rudolfolah / chaincrafter Goto Github PK

• write the library
• support OpenAI
• publish to NPM

The current code will overwrite the starting input vars using the output of the previous prompt.

The fix is to update the input vars dictionary that is passed to the prompt that is being run. This requires that input var names are unique.

• gpt4all
• llama.cpp

https://app.endpoints.anyscale.com/landing

I'm trying to run some of the examples and I keep on getting this error.

ImportError: cannot import name 'Chain' from partially initialized module 'chaincrafter' (most likely due to a circular import) (chaincrafter/__init__.py)

How I reached this point:

1. Forked and cloned repository onto local machine
2. Pip installed chaincrafter
3. Tried to run hello_world.py under ./python/examples

When I run the __init__.py script, there are no errors. I thought it might be an issue of importing incorrectly. The original import is

from chaincrafter import Chain, Prompt

while in the __init__.py script, these lines ares used

from chaincrafter.chains import Chain
from chaincrafter.prompts import Prompt

still getting the same error nonetheless.

the current README on https://pypi.org/project/chaincrafter/ is useful for local development

• let's move the sections to LOCAL_DEV.md and PUBLISH_BUILD.md
• copy the relevant parts of the top-level README into the package README

https://pypi.org/project/openai/

• convert some examples to use Jupyter and add the notebook to the repo
• convert some examples and use them in Google Colab
• add example of document loading/splitting using Unstructured, based on this:
  • writing_suggestions_document_loader.py
    
  • loaders.py
  • splitters.py
• create example that uses outlines

To test this:

1. create an account at https://cerebrium.ai/
2. deploy a pre-built model

• new CerebriumAI model based on ChatModel
• support for the pre-built model endpoints, example

When adding code from any language that uses braces such as JavaScript, the { needs to be escaped. Alternatively, there could be an option to skip the format-string entirely and treat the prompt as already formatted.

https://github.com/BerriAI/litellm