Cached LLM results aren't getting reset after moving a pipeline

Ask the Community
1

I’ve split out a transform from a large pipeline into a smaller one. The transform had 3 cached LLM blocks that I’m trying to “sever” from the existing cache. Whatever I do, I can’t seem to fix these deployment errors!

I’ve tried a few things to no avail:

  • Run the new pipeline once without caching (to see if that’d reset things);
  • Deployed the old pipeline after deleting a board’s cached results;
  • Recreated the LLM nodes in the new pipeline (i.e., not just copying and pasting);
  • Changed the names of the output columns (in the transforms);
  • Adjusted the LLM prompt and settings and saving the pipeline.

image
image656×656 30.2 KB

How does the LLM cache actually link to a transform board? And how do I sever/reset it so I can build my new pipeline?

2

Interesting… even though the two outputs of my new transform appeared to be in the same group, checking them off together + applying a profile to them (the same as the default profile) eliminated the errors. So it seems like it was an issue with the grouping, not with a conflicting/preexisting cache.

3

Was the first group a custom group or was it the default grouping? The default grouping is actually having every output in it’s own individual group

4

Correct! Everything was in the default group to start out, and moving them to a group fixed the problem. (But this was also a bit confusing, since if the 3 cached results are hidden datasets behind the 1 visible dataset, it seems like those 4 should be in a single group by default.)