[FEATURE] `SelfInstruct` predefined structured output

argilla-io / distilabel

Distilabel is a framework for synthetic data and AI feedback for engineers who need fast, reliable and scalable pipelines based on verified research papers.

https://distilabel.argilla.io

Apache License 2.0

1.42k stars 103 forks source link

[FEATURE] `SelfInstruct` predefined structured output #881

Open plaguss opened 1 month ago

plaguss commented 1 month ago

Description

As part of issue https://github.com/argilla-io/distilabel/issues/832, we started adding default structured outputs for tasks that could benefit from it. SelfInstruct is a good candidate.

PR https://github.com/argilla-io/distilabel/pull/868 can be used as a guide on what to update.

ignacioct commented 1 week ago

Hi @plaguss ! Do you think this issue is a good (re)starting point?

plaguss commented 1 week ago

Hey @ignacioct! Sure, let me know if you need help, there should be examples in other steps (like UtraFeedback) to see how this can be implemented.

ignacioct commented 1 week ago

@plaguss I started to play a bit with distilabel yesterday, and I tried to run it locally using Ollama models, but was not succesful. As I don't have access to any API like OpenAI's, do you know what's the best strategy for running LLMs locally with distilabel? Maybe vLLMs? We can discuss it via Discord if you wanna, I just couldn't find you there earlier :)