Implementing personalization with Gorilla

.gitignore

@@ -7,3 +7,4 @@ dist
 .vscode
 .idea
 .editorconfig
+.env

go_cli.py

@@ -19,25 +19,32 @@
 import fcntl
 import platform
 import requests
+from openai import OpenAI
+import json
 import subprocess
 import argparse
 import termios
 import urllib.parse
 import sys
 from halo import Halo
 import go_questionary
+from personalization.personalize import GorillaPersonalizer
+from personalization.setup import PersonalizationSetup
 
-__version__ = "0.0.11" # current version
+
+__version__ = "0.0.12" # current version
 SERVER_URL = "https://cli.gorilla-llm.com"
 UPDATE_CHECK_FILE = os.path.expanduser("~/.gorilla-cli-last-update-check")
 USERID_FILE = os.path.expanduser("~/.gorilla-cli-userid")
 HISTORY_FILE = os.path.expanduser("~/.gorilla_cli_history")
+CONFIG_FILE = os.path.expanduser("~/.gorilla-cli-config.json")
 ISSUE_URL = f"https://github.com/gorilla-llm/gorilla-cli/issues/new"
 GORILLA_EMOJI = "🦍 " if go_questionary.try_encode_gorilla() else ""
 HISTORY_LENGTH = 10
 WELCOME_TEXT = f"""===***===
 {GORILLA_EMOJI}Welcome to Gorilla-CLI! Enhance your Command Line with the power of LLMs! 
 
+
 Simply use `gorilla <your desired operation>` and Gorilla will do the rest. For instance:
  gorilla generate 100 random characters into a file called test.txt
  gorilla get the image ids of all pods running in all namespaces in kubernetes
@@ -54,6 +61,13 @@
 def generate_random_uid():
  return str(uuid.uuid4())
 
+def load_config():
+ # Load the user's configuration file and perform any necessary checks
+ if os.path.isfile(CONFIG_FILE):
+ with open(CONFIG_FILE, "r") as config_file:
+ config_json = json.load(config_file)
+ return config_json
+
 def get_git_email():
  return subprocess.check_output(["git", "config", "--global", "user.email"]).decode("utf-8").strip()
 
@@ -122,8 +136,32 @@ def check_for_updates():
  except Exception as e:
  print("Unable to write update check file:", e)
 
+def setup_config_file(user_id):
+ config_json = { "user_id": user_id }
+ with open(CONFIG_FILE, "w") as config_file:
+ json.dump(config_json, config_file)
 
 def get_user_id():
+ if os.path.isfile(CONFIG_FILE):
+ with open(CONFIG_FILE, "r") as config_file:
+ try:
+ config_json = json.load(config_file)
+ if "user_id" not in config_json:
+ user_id = get_user_id_deprecated()
+ setup_config_file(user_id)
+ return get_user_id_deprecated()
+ else: 
+ return config_json["user_id"] 
+ except:
+ user_id = get_user_id_deprecated()
+ setup_config_file(user_id)
+ return get_user_id_deprecated
+ else:
+ user_id = get_user_id_deprecated()
+ setup_config_file(user_id)
+ return get_user_id_deprecated()
+
+def get_user_id_deprecated():
  # Unique user identifier for authentication and load balancing
  # Gorilla-CLI is hosted by UC Berkeley Sky lab for FREE as a
  # research prototype. Please don't spam the system or use it
@@ -184,6 +222,9 @@ def append_string_to_file_if_missing(file_path, target_string):
  file.write(target_string)
 
 
+
+
+
 def main():
  def execute_command(cmd):
  cmd = format_command(cmd)
@@ -220,6 +261,13 @@ def get_history_commands(history_file):
  user_id = get_user_id()
  system_info = get_system_info()
 
+ personalization_setup = PersonalizationSetup()
+ # personalization_setup.request_personalization()
+ personalization = personalization_setup.personalization
+ open_ai_key = personalization_setup.open_ai_key
+
+ personalized_history = None
+
 
  # Parse command-line arguments
  parser = argparse.ArgumentParser(description="Gorilla CLI Help Doc")
@@ -231,13 +279,24 @@ def get_history_commands(history_file):
  # Generate a unique interaction ID
  interaction_id = str(uuid.uuid4())
 
+
+
+ commands = []
  if args.history:
  commands = get_history_commands(HISTORY_FILE)
+
+ if personalization:
+ personalizer = GorillaPersonalizer(open_ai_key)
+ personalized_history = personalizer.personalize(user_input, commands)
+ print (personalized_history)
+
+
  else:
  with Halo(text=f"{GORILLA_EMOJI}Loading", spinner="dots"):
  try:
  data_json = {
  "user_id": user_id,
+ #"synthesized_history": personalized_history,
  "user_input": user_input,
  "interaction_id": interaction_id,
  "system_info": system_info

personalization/personalize.py

@@ -0,0 +1,106 @@
+import os
+from presidio_analyzer import AnalyzerEngine, PatternRecognizer
+from presidio_anonymizer import AnonymizerEngine
+from presidio_anonymizer.entities import OperatorConfig
+import json
+from pprint import pprint
+from openai import OpenAI
+from personalization.prompts import get_system_prompt, get_user_prompt
+
+class GorillaPersonalizer:
+ """A class to personalize the user's bash history and query to provide the model with relevant context.
+
+ Attributes:
+ client: The OpenAI client.
+
+ """
+
+ def __init__(self, open_ai_key):
+ """ Initializes the GorillaPersonalizer class."""
+ self.client = OpenAI(api_key=open_ai_key)
+
+ def get_bash_history(self):
+ """
+ Retrieves the user's bash history. (Last 10 commands)
+ """
+ history_file = os.path.expanduser("~/.bash_history")
+ prev_operations = ""
+ try:
+ with open(history_file, "r") as file:
+ history = file.readlines()
+ except FileNotFoundError:
+ return "No bash history was found."
+ return history[:-10]
+
+ def anonymize_bash_history(self, operations):
+ """
+ Uses Microsoft's Presidio to anonymize the user's bash history.
+
+ Args:
+ operations: The user's bash history.
+
+ """
+ analyzer = AnalyzerEngine()
+ analyzer_results = analyzer.analyze(text=operations, language="en")
+ anonymizer = AnonymizerEngine()
+ anonymized_results = anonymizer.anonymize(
+ text=operations, analyzer_results=analyzer_results
+ )
+ return anonymized_results.text
+
+ def remove_duplicates(self, operations: list[str]):
+ """Removes duplicates from the user's bash history
+
+ Args:
+ operations: The user's bash history.
+
+ """
+ return list(set(operations))
+
+ def stringify_bash_history(self, operations: list[str]):
+ """Stringifies the user's bash history.
+
+ Args:
+ operations: The user's bash history.
+
+ """
+ return "\n".join(operations)
+
+ def synthesize_bash_history(self, desired_operation, gorilla_history, history):
+ """Uses OpenAI api to synthesize the user's bash, gorilla history.
+ It synthesizes this history with the current operation in mind, with the goal of providing the model with relevant context.
+
+ Args:
+ client: The OpenAI client.
+ desired_operation: The operation the user wants to perform.
+ gorilla_history: The user's previous operations with the Gorilla
+ history: The user's bash history.
+
+ """
+ system_prompt = get_system_prompt()
+ user_prompt = get_user_prompt(history, gorilla_history, desired_operation)
+ response = self.client.chat.completions.create(
+ model="gpt-4",
+ messages=[
+ {"role": "system", "content": system_prompt},
+ {"role": "user", "content": user_prompt},
+ ],
+ )
+ return response.choices[0].message.content
+
+ def personalize(self, query, gorilla_history, pi_removal=True):
+ """Personalizes the user's bash history and query to provide the model with relevant context.
+
+ Args:
+ query: The operation the user wants to perform.
+ gorila_history: The user's previous operations with the Gorilla
+ open_ai_key: The OpenAI API key.
+ pi_removal: Whether to remove personally identifiable information from the user's bash history.
+
+ """
+
+ history = self.stringify_bash_history(self.remove_duplicates(self.get_bash_history()))
+ if pi_removal:
+ history = self.anonymize_bash_history(history)
+ summary = self.synthesize_bash_history(query, gorilla_history, history)
+ return summary

personalization/prompts.py

@@ -0,0 +1,26 @@
+
+
+def get_system_prompt():
+ return f"""
+You are an assistant for a developer who wants to find the right API call for a specific task.
+The developer has bash history that contains the command they used to perform a task.
+Synthesize their bash history to provide the API call prediction model with extra context about the task.
+For reference, the API call prediction model, called Gorilla, is trained on a large dataset of API calls and their associated tasks.
+You may see the developer's previous operations with the API calling tool in their bash history.
+Use the previous bash history as well as their query to provide the model with a short paragraph of possible relevant context.
+There is a chance that their query has nothing to do with the bash history, so in that case, return 'No relevant context found'.
+"""
+
+def get_user_prompt(history, gorilla_history, desired_operation):
+ return f"""
+The user's bash history is:
+{history}
+
+The user's previous operations with the API calling tool are:
+{gorilla_history}
+
+The query of the user is:
+{desired_operation}
+
+Us this information to provide the model with a short paragraph of possible relevant context.
+ """