Why text-davinci-003?

[vateseeb]

Hi,

first, thx for sharing this neat plugin.

I noticed that the text-davinci-003 model is currently being used for the agents in this project. However, I am curious about the decision behind this choice, especially since the gpt-3.5-turbo model is more cost-effective and allows for defining the AI's role separately.

Based on your experience, is the davinci model more effective in completing tasks? I would be interested in hearing your thoughts on this topic.

Thank you,
Sebastian

[Doriandarko]

I actually had a version with chat api 3.5 turbo, but I didn't like how it was returning the tasks and executing them.

…

On Fri, Apr 14, 2023 at 5:46 PM Sebastian Vates ***@***.***> wrote: Hi, first, thx for sharing this neat plugin. I noticed that the text-davinci-003 model is currently being used for the agents in this project. However, I am curious about the decision behind this choice, especially since the gpt-3.5-turbo model is more cost-effective and allows for defining the AI's role separately. Based on your experience, is the davinci model more effective in completing tasks? I would be interested in hearing your thoughts on this topic. Thank you, Sebastian — Reply to this email directly, view it on GitHub <#1>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJQPTAQLN5RV4P4KKXRJ4DXBHAR5ANCNFSM6AAAAAAW657DEU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

--

_________________________ *Pietro Schirano* Senior Product Designer | Brex <http://brex.com>

[cometbus]

from typing import Dict, List
from quart import Quart, request, jsonify, send_file
import quart_cors
from collections import deque
import openai
import pinecone
import aiohttp

app = Quart(name)
quart_cors.cors(app, allow_origin="https://chat.openai.com")

Set API Keys

OPENAI_API_KEY = ""
PINECONE_API_KEY = ""
GOOGLE_API_KEY = ""
CUSTOM_SEARCH_ENGINE_ID = ""

class BabyAGI:

def init(self,
openai_api_key,
pinecone_api_key,
google_api_key,
custom_search_engine_id,
pinecone_environment="us-central1-gcp",
table_name="test-table",
first_task="Develop a task list"):
self.openai_api_key = openai_api_key
self.pinecone_api_key = pinecone_api_key
self.google_api_key = google_api_key
self.custom_search_engine_id = custom_search_engine_id
self.pinecone_environment = pinecone_environment
self.task_list = deque([])
self.objective = ""
self.table_name = table_name
self.first_task = first_task

openai.api_key = self.openai_api_key
pinecone.init(api_key=self.pinecone_api_key,
              environment=self.pinecone_environment)

def set_objective(self, objective):
self.objective = objective

def add_task(self, task: Dict):
if not self.task_list:
task_id = 1
else:
task_id = self.task_list[-1]["task_id"] + 1
task.update({"task_id": task_id})
self.task_list.append(task)

def get_ada_embedding(self, text):
text = text.replace("\n", " ")
return openai.Embedding.create(
input=[text], model="text-embedding-ada-002")["data"][0]["embedding"]

def task_creation_agent(self, objective: str, result: str,
task_description: str, task_list: List[str]):
messages = [
{"role": "system", "content": "You are a task creation AI that uses the result of an execution agent to create new tasks."},
{"role": "user", "content": f"Objective: {objective}\nResult: {result}\nTask description: {task_description}\nIncomplete tasks: {', '.join(task_list)}"},
]

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=messages
)

new_tasks_text = response.choices[0].message['content'].strip()
new_tasks = new_tasks_text.split('\n')
return [{"task_name": task_name} for task_name in new_tasks]

async def execution_agent(self, objective: str, task: str) -> str:
context = await self.context_agent(index="test-table",
query=objective,
n=5)
context_str = '\n'.join(context)

# Prepare messages for gpt-3.5-turbo
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": f"Objective: {objective}. Task: {task}."},
]
for snippet in context_str.split('\n'):
    messages.append({"role": "assistant", "content": snippet})

response = openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=messages,
  max_tokens=2000,
  temperature=0.7,
)
return response['choices'][0]['message']['content'].strip()

async def context_agent(self, query: str, index: str, n: int):
# Get search results
search_results = await self.search_internet(query, self.google_api_key,
self.custom_search_engine_id)
search_snippets = [
item["snippet"] for item in search_results.get("items", [])
]

query_embedding = self.get_ada_embedding(query)
index = pinecone.Index(index_name=index)
results = index.query(query_embedding, top_k=n, include_metadata=True)
sorted_results = sorted(results.matches,
                        key=lambda x: x.score,
                        reverse=True)
return search_snippets + [(str(item.metadata['task']))
                          for item in sorted_results]

async def search_internet(self, query, api_key, custom_search_engine_id):
print("search_internet called with query:", query)
base_url = "https://www.googleapis.com/customsearch/v1"
params = {"key": api_key, "cx": custom_search_engine_id, "q": query}
async with aiohttp.ClientSession() as session:
async with session.get(base_url, params=params) as response:
result = await response.json()
print("Search results:", result)
return result

def clear_task_list(self):
self.task_list.clear()
self.task_id_counter = 0

baby_agi = BabyAGI(OPENAI_API_KEY,
PINECONE_API_KEY,
GOOGLE_API_KEY,
CUSTOM_SEARCH_ENGINE_ID,
table_name="test-table",
first_task="Develop a task list")

@app.route("/set_objective", methods=["POST"])
async def set_objective():
global baby_agi
data = await request.json
objective = data["objective"]
baby_agi.clear_task_list(
) # Clear the task list and reset the task ID counter
baby_agi.set_objective(objective)
return jsonify({"status": "Objective set", "objective": objective})

@app.route("/add_task", methods=["POST"])
async def add_task():
global baby_agi
data = await request.json
task_name = data["task_name"]
task = {"task_name": task_name}
baby_agi.add_task(task)
return jsonify({"status": "Task added"})

@app.route("/execute_next_task", methods=["POST"])
async def execute_next_task():
global baby_agi
if not baby_agi.task_list:
return jsonify({"status": "No tasks in the list"})

task = baby_agi.task_list.popleft()
result = await baby_agi.execution_agent(baby_agi.objective,
task["task_name"])

new_tasks = baby_agi.task_creation_agent(
baby_agi.objective, result, task["task_name"],
[t["task_name"] for t in baby_agi.task_list])
for new_task in new_tasks:
baby_agi.add_task(new_task)

response = {
"task_id": task["task_id"],
"task_name": task["task_name"],
"result": result,
"new_tasks": new_tasks,
}

return jsonify(response)

@app.route("/get_task_list", methods=["GET"])
async def get_task_list():
global baby_agi
task_list = [task["task_name"] for task in baby_agi.task_list]
return jsonify({"task_list": task_list})

@app.route("/openapi.yaml", methods=["GET"])
async def get_openapi_yaml():
return await send_file("openapi.yaml",
mimetype="application/vnd.oai.openapi")

@app.route("/logo.png", methods=["GET"])
async def get_logo():
return await send_file("logo.png", mimetype="image/png")

@app.route("/.well-known/ai-plugin.json", methods=["GET"])
async def get_ai_plugin_json():
return await send_file("ai-plugin.json", mimetype="application/json")

@app.route("/test_google_api", methods=["GET"])
async def test_google_api():
global baby_agi
query = request.args.get("query", "test")
search_results = await baby_agi.search_internet(
query, baby_agi.google_api_key, baby_agi.custom_search_engine_id)
return jsonify(search_results)

if name == "main":
app.run(debug=True, host="0.0.0.0", port=5001)

[vateseeb]

Thx for the reply and sharing the code!