
 
 def call_main(purpose, task, history, directory, action_input):
     resp = run_gpt(
         ACTION_PROMPT,
-        stop_tokens=["observation:", "task:", "action:","though:"],
+        stop_tokens=["observation:", "task:", "action:", "thought:"],
         max_tokens=5096,
         purpose=purpose,
         task=task,
         history=history,
     )
+    
     lines = resp.strip().strip("\n").split("\n")
     for line in lines:
         if line == "":
@@ -137,9 +130,7 @@ def call_main(purpose, task, history, directory, action_input):
                 return action_name, action_input, history, task
         else:
             history += "{}\n".format(line)
-            #history += "observation: the following command did not produce any useful output: '{}', I need to check the commands syntax, or use a different command\n".format(line)
-            #return action_name, action_input, history, task
-            #assert False, "unknown action: {}".format(line)
+    
     return "MAIN", None, history, task
 
 def call_set_task(purpose, task, history, directory, action_input):
@@ -151,6 +142,7 @@ def call_set_task(purpose, task, history, directory, action_input):
         task=task,
         history=history,
     ).strip("\n")
+    
     history += "observation: task has been updated to: {}\n".format(task)
     return "MAIN", None, history, task
 
@@ -178,10 +170,12 @@ def run_action(purpose, task, history, directory, action_name, action_input):
             if VERBOSE:
                 print("COMPRESSING HISTORY")
             history = compress_history(purpose, task, history, directory)
+        
         if not action_name in NAME_TO_FUNC:
             action_name = "MAIN"
         if action_name == "" or action_name is None:
             action_name = "MAIN"
+        
         assert action_name in NAME_TO_FUNC
 
         print("RUN: ", action_name, action_input)
@@ -218,22 +212,14 @@ def run(purpose, history):
             action_name,
             action_input,
         )
+        
         yield (history)
-        #yield ("",[(purpose,history)])
+        
         if task == "END":
             return (history)
-            #return ("", [(purpose,history)])
 
 ################################################
 
-def format_prompt(message, history):
-  prompt = "<s>"
-  for user_prompt, bot_response in history:
-    prompt += f"[INST] {user_prompt} [/INST]"
-    prompt += f" {bot_response}</s> "
-  prompt += f"[INST] {message} [/INST]"
-  return prompt
-
 agents = [
     "WEB_DEV",
     "AI_SYSTEM_PROMPT",
@@ -244,14 +230,15 @@ def generate(
         prompt, history, agent_name=agents[0], sys_prompt="", temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.7,
 ):
     seed = random.randint(1, 1111111111111111)
-
     agent = prompts.WEB_DEV
+
     if agent_name == "WEB_DEV":
         agent = prompts.WEB_DEV
-    if agent_name == "AI_SYSTEM_PROMPT":
+    elif agent_name == "AI_SYSTEM_PROMPT":
         agent = prompts.AI_SYSTEM_PROMPT
-    if agent_name == "PYTHON_CODE_DEV":
+    elif agent_name == "PYTHON_CODE_DEV":
         agent = prompts.PYTHON_CODE_DEV
+    
     system_prompt = agent
     temperature = float(temperature)
     if temperature < 1e-2:
@@ -274,6 +261,7 @@ def generate(
     for response in stream:
         output += response.token.text
         yield output
+    
     return output
 
 additional_inputs = [
@@ -297,7 +285,6 @@ additional_inputs = [
         interactive=True,
         info="Higher values produce more diverse outputs",
     ),
-
     gr.Slider(
         label="Max new tokens",
         value=1048 * 10,
@@ -328,13 +315,13 @@ additional_inputs = [
 ]
 
 examples = [
-    ["Based on previous interactions, generate an interactive preview of the user's requested application.", None, None, None, None, None, ],
-    ["Utilize the relevant code snippets and components from previous interactions.", None, None, None, None, None, ],
-    ["Assemble a working demo that showcases the core functionality of the application.", None, None, None, None, None, ],
-    ["Present the demo in an interactive environment within the Gradio interface.", None, None, None, None, None, ],
-    ["Allow the user to explore and interact with the demo to test its features.", None, None, None, None, None, ],
-    ["Gather feedback from the user about the demo and potential improvements.", None, None, None, None, None, ],
-    ["If the user approves of the app's running state you should provide a bash script that will automate all aspects of a local run and also a docker image for ease-of-launch in addition to the huggingface-ready app.py with all functions and gui and the requirements.txt file comprised of all required libraries and packages the application is dependent on, avoiding openai api at all points as we only use huggingface transformers, models, agents, libraries, api.", None, None, None, None, None, ],
+    ["Based on previous interactions, generate an interactive preview of the user's requested application.", None, None, None, None, None],
+    ["Utilize the relevant code snippets and components from previous interactions.", None, None, None, None, None],
+    ["Assemble a working demo that showcases the core functionality of the application.", None, None, None, None, None],
+    ["Present the demo in an interactive environment within the Gradio interface.", None, None, None, None, None],
+    ["Allow the user to explore and interact with the demo to test its features.", None, None, None, None, None],
+    ["Gather feedback from the user about the demo and potential improvements.", None, None, None, None, None],
+    ["If the user approves of the app's running state, provide a bash script that will automate all aspects of a local run and a docker image for ease-of-launch in addition to the huggingface-ready app.py with all functions and GUI, and the requirements.txt file comprised of all required libraries and packages the application is dependent on, avoiding OpenAI API at all points since we only use Hugging Face transformers, models, agents, libraries, and API.", None, None, None, None, None],
 ]
 
 def create_interface():
@@ -366,7 +353,9 @@ with iface:
     top_p = gr.Slider(label="Top-p (nucleus sampling)", value=0.90, minimum=0.0, maximum=1, step=0.05)
     repetition_penalty = gr.Slider(label="Repetition penalty", value=1.2, minimum=1.0, maximum=2.0, step=0.05)
 
-    msg.submit(chat_interface, [msg, chatbot, agent_dropdown, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty], [chatbot, msg])
+    msg.submit(chat_interface, 
+               [msg, chatbot, agent_dropdown, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty], 
+               [chatbot, msg])
     clear.click(lambda: None, None, chatbot, queue=False)
 
     gr.Examples(examples, [msg, agent_dropdown, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty])