Files
gpt_academic/autogpt/agent/agent.py
2023-05-30 15:44:39 +08:00

363 lines
15 KiB
Python

import signal
import sys
from datetime import datetime
from colorama import Fore, Style
from autogpt.app import execute_command, get_command
from autogpt.commands.command import CommandRegistry
from autogpt.config import Config
from autogpt.config.ai_config import AIConfig
from autogpt.json_utils.json_fix_llm import fix_json_using_multiple_techniques
from autogpt.json_utils.utilities import LLM_DEFAULT_RESPONSE_FORMAT, validate_json
from autogpt.llm.base import ChatSequence
from autogpt.llm.chat import chat_with_ai, create_chat_completion
from autogpt.llm.utils import count_string_tokens
from autogpt.log_cycle.log_cycle import (
FULL_MESSAGE_HISTORY_FILE_NAME,
NEXT_ACTION_FILE_NAME,
PROMPT_SUPERVISOR_FEEDBACK_FILE_NAME,
SUPERVISOR_FEEDBACK_FILE_NAME,
USER_INPUT_FILE_NAME,
LogCycleHandler,
)
from autogpt.logs import logger, print_assistant_thoughts
from autogpt.memory.message_history import MessageHistory
from autogpt.memory.vector import VectorMemory
from autogpt.speech import say_text
from autogpt.spinner import Spinner
from autogpt.utils import clean_input
from autogpt.workspace import Workspace
class Agent:
"""Agent class for interacting with Auto-GPT.
Attributes:
ai_name: The name of the agent.
memory: The memory object to use.
next_action_count: The number of actions to execute.
system_prompt: The system prompt is the initial prompt that defines everything
the AI needs to know to achieve its task successfully.
Currently, the dynamic and customizable information in the system prompt are
ai_name, description and goals.
triggering_prompt: The last sentence the AI will see before answering.
For Auto-GPT, this prompt is:
Determine which next command to use, and respond using the format specified
above:
The triggering prompt is not part of the system prompt because between the
system prompt and the triggering
prompt we have contextual information that can distract the AI and make it
forget that its goal is to find the next task to achieve.
SYSTEM PROMPT
CONTEXTUAL INFORMATION (memory, previous conversations, anything relevant)
TRIGGERING PROMPT
The triggering prompt reminds the AI about its short term meta task
(defining the next task)
"""
def __init__(
self,
ai_name: str,
memory: VectorMemory,
next_action_count: int,
command_registry: CommandRegistry,
config: AIConfig,
system_prompt: str,
triggering_prompt: str,
workspace_directory: str,
):
cfg = Config()
self.ai_name = ai_name
self.memory = memory
self.history = MessageHistory(self)
self.next_action_count = next_action_count
self.command_registry = command_registry
self.config = config
self.system_prompt = system_prompt
self.triggering_prompt = triggering_prompt
self.workspace = Workspace(workspace_directory, cfg.restrict_to_workspace)
self.created_at = datetime.now().strftime("%Y%m%d_%H%M%S")
self.cycle_count = 0
self.log_cycle_handler = LogCycleHandler()
def start_interaction_loop(self):
# Interaction Loop
cfg = Config()
self.cycle_count = 0
command_name = None
arguments = None
user_input = ""
# Signal handler for interrupting y -N
def signal_handler(signum, frame):
if self.next_action_count == 0:
sys.exit()
else:
print(
Fore.RED
+ "Interrupt signal received. Stopping continuous command execution."
+ Style.RESET_ALL
)
self.next_action_count = 0
signal.signal(signal.SIGINT, signal_handler)
while True:
# Discontinue if continuous limit is reached
self.cycle_count += 1
self.log_cycle_handler.log_count_within_cycle = 0
self.log_cycle_handler.log_cycle(
self.config.ai_name,
self.created_at,
self.cycle_count,
[m.raw() for m in self.history],
FULL_MESSAGE_HISTORY_FILE_NAME,
)
if (
cfg.continuous_mode
and cfg.continuous_limit > 0
and self.cycle_count > cfg.continuous_limit
):
logger.typewriter_log(
"Continuous Limit Reached: ", Fore.YELLOW, f"{cfg.continuous_limit}"
)
break
# Send message to AI, get response
with Spinner("Thinking... ", plain_output=cfg.plain_output):
assistant_reply = chat_with_ai(
cfg,
self,
self.system_prompt,
self.triggering_prompt,
cfg.fast_token_limit,
cfg.fast_llm_model,
)
assistant_reply_json = fix_json_using_multiple_techniques(assistant_reply)
for plugin in cfg.plugins:
if not plugin.can_handle_post_planning():
continue
assistant_reply_json = plugin.post_planning(assistant_reply_json)
# Print Assistant thoughts
if assistant_reply_json != {}:
validate_json(assistant_reply_json, LLM_DEFAULT_RESPONSE_FORMAT)
# Get command name and arguments
try:
print_assistant_thoughts(
self.ai_name, assistant_reply_json, cfg.speak_mode
)
command_name, arguments = get_command(assistant_reply_json)
if cfg.speak_mode:
say_text(f"I want to execute {command_name}")
arguments = self._resolve_pathlike_command_args(arguments)
except Exception as e:
logger.error("Error: \n", str(e))
self.log_cycle_handler.log_cycle(
self.config.ai_name,
self.created_at,
self.cycle_count,
assistant_reply_json,
NEXT_ACTION_FILE_NAME,
)
logger.typewriter_log(
"NEXT ACTION: ",
Fore.CYAN,
f"COMMAND = {Fore.CYAN}{command_name}{Style.RESET_ALL} "
f"ARGUMENTS = {Fore.CYAN}{arguments}{Style.RESET_ALL}",
)
if not cfg.continuous_mode and self.next_action_count == 0:
# ### GET USER AUTHORIZATION TO EXECUTE COMMAND ###
# Get key press: Prompt the user to press enter to continue or escape
# to exit
self.user_input = ""
logger.info(
"Enter 'y' to authorise command, 'y -N' to run N continuous commands, 's' to run self-feedback commands, "
"'n' to exit program, or enter feedback for "
f"{self.ai_name}..."
)
while True:
if cfg.chat_messages_enabled:
console_input = clean_input("Waiting for your response...")
else:
console_input = clean_input(
Fore.MAGENTA + "Input:" + Style.RESET_ALL
)
if console_input.lower().strip() == cfg.authorise_key:
user_input = "GENERATE NEXT COMMAND JSON"
break
elif console_input.lower().strip() == "s":
logger.typewriter_log(
"-=-=-=-=-=-=-= THOUGHTS, REASONING, PLAN AND CRITICISM WILL NOW BE VERIFIED BY AGENT -=-=-=-=-=-=-=",
Fore.GREEN,
"",
)
thoughts = assistant_reply_json.get("thoughts", {})
self_feedback_resp = self.get_self_feedback(
thoughts, cfg.fast_llm_model
)
logger.typewriter_log(
f"SELF FEEDBACK: {self_feedback_resp}",
Fore.YELLOW,
"",
)
user_input = self_feedback_resp
command_name = "self_feedback"
break
elif console_input.lower().strip() == "":
logger.warn("Invalid input format.")
continue
elif console_input.lower().startswith(f"{cfg.authorise_key} -"):
try:
self.next_action_count = abs(
int(console_input.split(" ")[1])
)
user_input = "GENERATE NEXT COMMAND JSON"
except ValueError:
logger.warn(
"Invalid input format. Please enter 'y -n' where n is"
" the number of continuous tasks."
)
continue
break
elif console_input.lower() == cfg.exit_key:
user_input = "EXIT"
break
else:
user_input = console_input
command_name = "human_feedback"
self.log_cycle_handler.log_cycle(
self.config.ai_name,
self.created_at,
self.cycle_count,
user_input,
USER_INPUT_FILE_NAME,
)
break
if user_input == "GENERATE NEXT COMMAND JSON":
logger.typewriter_log(
"-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-=",
Fore.MAGENTA,
"",
)
elif user_input == "EXIT":
logger.info("Exiting...")
break
else:
# Print authorized commands left value
logger.typewriter_log(
f"{Fore.CYAN}AUTHORISED COMMANDS LEFT: {Style.RESET_ALL}{self.next_action_count}"
)
# Execute command
if command_name is not None and command_name.lower().startswith("error"):
result = f"Could not execute command: {arguments}"
elif command_name == "human_feedback":
result = f"Human feedback: {user_input}"
elif command_name == "self_feedback":
result = f"Self feedback: {user_input}"
else:
for plugin in cfg.plugins:
if not plugin.can_handle_pre_command():
continue
command_name, arguments = plugin.pre_command(
command_name, arguments
)
command_result = execute_command(
self.command_registry,
command_name,
arguments,
self.config.prompt_generator,
config=cfg,
)
result = f"Command {command_name} returned: " f"{command_result}"
result_tlength = count_string_tokens(
str(command_result), cfg.fast_llm_model
)
memory_tlength = count_string_tokens(
str(self.history.summary_message()), cfg.fast_llm_model
)
if result_tlength + memory_tlength + 600 > cfg.fast_token_limit:
result = f"Failure: command {command_name} returned too much output. \
Do not execute this command again with the same arguments."
for plugin in cfg.plugins:
if not plugin.can_handle_post_command():
continue
result = plugin.post_command(command_name, result)
if self.next_action_count > 0:
self.next_action_count -= 1
# Check if there's a result from the command append it to the message
# history
if result is not None:
self.history.add("system", result, "action_result")
logger.typewriter_log("SYSTEM: ", Fore.YELLOW, result)
else:
self.history.add("system", "Unable to execute command", "action_result")
logger.typewriter_log(
"SYSTEM: ", Fore.YELLOW, "Unable to execute command"
)
def _resolve_pathlike_command_args(self, command_args):
if "directory" in command_args and command_args["directory"] in {"", "/"}:
command_args["directory"] = str(self.workspace.root)
else:
for pathlike in ["filename", "directory", "clone_path"]:
if pathlike in command_args:
command_args[pathlike] = str(
self.workspace.get_path(command_args[pathlike])
)
return command_args
def get_self_feedback(self, thoughts: dict, llm_model: str) -> str:
"""Generates a feedback response based on the provided thoughts dictionary.
This method takes in a dictionary of thoughts containing keys such as 'reasoning',
'plan', 'thoughts', and 'criticism'. It combines these elements into a single
feedback message and uses the create_chat_completion() function to generate a
response based on the input message.
Args:
thoughts (dict): A dictionary containing thought elements like reasoning,
plan, thoughts, and criticism.
Returns:
str: A feedback response generated using the provided thoughts dictionary.
"""
ai_role = self.config.ai_role
feedback_prompt = f"Below is a message from me, an AI Agent, assuming the role of {ai_role}. whilst keeping knowledge of my slight limitations as an AI Agent Please evaluate my thought process, reasoning, and plan, and provide a concise paragraph outlining potential improvements. Consider adding or removing ideas that do not align with my role and explaining why, prioritizing thoughts based on their significance, or simply refining my overall thought process."
reasoning = thoughts.get("reasoning", "")
plan = thoughts.get("plan", "")
thought = thoughts.get("thoughts", "")
feedback_thoughts = thought + reasoning + plan
prompt = ChatSequence.for_model(llm_model)
prompt.add("user", feedback_prompt + feedback_thoughts)
self.log_cycle_handler.log_cycle(
self.config.ai_name,
self.created_at,
self.cycle_count,
prompt.raw(),
PROMPT_SUPERVISOR_FEEDBACK_FILE_NAME,
)
feedback = create_chat_completion(prompt)
self.log_cycle_handler.log_cycle(
self.config.ai_name,
self.created_at,
self.cycle_count,
feedback,
SUPERVISOR_FEEDBACK_FILE_NAME,
)
return feedback