Significant-Gravitas
/
AutoGPT
mirror of https://github.com/Significant-Gravitas/AutoGPT.git
1
0
Fork 1
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/master'

pull/1381/head
Jedakiah 2023-04-14 20:18:04 +02:00
parent d0dd107f39 a17a850b25
commit 7daa3fc8f9
12 changed files with 321 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
.env.template
View File

@ -3,6 +3,14 @@
################################################################################
# EXECUTE_LOCAL_COMMANDS - Allow local command execution (Example: False)
EXECUTE_LOCAL_COMMANDS=False
# BROWSE_CHUNK_MAX_LENGTH - When browsing website, define the length of chunk stored in memory
BROWSE_CHUNK_MAX_LENGTH=8192
# BROWSE_SUMMARY_MAX_TOKEN - Define the maximum length of the summary generated by GPT agent when browsing website
BROWSE_SUMMARY_MAX_TOKEN=300
# USER_AGENT - Define the user-agent used by the requests library to browse website (string)
# USER_AGENT="Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/83.0.4103.97 Safari/537.36"
# AI_SETTINGS_FILE - Specifies which AI Settings file to use (defaults to ai_settings.yaml)
AI_SETTINGS_FILE=ai_settings.yaml
################################################################################
### LLM PROVIDER

1
.gitignore vendored
View File

@ -16,6 +16,7 @@ last_run_ai_settings.yaml
.idea/*
auto-gpt.json
log.txt
log-ingestion.txt
logs
# Coverage reports

55
README.md
View File

@ -46,6 +46,7 @@ Your support is greatly appreciated
- [Setting up environment variables](#setting-up-environment-variables-1)
- [Setting Your Cache Type](#setting-your-cache-type)
- [View Memory Usage](#view-memory-usage)
- [🧠 Memory pre-seeding](#memory-pre-seeding)
- [💀 Continuous Mode ⚠️](#-continuous-mode-️)
- [GPT3.5 ONLY Mode](#gpt35-only-mode)
- [🖼 Image Generation](#-image-generation)
@ -137,6 +138,14 @@ To output debug logs:
```
python scripts/main.py --debug
```
### Command Line Arguments
Here are some common arguments you can use when running Auto-GPT:
> Replace anything in angled brackets (<>) to a value you want to specify
* `python scripts/main.py --help` to see a list of all available command line arguments.
* `python scripts/main.py --ai-settings <filename>` to run Auto-GPT with a different AI Settings file.
* `python scripts/main.py --use-memory <memory-backend>` to specify one of 3 memory backends: `local`, `redis`, `pinecone` or 'no_memory'.
> **NOTE**: There are shorthands for some of these flags, for example `-m` for `--use-memory`. Use `python scripts/main.py --help` for more information
## 🗣️ Speech Mode
@ -266,6 +275,52 @@ To switch to either, change the `MEMORY_BACKEND` env variable to the value that
1. View memory usage by using the `--debug` flag :)
## 🧠 Memory pre-seeding
```
# python scripts/data_ingestion.py -h
usage: data_ingestion.py [-h] (--file FILE | --dir DIR) [--init] [--overlap OVERLAP] [--max_length MAX_LENGTH]
Ingest a file or a directory with multiple files into memory. Make sure to set your .env before running this script.
options:
-h, --help show this help message and exit
--file FILE The file to ingest.
--dir DIR The directory containing the files to ingest.
--init Init the memory and wipe its content (default: False)
--overlap OVERLAP The overlap size between chunks when ingesting files (default: 200)
--max_length MAX_LENGTH The max_length of each chunk when ingesting files (default: 4000
# python scripts/data_ingestion.py --dir seed_data --init --overlap 200 --max_length 1000
```
This script located at scripts/data_ingestion.py, allows you to ingest files into memory and pre-seed it before running Auto-GPT.
Memory pre-seeding is a technique that involves ingesting relevant documents or data into the AI's memory so that it can use this information to generate more informed and accurate responses.
To pre-seed the memory, the content of each document is split into chunks of a specified maximum length with a specified overlap between chunks, and then each chunk is added to the memory backend set in the .env file. When the AI is prompted to recall information, it can then access those pre-seeded memories to generate more informed and accurate responses.
This technique is particularly useful when working with large amounts of data or when there is specific information that the AI needs to be able to access quickly.
By pre-seeding the memory, the AI can retrieve and use this information more efficiently, saving time, API call and improving the accuracy of its responses.
You could for example download the documentation of an API, a Github repository, etc. and ingest it into memory before running Auto-GPT.
⚠️ If you use Redis as your memory, make sure to run Auto-GPT with the WIPE_REDIS_ON_START set to False in your .env file.
⚠️For other memory backend, we currently forcefully wipe the memory when starting Auto-GPT. To ingest data with those memory backend, you can call the data_ingestion.py script anytime during an Auto-GPT run.
Memories will be available to the AI immediately as they are ingested, even if ingested while Auto-GPT is running.
In the example above, the script initializes the memory, ingests all files within the seed_data directory into memory with an overlap between chunks of 200 and a maximum length of each chunk of 4000.
Note that you can also use the --file argument to ingest a single file into memory and that the script will only ingest files within the auto_gpt_workspace directory.
You can adjust the max_length and overlap parameters to fine-tune the way the docuents are presented to the AI when it "recall" that memory:
- Adjusting the overlap value allows the AI to access more contextual information from each chunk when recalling information, but will result in more chunks being created and therefore increase memory backend usage and OpenAI API requests.
- Reducing the max_length value will create more chunks, which can save prompt tokens by allowing for more message history in the context, but will also increase the number of chunks.
- Increasing the max_length value will provide the AI with more contextual information from each chunk, reducing the number of chunks created and saving on OpenAI API requests. However, this may also use more prompt tokens and decrease the overall context available to the AI.
## 💀 Continuous Mode ⚠️
Run the AI **without** user authorisation, 100% automated.

30
scripts/browse.py
View File

@ -1,10 +1,15 @@
import requests
from bs4 import BeautifulSoup
from memory import get_memory
from config import Config
from llm_utils import create_chat_completion
from urllib.parse import urlparse, urljoin
cfg = Config()
memory = get_memory(cfg)
session = requests.Session()
session.headers.update({'User-Agent': cfg.user_agent})
# Function to check if the URL is valid
@ -27,7 +32,7 @@ def check_local_file_access(url):
return any(url.startswith(prefix) for prefix in local_prefixes)
def get_response(url, headers=cfg.user_agent_header, timeout=10):
def get_response(url, timeout=10):
try:
# Restrict access to local files
if check_local_file_access(url):
@ -39,7 +44,7 @@ def get_response(url, headers=cfg.user_agent_header, timeout=10):
sanitized_url = sanitize_url(url)
response = requests.get(sanitized_url, headers=headers, timeout=timeout)
response = session.get(sanitized_url, timeout=timeout)
# Check if the response contains an HTTP error
if response.status_code >= 400:
@ -106,7 +111,7 @@ def scrape_links(url):
return format_hyperlinks(hyperlinks)
def split_text(text, max_length=8192):
def split_text(text, max_length=cfg.browse_chunk_max_length):
"""Split text into chunks of a maximum length"""
paragraphs = text.split("\n")
current_length = 0
@ -133,7 +138,7 @@ def create_message(chunk, question):
}
def summarize_text(text, question):
def summarize_text(url, text, question):
"""Summarize text using the LLM model"""
if not text:
return "Error: No text to summarize"
@ -145,15 +150,28 @@ def summarize_text(text, question):
chunks = list(split_text(text))
for i, chunk in enumerate(chunks):
print(f"Adding chunk {i + 1} / {len(chunks)} to memory")
memory_to_add = f"Source: {url}\n" \
f"Raw content part#{i + 1}: {chunk}"
memory.add(memory_to_add)
print(f"Summarizing chunk {i + 1} / {len(chunks)}")
messages = [create_message(chunk, question)]
summary = create_chat_completion(
model=cfg.fast_llm_model,
messages=messages,
max_tokens=300,
max_tokens=cfg.browse_summary_max_token,
)
summaries.append(summary)
print(f"Added chunk {i + 1} summary to memory")
memory_to_add = f"Source: {url}\n" \
f"Content summary part#{i + 1}: {summary}"
memory.add(memory_to_add)
print(f"Summarized {len(chunks)} chunks.")
@ -163,7 +181,7 @@ def summarize_text(text, question):
final_summary = create_chat_completion(
model=cfg.fast_llm_model,
messages=messages,
max_tokens=300,
max_tokens=cfg.browse_summary_max_token,
)
return final_summary

2
scripts/commands.py
View File

@ -194,7 +194,7 @@ def browse_website(url, question):
def get_text_summary(url, question):
"""Return the results of a google search"""
text = browse.scrape_text(url)
summary = browse.summarize_text(text, question)
summary = browse.summarize_text(url, text, question)
return """ "Result" : """ + summary

14
scripts/config.py
View File

@ -38,11 +38,15 @@ class Config(metaclass=Singleton):
self.continuous_mode = False
self.continuous_limit = 0
self.speak_mode = False
self.skip_reprompt = False
self.ai_settings_file = os.getenv("AI_SETTINGS_FILE", "ai_settings.yaml")
self.fast_llm_model = os.getenv("FAST_LLM_MODEL", "gpt-3.5-turbo")
self.smart_llm_model = os.getenv("SMART_LLM_MODEL", "gpt-4")
self.fast_token_limit = int(os.getenv("FAST_TOKEN_LIMIT", 4000))
self.smart_token_limit = int(os.getenv("SMART_TOKEN_LIMIT", 8000))
self.browse_chunk_max_length = int(os.getenv("BROWSE_CHUNK_MAX_LENGTH", 8192))
self.browse_summary_max_token = int(os.getenv("BROWSE_SUMMARY_MAX_TOKEN", 300))
self.openai_api_key = os.getenv("OPENAI_API_KEY")
self.temperature = float(os.getenv("TEMPERATURE", "1"))
@ -76,7 +80,7 @@ class Config(metaclass=Singleton):
# User agent headers to use when browsing web
# Some websites might just completely deny request with an error code if no user agent was found.
self.user_agent_header = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/83.0.4103.97 Safari/537.36"}
self.user_agent = os.getenv("USER_AGENT", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/83.0.4103.97 Safari/537.36")
self.redis_host = os.getenv("REDIS_HOST", "localhost")
self.redis_port = os.getenv("REDIS_PORT", "6379")
self.redis_password = os.getenv("REDIS_PASSWORD", "")
@ -157,6 +161,14 @@ class Config(metaclass=Singleton):
"""Set the smart token limit value."""
self.smart_token_limit = value
def set_browse_chunk_max_length(self, value: int):
"""Set the browse_website command chunk max length value."""
self.browse_chunk_max_length = value
def set_browse_summary_max_token(self, value: int):
"""Set the browse_website command summary max token value."""
self.browse_summary_max_token = value
def set_openai_api_key(self, value: str):
"""Set the OpenAI API key value."""
self.openai_api_key = value

70
scripts/data_ingestion.py Normal file
View File

@ -0,0 +1,70 @@
import argparse
import logging
from config import Config
from memory import get_memory
from file_operations import ingest_file, search_files
cfg = Config()
def configure_logging():
logging.basicConfig(filename='log-ingestion.txt',
filemode='a',
format='%(asctime)s,%(msecs)d %(name)s %(levelname)s %(message)s',
datefmt='%H:%M:%S',
level=logging.DEBUG)
return logging.getLogger('AutoGPT-Ingestion')
def ingest_directory(directory, memory, args):
"""
Ingest all files in a directory by calling the ingest_file function for each file.
:param directory: The directory containing the files to ingest
:param memory: An object with an add() method to store the chunks in memory
"""
try:
files = search_files(directory)
for file in files:
ingest_file(file, memory, args.max_length, args.overlap)
except Exception as e:
print(f"Error while ingesting directory '{directory}': {str(e)}")
def main():
logger = configure_logging()
parser = argparse.ArgumentParser(description="Ingest a file or a directory with multiple files into memory. Make sure to set your .env before running this script.")
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument("--file", type=str, help="The file to ingest.")
group.add_argument("--dir", type=str, help="The directory containing the files to ingest.")
parser.add_argument("--init", action='store_true', help="Init the memory and wipe its content (default: False)", default=False)
parser.add_argument("--overlap", type=int, help="The overlap size between chunks when ingesting files (default: 200)", default=200)
parser.add_argument("--max_length", type=int, help="The max_length of each chunk when ingesting files (default: 4000)", default=4000)
args = parser.parse_args()
# Initialize memory
memory = get_memory(cfg, init=args.init)
print('Using memory of type: ' + memory.__class__.__name__)
if args.file:
try:
ingest_file(args.file, memory, args.max_length, args.overlap)
print(f"File '{args.file}' ingested successfully.")
except Exception as e:
logger.error(f"Error while ingesting file '{args.file}': {str(e)}")
print(f"Error while ingesting file '{args.file}': {str(e)}")
elif args.dir:
try:
ingest_directory(args.dir, memory, args)
print(f"Directory '{args.dir}' ingested successfully.")
except Exception as e:
logger.error(f"Error while ingesting directory '{args.dir}': {str(e)}")
print(f"Error while ingesting directory '{args.dir}': {str(e)}")
else:
print("Please provide either a file path (--file) or a directory name (--dir) inside the auto_gpt_workspace directory as input.")
if __name__ == "__main__":
main()

93
scripts/execute_code.py
View File

@ -19,53 +19,60 @@ def execute_python_file(file):
if not os.path.isfile(file_path):
return f"Error: File '{file}' does not exist."
try:
client = docker.from_env()
image_name = 'python:3.10'
if we_are_running_in_a_docker_container():
result = subprocess.run(f'python {file_path}', capture_output=True, encoding="utf8", shell=True)
if result.returncode == 0:
return result.stdout
else:
return f"Error: {result.stderr}"
else:
try:
client.images.get(image_name)
print(f"Image '{image_name}' found locally")
except docker.errors.ImageNotFound:
print(f"Image '{image_name}' not found locally, pulling from Docker Hub")
# Use the low-level API to stream the pull response
low_level_client = docker.APIClient()
for line in low_level_client.pull(image_name, stream=True, decode=True):
# Print the status and progress, if available
status = line.get('status')
progress = line.get('progress')
if status and progress:
print(f"{status}: {progress}")
elif status:
print(status)
client = docker.from_env()
# You can replace 'python:3.8' with the desired Python image/version
# You can find available Python images on Docker Hub:
# https://hub.docker.com/_/python
container = client.containers.run(
image_name,
f'python {file}',
volumes={
os.path.abspath(WORKSPACE_FOLDER): {
'bind': '/workspace',
'mode': 'ro'}},
working_dir='/workspace',
stderr=True,
stdout=True,
detach=True,
)
image_name = 'python:3.10'
try:
client.images.get(image_name)
print(f"Image '{image_name}' found locally")
except docker.errors.ImageNotFound:
print(f"Image '{image_name}' not found locally, pulling from Docker Hub")
# Use the low-level API to stream the pull response
low_level_client = docker.APIClient()
for line in low_level_client.pull(image_name, stream=True, decode=True):
# Print the status and progress, if available
status = line.get('status')
progress = line.get('progress')
if status and progress:
print(f"{status}: {progress}")
elif status:
print(status)
output = container.wait()
logs = container.logs().decode('utf-8')
container.remove()
# You can replace 'python:3.8' with the desired Python image/version
# You can find available Python images on Docker Hub:
# https://hub.docker.com/_/python
container = client.containers.run(
image_name,
f'python {file}',
volumes={
os.path.abspath(WORKSPACE_FOLDER): {
'bind': '/workspace',
'mode': 'ro'}},
working_dir='/workspace',
stderr=True,
stdout=True,
detach=True,
)
# print(f"Execution complete. Output: {output}")
# print(f"Logs: {logs}")
output = container.wait()
logs = container.logs().decode('utf-8')
container.remove()
return logs
# print(f"Execution complete. Output: {output}")
# print(f"Logs: {logs}")
except Exception as e:
return f"Error: {str(e)}"
return logs
except Exception as e:
return f"Error: {str(e)}"
def execute_shell(command_line):
@ -86,3 +93,7 @@ def execute_shell(command_line):
os.chdir(current_dir)
return output
def we_are_running_in_a_docker_container():
os.path.exists('/.dockerenv')

54
scripts/file_operations.py
View File

@ -20,6 +20,29 @@ def safe_join(base, *paths):
return norm_new_path
def split_file(content, max_length=4000, overlap=0):
"""
Split text into chunks of a specified maximum length with a specified overlap
between chunks.
:param text: The input text to be split into chunks
:param max_length: The maximum length of each chunk, default is 4000 (about 1k token)
:param overlap: The number of overlapping characters between chunks, default is no overlap
:return: A generator yielding chunks of text
"""
start = 0
content_length = len(content)
while start < content_length:
end = start + max_length
if end + overlap < content_length:
chunk = content[start:end+overlap]
else:
chunk = content[start:content_length]
yield chunk
start += max_length - overlap
def read_file(filename):
"""Read a file and return the contents"""
try:
@ -31,6 +54,37 @@ def read_file(filename):
return "Error: " + str(e)
def ingest_file(filename, memory, max_length=4000, overlap=200):
"""
Ingest a file by reading its content, splitting it into chunks with a specified
maximum length and overlap, and adding the chunks to the memory storage.
:param filename: The name of the file to ingest
:param memory: An object with an add() method to store the chunks in memory
:param max_length: The maximum length of each chunk, default is 4000
:param overlap: The number of overlapping characters between chunks, default is 200
"""
try:
print(f"Working with file {filename}")
content = read_file(filename)
content_length = len(content)
print(f"File length: {content_length} characters")
chunks = list(split_file(content, max_length=max_length, overlap=overlap))
num_chunks = len(chunks)
for i, chunk in enumerate(chunks):
print(f"Ingesting chunk {i + 1} / {num_chunks} into memory")
memory_to_add = f"Filename: {filename}\n" \
f"Content part#{i + 1}/{num_chunks}: {chunk}"
memory.add(memory_to_add)
print(f"Done ingesting {num_chunks} chunks from {filename}.")
except Exception as e:
print(f"Error while ingesting file '{filename}': {str(e)}")
def write_to_file(filename, text):
"""Write text to a file"""
try:

30
scripts/main.py
View File

@ -131,8 +131,12 @@ def print_assistant_thoughts(assistant_reply):
def construct_prompt():
"""Construct the prompt for the AI to respond to"""
config = AIConfig.load()
if config.ai_name:
config = AIConfig.load(cfg.ai_settings_file)
if cfg.skip_reprompt and config.ai_name:
logger.typewriter_log("Name :", Fore.GREEN, config.ai_name)
logger.typewriter_log("Role :", Fore.GREEN, config.ai_role)
logger.typewriter_log("Goals:", Fore.GREEN, config.ai_goals)
elif config.ai_name:
logger.typewriter_log(
f"Welcome back! ",
Fore.GREEN,
@ -220,13 +224,15 @@ def parse_arguments():
cfg.set_speak_mode(False)
parser = argparse.ArgumentParser(description='Process arguments.')
parser.add_argument('--continuous', action='store_true', help='Enable Continuous Mode')
parser.add_argument('--continuous', '-c', action='store_true', help='Enable Continuous Mode')
parser.add_argument('--continuous-limit', '-l', type=int, dest="continuous_limit", help='Defines the number of times to run in continuous mode')
parser.add_argument('--speak', action='store_true', help='Enable Speak Mode')
parser.add_argument('--debug', action='store_true', help='Enable Debug Mode')
parser.add_argument('--gpt3only', action='store_true', help='Enable GPT3.5 Only Mode')
parser.add_argument('--gpt4only', action='store_true', help='Enable GPT4 Only Mode')
parser.add_argument('--use-memory', '-m', dest="memory_type", help='Defines which Memory backend to use')
parser.add_argument('--skip-reprompt', '-y', dest='skip_reprompt', action='store_true', help='Skips the re-prompting messages at the beginning of the script')
parser.add_argument('--ai-settings', '-C', dest='ai_settings_file', help="Specifies which ai_settings.yaml file to use, will also automatically skip the re-prompt.")
args = parser.parse_args()
if args.debug:
@ -273,6 +279,24 @@ def parse_arguments():
else:
cfg.memory_backend = chosen
if args.skip_reprompt:
logger.typewriter_log("Skip Re-prompt: ", Fore.GREEN, "ENABLED")
cfg.skip_reprompt = True
if args.ai_settings_file:
file = args.ai_settings_file
# Validate file
(validated, message) = utils.validate_yaml_file(file)
if not validated:
logger.typewriter_log("FAILED FILE VALIDATION", Fore.RED, message)
logger.double_check()
exit(1)
logger.typewriter_log("Using AI Settings File:", Fore.GREEN, file)
cfg.ai_settings_file = file
cfg.skip_reprompt = True
def main():
global ai_name, memory

2
scripts/memory/__init__.py
View File

@ -3,7 +3,7 @@ from memory.no_memory import NoMemory
# List of supported memory backends
# Add a backend to this list if the import attempt is successful
supported_memory = ['local']
supported_memory = ['local', 'no_memory']
try:
from memory.redismem import RedisMemory

15
scripts/utils.py
View File

@ -1,3 +1,7 @@
import yaml
from colorama import Fore
def clean_input(prompt: str=''):
try:
return input(prompt)
@ -6,3 +10,14 @@ def clean_input(prompt: str=''):
print("Quitting...")
exit(0)
def validate_yaml_file(file: str):
try:
with open(file) as file:
yaml.load(file, Loader=yaml.FullLoader)
except FileNotFoundError:
return (False, f"The file {Fore.CYAN}`{file}`{Fore.RESET} wasn't found")
except yaml.YAMLError as e:
return (False, f"There was an issue while trying to read with your AI Settings file: {e}")
return (True, f"Successfully validated {Fore.CYAN}`{file}`{Fore.RESET}!")