mosaicml / llm-foundry

LLM training code for Databricks foundation models
https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm
Apache License 2.0
3.84k stars 503 forks source link

[MCLOUD-4623] Add more detailed exception when user has uppercase in their example case but could potentially match the exampe type #1245

Closed shitaoli-db closed 3 weeks ago

shitaoli-db commented 1 month ago

As title, we want to show user a better exception, i.e. to give lowercase keys, when the user example contain upper case keys like "Prompt" that could potentially be fixed by using lower cases.

shitaoli-db commented 1 month ago

This PR seems a bit overkill to me. It adds code just for detecting one possible you may have messed up your data (out of the many possible ways), when the error message would already tell you fairly clearly what was wrong. If this is an error we've seen a handful of times, maybe we can just update the UnknownExampleTypeError to mention that you may want to check your casing as the keys are case sensitive.

Yeah, I was thinking the same at first place, I think another idea is no change in the LLM foundry but rather to only change on https://github.com/databricks-mosaic/mcloud/pull/4083 for user facing error. @jjanezhang WDYT on this?

shitaoli-db commented 3 weeks ago

Closed for now since we agreeed that the change should be on mcloud side.