chain-ml/council

Council is an open-source platform for the rapid development and robust deployment of customized generative AI applications

council's Issues

Allow commonly used classes to be imported directly from `council`

Instead of:

from council.controllers import LLMController
from council.evaluators import LLMEvaluator
from council.agents import Agent
from council.filters import BasicFilter

Do:

from council import LLMController, LLMEvaluator, Agent, BasicFilter

It's much easier to remember and also extremely easy to implement (I can put in a PR). Basically everything in the Hello, World example should probably be able to be imported directly from council (or council.models which makes sense separately).

Config instantiation without parameters

Steps to Reproduce:

1. Run the following to llama_integration.ipynb: ! pip install python-dotenv llama-index council-ai
2. Run all fields in notebook in order

Error:

AttributeError Traceback (most recent call last)
in <cell line: 15>()
13 dotenv.load_dotenv()
14 config = OpenAILLMConfiguration.from_env()
---> 15 config.model = Option("gpt-3.5-turbo")
16 llm = OpenAILLM(config)
17

AttributeError: can't set attribute 'model'

Cause:
Option.model is no longer exposed after OpenAILLMConfiguration refactor in #113

Proposed Fix:

config = OpenAILLMConfiguration(
api_key=os.getenv("OPENAI_API_KEY"), model="gpt-3.5-turbo"
)
llm = OpenAILLM(config)

when to update?

It was said the repo will updated every two weeks. Hope the project will become more popular.

OPENAI_LLM_MODEL parameter appears to be required for OpenAILLM but not mentioned in quickstart

Hi team! Very cool project.

It appears that the OpenAI API requires the model parameter on every call (docs) but it's an Option in OpenAILLMConfiguration.

Configuring OPENAI_LLM_MODEL="gpt-3.5-turbo" works. Omitting OPENAI_LLM_MODEL from my environment causes the example to fail due to the API returning a 400 with error message you must provide a model parameter.

Here is a small repro using Council: https://gist.github.com/rbm/3016c5a8d4fff2ac63de5990cb225444

Just curling the OpenAI endpoint directly also fails, for clarity:

# Note "model" is omitted from data payload below
# Fails with "you must provide a model parameter"
curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
     "messages": [{"role": "user", "content": "Say this is a test!"}],
     "temperature": 0.7
   }'

Unsure for Council whether it would be better to default to gpt-3.5-turbo or make the configuration non-optional when using OpenAILLM.

Phrasing in First Example

first_example.ipynb

The following:
hw_chain = Chain(name="Hello World", description="Answers with a poem about titled Hello World", runners=[hw_skill])

Should read:
hw_chain = Chain(name="Hello World", description="Answers with a poem titled Hello World", runners=[hw_skill])

How does council compare with autogen?

Any suggestion when to use autogen, when to use council? Thanks.

README - Discord Invite is invalid

The discord invitation in the README does not properly link to a valid discord invitation. It should link to: https://discord.com/invite/DWNCftGQZ3