Harrison/move experimental (#8084)

This commit is contained in:
Harrison Chase
2023-07-21 10:36:28 -07:00
committed by GitHub
parent f35db9f43e
commit da04760de1
50 changed files with 3519 additions and 28 deletions

View File

@@ -0,0 +1,4 @@
from langchain.experimental.autonomous_agents.autogpt.agent import AutoGPT
from langchain.experimental.autonomous_agents.baby_agi.baby_agi import BabyAGI
__all__ = ["BabyAGI", "AutoGPT"]

View File

@@ -0,0 +1,142 @@
from __future__ import annotations
from typing import List, Optional
from langchain.chains.llm import LLMChain
from langchain.chat_models.base import BaseChatModel
from langchain.experimental.autonomous_agents.autogpt.output_parser import (
AutoGPTOutputParser,
BaseAutoGPTOutputParser,
)
from langchain.experimental.autonomous_agents.autogpt.prompt import AutoGPTPrompt
from langchain.experimental.autonomous_agents.autogpt.prompt_generator import (
FINISH_NAME,
)
from langchain.memory import ChatMessageHistory
from langchain.schema import (
BaseChatMessageHistory,
Document,
)
from langchain.schema.messages import AIMessage, HumanMessage, SystemMessage
from langchain.tools.base import BaseTool
from langchain.tools.human.tool import HumanInputRun
from langchain.vectorstores.base import VectorStoreRetriever
from pydantic import ValidationError
class AutoGPT:
"""Agent class for interacting with Auto-GPT."""
def __init__(
self,
ai_name: str,
memory: VectorStoreRetriever,
chain: LLMChain,
output_parser: BaseAutoGPTOutputParser,
tools: List[BaseTool],
feedback_tool: Optional[HumanInputRun] = None,
chat_history_memory: Optional[BaseChatMessageHistory] = None,
):
self.ai_name = ai_name
self.memory = memory
self.next_action_count = 0
self.chain = chain
self.output_parser = output_parser
self.tools = tools
self.feedback_tool = feedback_tool
self.chat_history_memory = chat_history_memory or ChatMessageHistory()
@classmethod
def from_llm_and_tools(
cls,
ai_name: str,
ai_role: str,
memory: VectorStoreRetriever,
tools: List[BaseTool],
llm: BaseChatModel,
human_in_the_loop: bool = False,
output_parser: Optional[BaseAutoGPTOutputParser] = None,
chat_history_memory: Optional[BaseChatMessageHistory] = None,
) -> AutoGPT:
prompt = AutoGPTPrompt(
ai_name=ai_name,
ai_role=ai_role,
tools=tools,
input_variables=["memory", "messages", "goals", "user_input"],
token_counter=llm.get_num_tokens,
)
human_feedback_tool = HumanInputRun() if human_in_the_loop else None
chain = LLMChain(llm=llm, prompt=prompt)
return cls(
ai_name,
memory,
chain,
output_parser or AutoGPTOutputParser(),
tools,
feedback_tool=human_feedback_tool,
chat_history_memory=chat_history_memory,
)
def run(self, goals: List[str]) -> str:
user_input = (
"Determine which next command to use, "
"and respond using the format specified above:"
)
# Interaction Loop
loop_count = 0
while True:
# Discontinue if continuous limit is reached
loop_count += 1
# Send message to AI, get response
assistant_reply = self.chain.run(
goals=goals,
messages=self.chat_history_memory.messages,
memory=self.memory,
user_input=user_input,
)
# Print Assistant thoughts
print(assistant_reply)
self.chat_history_memory.add_message(HumanMessage(content=user_input))
self.chat_history_memory.add_message(AIMessage(content=assistant_reply))
# Get command name and arguments
action = self.output_parser.parse(assistant_reply)
tools = {t.name: t for t in self.tools}
if action.name == FINISH_NAME:
return action.args["response"]
if action.name in tools:
tool = tools[action.name]
try:
observation = tool.run(action.args)
except ValidationError as e:
observation = (
f"Validation Error in args: {str(e)}, args: {action.args}"
)
except Exception as e:
observation = (
f"Error: {str(e)}, {type(e).__name__}, args: {action.args}"
)
result = f"Command {tool.name} returned: {observation}"
elif action.name == "ERROR":
result = f"Error: {action.args}. "
else:
result = (
f"Unknown command '{action.name}'. "
f"Please refer to the 'COMMANDS' list for available "
f"commands and only respond in the specified JSON format."
)
memory_to_add = (
f"Assistant Reply: {assistant_reply} " f"\nResult: {result} "
)
if self.feedback_tool is not None:
feedback = f"\n{self.feedback_tool.run('Input: ')}"
if feedback in {"q", "stop"}:
print("EXITING")
return "EXITING"
memory_to_add += feedback
self.memory.add_documents([Document(page_content=memory_to_add)])
self.chat_history_memory.add_message(SystemMessage(content=result))

View File

@@ -0,0 +1,29 @@
from typing import Any, Dict, List
from langchain.memory.chat_memory import BaseChatMemory, get_prompt_input_key
from langchain.vectorstores.base import VectorStoreRetriever
from pydantic import Field
class AutoGPTMemory(BaseChatMemory):
retriever: VectorStoreRetriever = Field(exclude=True)
"""VectorStoreRetriever object to connect to."""
@property
def memory_variables(self) -> List[str]:
return ["chat_history", "relevant_context"]
def _get_prompt_input_key(self, inputs: Dict[str, Any]) -> str:
"""Get the input key for the prompt."""
if self.input_key is None:
return get_prompt_input_key(inputs, self.memory_variables)
return self.input_key
def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
input_key = self._get_prompt_input_key(inputs)
query = inputs[input_key]
docs = self.retriever.get_relevant_documents(query)
return {
"chat_history": self.chat_memory.messages[-10:],
"relevant_context": docs,
}

View File

@@ -0,0 +1,60 @@
import json
import re
from abc import abstractmethod
from typing import Dict, NamedTuple
from langchain.schema import BaseOutputParser
class AutoGPTAction(NamedTuple):
name: str
args: Dict
class BaseAutoGPTOutputParser(BaseOutputParser):
@abstractmethod
def parse(self, text: str) -> AutoGPTAction:
"""Return AutoGPTAction"""
def preprocess_json_input(input_str: str) -> str:
"""Preprocesses a string to be parsed as json.
Replace single backslashes with double backslashes,
while leaving already escaped ones intact.
Args:
input_str: String to be preprocessed
Returns:
Preprocessed string
"""
corrected_str = re.sub(
r'(?<!\\)\\(?!["\\/bfnrt]|u[0-9a-fA-F]{4})', r"\\\\", input_str
)
return corrected_str
class AutoGPTOutputParser(BaseAutoGPTOutputParser):
def parse(self, text: str) -> AutoGPTAction:
try:
parsed = json.loads(text, strict=False)
except json.JSONDecodeError:
preprocessed_text = preprocess_json_input(text)
try:
parsed = json.loads(preprocessed_text, strict=False)
except Exception:
return AutoGPTAction(
name="ERROR",
args={"error": f"Could not parse invalid json: {text}"},
)
try:
return AutoGPTAction(
name=parsed["command"]["name"],
args=parsed["command"]["args"],
)
except (KeyError, TypeError):
# If the command is null or incomplete, return an erroneous tool
return AutoGPTAction(
name="ERROR", args={"error": f"Incomplete command args: {parsed}"}
)

View File

@@ -0,0 +1,77 @@
import time
from typing import Any, Callable, List
from langchain.experimental.autonomous_agents.autogpt.prompt_generator import get_prompt
from langchain.prompts.chat import (
BaseChatPromptTemplate,
)
from langchain.schema.messages import BaseMessage, HumanMessage, SystemMessage
from langchain.tools.base import BaseTool
from langchain.vectorstores.base import VectorStoreRetriever
from pydantic import BaseModel
class AutoGPTPrompt(BaseChatPromptTemplate, BaseModel):
ai_name: str
ai_role: str
tools: List[BaseTool]
token_counter: Callable[[str], int]
send_token_limit: int = 4196
def construct_full_prompt(self, goals: List[str]) -> str:
prompt_start = (
"Your decisions must always be made independently "
"without seeking user assistance.\n"
"Play to your strengths as an LLM and pursue simple "
"strategies with no legal complications.\n"
"If you have completed all your tasks, make sure to "
'use the "finish" command.'
)
# Construct full prompt
full_prompt = (
f"You are {self.ai_name}, {self.ai_role}\n{prompt_start}\n\nGOALS:\n\n"
)
for i, goal in enumerate(goals):
full_prompt += f"{i+1}. {goal}\n"
full_prompt += f"\n\n{get_prompt(self.tools)}"
return full_prompt
def format_messages(self, **kwargs: Any) -> List[BaseMessage]:
base_prompt = SystemMessage(content=self.construct_full_prompt(kwargs["goals"]))
time_prompt = SystemMessage(
content=f"The current time and date is {time.strftime('%c')}"
)
used_tokens = self.token_counter(base_prompt.content) + self.token_counter(
time_prompt.content
)
memory: VectorStoreRetriever = kwargs["memory"]
previous_messages = kwargs["messages"]
relevant_docs = memory.get_relevant_documents(str(previous_messages[-10:]))
relevant_memory = [d.page_content for d in relevant_docs]
relevant_memory_tokens = sum(
[self.token_counter(doc) for doc in relevant_memory]
)
while used_tokens + relevant_memory_tokens > 2500:
relevant_memory = relevant_memory[:-1]
relevant_memory_tokens = sum(
[self.token_counter(doc) for doc in relevant_memory]
)
content_format = (
f"This reminds you of these events "
f"from your past:\n{relevant_memory}\n\n"
)
memory_message = SystemMessage(content=content_format)
used_tokens += self.token_counter(memory_message.content)
historical_messages: List[BaseMessage] = []
for message in previous_messages[-10:][::-1]:
message_tokens = self.token_counter(message.content)
if used_tokens + message_tokens > self.send_token_limit - 1000:
break
historical_messages = [message] + historical_messages
used_tokens += message_tokens
input_message = HumanMessage(content=kwargs["user_input"])
messages: List[BaseMessage] = [base_prompt, time_prompt, memory_message]
messages += historical_messages
messages.append(input_message)
return messages

View File

@@ -0,0 +1,186 @@
import json
from typing import List
from langchain.tools.base import BaseTool
FINISH_NAME = "finish"
class PromptGenerator:
"""A class for generating custom prompt strings.
Does this based on constraints, commands, resources, and performance evaluations.
"""
def __init__(self) -> None:
"""Initialize the PromptGenerator object.
Starts with empty lists of constraints, commands, resources,
and performance evaluations.
"""
self.constraints: List[str] = []
self.commands: List[BaseTool] = []
self.resources: List[str] = []
self.performance_evaluation: List[str] = []
self.response_format = {
"thoughts": {
"text": "thought",
"reasoning": "reasoning",
"plan": "- short bulleted\n- list that conveys\n- long-term plan",
"criticism": "constructive self-criticism",
"speak": "thoughts summary to say to user",
},
"command": {"name": "command name", "args": {"arg name": "value"}},
}
def add_constraint(self, constraint: str) -> None:
"""
Add a constraint to the constraints list.
Args:
constraint (str): The constraint to be added.
"""
self.constraints.append(constraint)
def add_tool(self, tool: BaseTool) -> None:
self.commands.append(tool)
def _generate_command_string(self, tool: BaseTool) -> str:
output = f"{tool.name}: {tool.description}"
output += f", args json schema: {json.dumps(tool.args)}"
return output
def add_resource(self, resource: str) -> None:
"""
Add a resource to the resources list.
Args:
resource (str): The resource to be added.
"""
self.resources.append(resource)
def add_performance_evaluation(self, evaluation: str) -> None:
"""
Add a performance evaluation item to the performance_evaluation list.
Args:
evaluation (str): The evaluation item to be added.
"""
self.performance_evaluation.append(evaluation)
def _generate_numbered_list(self, items: list, item_type: str = "list") -> str:
"""
Generate a numbered list from given items based on the item_type.
Args:
items (list): A list of items to be numbered.
item_type (str, optional): The type of items in the list.
Defaults to 'list'.
Returns:
str: The formatted numbered list.
"""
if item_type == "command":
command_strings = [
f"{i + 1}. {self._generate_command_string(item)}"
for i, item in enumerate(items)
]
finish_description = (
"use this to signal that you have finished all your objectives"
)
finish_args = (
'"response": "final response to let '
'people know you have finished your objectives"'
)
finish_string = (
f"{len(items) + 1}. {FINISH_NAME}: "
f"{finish_description}, args: {finish_args}"
)
return "\n".join(command_strings + [finish_string])
else:
return "\n".join(f"{i+1}. {item}" for i, item in enumerate(items))
def generate_prompt_string(self) -> str:
"""Generate a prompt string.
Returns:
str: The generated prompt string.
"""
formatted_response_format = json.dumps(self.response_format, indent=4)
prompt_string = (
f"Constraints:\n{self._generate_numbered_list(self.constraints)}\n\n"
f"Commands:\n"
f"{self._generate_numbered_list(self.commands, item_type='command')}\n\n"
f"Resources:\n{self._generate_numbered_list(self.resources)}\n\n"
f"Performance Evaluation:\n"
f"{self._generate_numbered_list(self.performance_evaluation)}\n\n"
f"You should only respond in JSON format as described below "
f"\nResponse Format: \n{formatted_response_format} "
f"\nEnsure the response can be parsed by Python json.loads"
)
return prompt_string
def get_prompt(tools: List[BaseTool]) -> str:
"""This function generates a prompt string.
It includes various constraints, commands, resources, and performance evaluations.
Returns:
str: The generated prompt string.
"""
# Initialize the PromptGenerator object
prompt_generator = PromptGenerator()
# Add constraints to the PromptGenerator object
prompt_generator.add_constraint(
"~4000 word limit for short term memory. "
"Your short term memory is short, "
"so immediately save important information to files."
)
prompt_generator.add_constraint(
"If you are unsure how you previously did something "
"or want to recall past events, "
"thinking about similar events will help you remember."
)
prompt_generator.add_constraint("No user assistance")
prompt_generator.add_constraint(
'Exclusively use the commands listed in double quotes e.g. "command name"'
)
# Add commands to the PromptGenerator object
for tool in tools:
prompt_generator.add_tool(tool)
# Add resources to the PromptGenerator object
prompt_generator.add_resource(
"Internet access for searches and information gathering."
)
prompt_generator.add_resource("Long Term memory management.")
prompt_generator.add_resource(
"GPT-3.5 powered Agents for delegation of simple tasks."
)
prompt_generator.add_resource("File output.")
# Add performance evaluations to the PromptGenerator object
prompt_generator.add_performance_evaluation(
"Continuously review and analyze your actions "
"to ensure you are performing to the best of your abilities."
)
prompt_generator.add_performance_evaluation(
"Constructively self-criticize your big-picture behavior constantly."
)
prompt_generator.add_performance_evaluation(
"Reflect on past decisions and strategies to refine your approach."
)
prompt_generator.add_performance_evaluation(
"Every command has a cost, so be smart and efficient. "
"Aim to complete tasks in the least number of steps."
)
# Generate the prompt string
prompt_string = prompt_generator.generate_prompt_string()
return prompt_string

View File

@@ -0,0 +1,17 @@
from langchain.experimental.autonomous_agents.baby_agi.baby_agi import BabyAGI
from langchain.experimental.autonomous_agents.baby_agi.task_creation import (
TaskCreationChain,
)
from langchain.experimental.autonomous_agents.baby_agi.task_execution import (
TaskExecutionChain,
)
from langchain.experimental.autonomous_agents.baby_agi.task_prioritization import (
TaskPrioritizationChain,
)
__all__ = [
"BabyAGI",
"TaskPrioritizationChain",
"TaskExecutionChain",
"TaskCreationChain",
]

View File

@@ -0,0 +1,202 @@
"""BabyAGI agent."""
from collections import deque
from typing import Any, Dict, List, Optional
from langchain.callbacks.manager import CallbackManagerForChainRun
from langchain.chains.base import Chain
from langchain.experimental.autonomous_agents.baby_agi.task_creation import (
TaskCreationChain,
)
from langchain.experimental.autonomous_agents.baby_agi.task_execution import (
TaskExecutionChain,
)
from langchain.experimental.autonomous_agents.baby_agi.task_prioritization import (
TaskPrioritizationChain,
)
from langchain.schema.language_model import BaseLanguageModel
from langchain.vectorstores.base import VectorStore
from pydantic import BaseModel, Field
class BabyAGI(Chain, BaseModel):
"""Controller model for the BabyAGI agent."""
task_list: deque = Field(default_factory=deque)
task_creation_chain: Chain = Field(...)
task_prioritization_chain: Chain = Field(...)
execution_chain: Chain = Field(...)
task_id_counter: int = Field(1)
vectorstore: VectorStore = Field(init=False)
max_iterations: Optional[int] = None
class Config:
"""Configuration for this pydantic object."""
arbitrary_types_allowed = True
def add_task(self, task: Dict) -> None:
self.task_list.append(task)
def print_task_list(self) -> None:
print("\033[95m\033[1m" + "\n*****TASK LIST*****\n" + "\033[0m\033[0m")
for t in self.task_list:
print(str(t["task_id"]) + ": " + t["task_name"])
def print_next_task(self, task: Dict) -> None:
print("\033[92m\033[1m" + "\n*****NEXT TASK*****\n" + "\033[0m\033[0m")
print(str(task["task_id"]) + ": " + task["task_name"])
def print_task_result(self, result: str) -> None:
print("\033[93m\033[1m" + "\n*****TASK RESULT*****\n" + "\033[0m\033[0m")
print(result)
@property
def input_keys(self) -> List[str]:
return ["objective"]
@property
def output_keys(self) -> List[str]:
return []
def get_next_task(
self, result: str, task_description: str, objective: str, **kwargs: Any
) -> List[Dict]:
"""Get the next task."""
task_names = [t["task_name"] for t in self.task_list]
incomplete_tasks = ", ".join(task_names)
response = self.task_creation_chain.run(
result=result,
task_description=task_description,
incomplete_tasks=incomplete_tasks,
objective=objective,
**kwargs,
)
new_tasks = response.split("\n")
return [
{"task_name": task_name} for task_name in new_tasks if task_name.strip()
]
def prioritize_tasks(
self, this_task_id: int, objective: str, **kwargs: Any
) -> List[Dict]:
"""Prioritize tasks."""
task_names = [t["task_name"] for t in list(self.task_list)]
next_task_id = int(this_task_id) + 1
response = self.task_prioritization_chain.run(
task_names=", ".join(task_names),
next_task_id=str(next_task_id),
objective=objective,
**kwargs,
)
new_tasks = response.split("\n")
prioritized_task_list = []
for task_string in new_tasks:
if not task_string.strip():
continue
task_parts = task_string.strip().split(".", 1)
if len(task_parts) == 2:
task_id = task_parts[0].strip()
task_name = task_parts[1].strip()
prioritized_task_list.append(
{"task_id": task_id, "task_name": task_name}
)
return prioritized_task_list
def _get_top_tasks(self, query: str, k: int) -> List[str]:
"""Get the top k tasks based on the query."""
results = self.vectorstore.similarity_search(query, k=k)
if not results:
return []
return [str(item.metadata["task"]) for item in results]
def execute_task(self, objective: str, task: str, k: int = 5, **kwargs: Any) -> str:
"""Execute a task."""
context = self._get_top_tasks(query=objective, k=k)
return self.execution_chain.run(
objective=objective, context="\n".join(context), task=task, **kwargs
)
def _call(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
"""Run the agent."""
_run_manager = run_manager or CallbackManagerForChainRun.get_noop_manager()
objective = inputs["objective"]
first_task = inputs.get("first_task", "Make a todo list")
self.add_task({"task_id": 1, "task_name": first_task})
num_iters = 0
while True:
if self.task_list:
self.print_task_list()
# Step 1: Pull the first task
task = self.task_list.popleft()
self.print_next_task(task)
# Step 2: Execute the task
result = self.execute_task(
objective, task["task_name"], callbacks=_run_manager.get_child()
)
this_task_id = int(task["task_id"])
self.print_task_result(result)
# Step 3: Store the result in Pinecone
result_id = f"result_{task['task_id']}"
self.vectorstore.add_texts(
texts=[result],
metadatas=[{"task": task["task_name"]}],
ids=[result_id],
)
# Step 4: Create new tasks and reprioritize task list
new_tasks = self.get_next_task(
result,
task["task_name"],
objective,
callbacks=_run_manager.get_child(),
)
for new_task in new_tasks:
self.task_id_counter += 1
new_task.update({"task_id": self.task_id_counter})
self.add_task(new_task)
self.task_list = deque(
self.prioritize_tasks(
this_task_id, objective, callbacks=_run_manager.get_child()
)
)
num_iters += 1
if self.max_iterations is not None and num_iters == self.max_iterations:
print(
"\033[91m\033[1m" + "\n*****TASK ENDING*****\n" + "\033[0m\033[0m"
)
break
return {}
@classmethod
def from_llm(
cls,
llm: BaseLanguageModel,
vectorstore: VectorStore,
verbose: bool = False,
task_execution_chain: Optional[Chain] = None,
**kwargs: Dict[str, Any],
) -> "BabyAGI":
"""Initialize the BabyAGI Controller."""
task_creation_chain = TaskCreationChain.from_llm(llm, verbose=verbose)
task_prioritization_chain = TaskPrioritizationChain.from_llm(
llm, verbose=verbose
)
if task_execution_chain is None:
execution_chain: Chain = TaskExecutionChain.from_llm(llm, verbose=verbose)
else:
execution_chain = task_execution_chain
return cls(
task_creation_chain=task_creation_chain,
task_prioritization_chain=task_prioritization_chain,
execution_chain=execution_chain,
vectorstore=vectorstore,
**kwargs,
)

View File

@@ -0,0 +1,30 @@
from langchain import LLMChain, PromptTemplate
from langchain.schema.language_model import BaseLanguageModel
class TaskCreationChain(LLMChain):
"""Chain to generates tasks."""
@classmethod
def from_llm(cls, llm: BaseLanguageModel, verbose: bool = True) -> LLMChain:
"""Get the response parser."""
task_creation_template = (
"You are an task creation AI that uses the result of an execution agent"
" to create new tasks with the following objective: {objective},"
" The last completed task has the result: {result}."
" This result was based on this task description: {task_description}."
" These are incomplete tasks: {incomplete_tasks}."
" Based on the result, create new tasks to be completed"
" by the AI system that do not overlap with incomplete tasks."
" Return the tasks as an array."
)
prompt = PromptTemplate(
template=task_creation_template,
input_variables=[
"result",
"task_description",
"incomplete_tasks",
"objective",
],
)
return cls(prompt=prompt, llm=llm, verbose=verbose)

View File

@@ -0,0 +1,21 @@
from langchain import LLMChain, PromptTemplate
from langchain.schema.language_model import BaseLanguageModel
class TaskExecutionChain(LLMChain):
"""Chain to execute tasks."""
@classmethod
def from_llm(cls, llm: BaseLanguageModel, verbose: bool = True) -> LLMChain:
"""Get the response parser."""
execution_template = (
"You are an AI who performs one task based on the following objective: "
"{objective}."
"Take into account these previously completed tasks: {context}."
" Your task: {task}. Response:"
)
prompt = PromptTemplate(
template=execution_template,
input_variables=["objective", "context", "task"],
)
return cls(prompt=prompt, llm=llm, verbose=verbose)

View File

@@ -0,0 +1,24 @@
from langchain import LLMChain, PromptTemplate
from langchain.schema.language_model import BaseLanguageModel
class TaskPrioritizationChain(LLMChain):
"""Chain to prioritize tasks."""
@classmethod
def from_llm(cls, llm: BaseLanguageModel, verbose: bool = True) -> LLMChain:
"""Get the response parser."""
task_prioritization_template = (
"You are a task prioritization AI tasked with cleaning the formatting of "
"and reprioritizing the following tasks: {task_names}."
" Consider the ultimate objective of your team: {objective}."
" Do not remove any tasks. Return the result as a numbered list, like:"
" #. First task"
" #. Second task"
" Start the task list with number {next_task_id}."
)
prompt = PromptTemplate(
template=task_prioritization_template,
input_variables=["task_names", "next_task_id", "objective"],
)
return cls(prompt=prompt, llm=llm, verbose=verbose)