distilabel
distilabel copied to clipboard
[FEATURE] Include custom steps, requirements in pipeline config file and `allow_remote_code` flag
Description
Currently, the pipelines that can be executed directly from the CLI or when loaded with Pipeline.from_yaml
are those which include only steps from distilabel
package.
The idea is to also serialize custom steps (as safe as possible) and include them in the pipeline config file, so if an user desires to execute the pipeline from the CLI, it can be executed if allow_remote_code=True
.
In addition, some pipelines requires some dependencies to be executed (transformers
, openai
, etc). These requirements should be also added to the pipeline config file and when executing from CLI and if not installed, then prompt the user to consent the install of these requirements.