Skip to content

Conversation

@david-thrower
Copy link
Owner

Merge in updates to #240: Improved testing printouts for human - subjective evaluation of generated text.

  • Print out samples with different permutations of generation parameters.
  • Conditional printout if the perplexity is low enough it will print meaningful text, prevent logs from being flooded with malarkey generated by bad models in the NAS pipeline.

Added some tests to print out text of lowest perplexity trials with different permutations of generation params.
Trigger tests to run.
Add conditional filtering for result < result_cutoff, so verbose prints only print when there is a result that makes sense to generate text from.
@david-thrower david-thrower merged commit 1ef29f4 into 240-branch-to-diverge-cicd-scale-nlp-hpo-from-at-scale-study Sep 28, 2025
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants