Add prompts layer to ModelKit
Describe the problem you're trying to solve When developing a gen AI model the system and user prompts used for validation, in-production use are critical to the end result.
Describe the solution you'd like It would be great to have the option of storing these in the ModelKit directly, or as a reference to git. Making prompts its own layer would make it trivial to pull only the prompts in cases where a dev was iterating on those.
Describe alternatives you've considered
Storing prompts in git works, and you can even add the repo to the code layer in a ModelKit, but being able to have a dedicated layer would make things easier to track and rev.
I made some changes to the repository code to add a new layer to the docker image dedicated to just all the prompt files. I changed these files
mediaType.go FiltersFromUnpackConf | filter.go | shouldUnpackLayer | filter.go | saveKitfileLayers | local-storage.go | unpackRecursive | core.go |
Am i missing some change here? I am new to the codebase.
@Eshan-Jogwar it sounds correct to me but you can issue a draft pull-request and we can have a quick look.
hey after those changes i am getting an error in unpacking the code layer
if there is some context you could provide me about this error.
the pack command runs correctly and also gives out that the prompt layer is made with the sha256
this is the manifest inspect after pulling from the registry
This is the final error
hey i figured the error out. now it is working.
should i also include a --prompt in the cli to only unpack the prompts layer like the code layer?