> With GenAI ... the same set of actions can

> produce, often wildly, different results.

If you are using sampling decoding that uses pseudo-randomness to produce "creative" output, you need to log the random seed used with a given prompt so support can reproduce the same output.

See:

https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-model-parameters.html?context=wx&locale=en&audience=wdp#random-seed