Readme: Brought "Knowledge" ahead of "Skills"

[rfrht]

For someone training their models, the most common use case is creating knowledge - and not skills. Knowledge QNAs have strict requirements, needs schema version 3 - and an interested party can be misled with the "skills" examples.

This commit brings Knowledge QNA ahead of Skills, highlight the difference between them - and also highlights the context and qna pair requirements.

signed-off-by: [email protected]

[jjasghar]

Thank you for this, but we are attempting to migrate to docs.instructlab.ai. This rewrite maybe worth looking at that repo instead

[rfrht]

Thank you for this, but we are attempting to migrate to docs.instructlab.ai. This rewrite maybe worth looking at that repo instead

Hello, Asghar! Thanks for your review. I'll be definitely doing it - already forked it and will propose the changes over there too.

Meanwhile, can you approve this PR for this repo, hoping to help our users? I took issue on the thing because of the amount of time that I lost figuring out the qna :-)

Best regards,

• RF.

[rfrht]

Added another commit to this batch - Another important one clarifying the requirements for Q&A block: I learnt the hard way that the Q&A file requires exactly 5 contexts, with exactly 3 Q/A pairs.

Reinforcing the verbiage.

CC: @mairin

[jjasghar]

After continued discussion at the triage meeting, please make these changes at the docs.instructlab.ai, we want to cut down these readmes (across the project) and centralize non-developer content at docs.instructlab.ai.

[bjhargrave]

I learnt the hard way that the Q&A file requires exactly 5 contexts, with exactly 3 Q/A pairs

I am curious about this. I have heard it before but could not find anywhere in the sdg code where this was enforced or required. Do you have a pointer? If this is truly the case, then the schema should also enforce this. Thanks.

[bjhargrave]

For someone training their models, the most common use case is creating knowledge - and not skills.

I am not sure this is true for all. It may be true for you in the use cases you are working on. I have seen other cases where it was almost all compositional skills with little to no knowledge.

Knowledge and compositional skills are both relevant but one is not more important than another.

[bjhargrave]

Knowledge QNAs have strict requirements, needs schema version 3

Compositional skills also need schema version 3 when contributing to this repo.

[rfrht]

Hello @bjhargrave

Compositional skills also need schema version 3 when contributing to this repo.

Alright, the existing documentation in README.md shows schema Version 2 for skills, all around.

Where can I find the proper Schema V3 for Skill, so I can update this content?

Best,

• RF.

[bjhargrave]

V3 Compositional Skills schema is the same schema as for V2. So the only change needed is to update the version to 3.

The schemas are at https://github.com/instructlab/schema/tree/main/src/instructlab/schema

[rfrht]

Hello, BJ!

Added a new commit to the batch. Check the diff here and the final rendered page here.

[jjasghar]

I really think we should be cutting this README down and moving to http://github.com/instructlab/docs.instructlab.ai for any of these changes.

Our readmes are so much, and confusing, and any effort here will just be eventually moved over anyway.

[rfrht]

I wholeheartedly agree with that. This Readme should be wiped out, have some few words explaining what it is about and then referring to the new documentation URL.

Meanwhile, I think that we are sitting on the content and people are consuming outdated content.