Mistral interviews
Mistral interviewsPublic evaluation and observability product design.
Keep a result’s lineage
“Reproduce results or explain regressions”
MistralProduct announcement
A gap teams struggle to close
Mistral names reproduction as a gap. Which versions would you preserve to investigate it?
Example“Define a dataset, a task, and scorers”
MistralEvaluation documentation
Let production change the test set
“Online evaluations score real sessions in production”
MistralRelease notes
Last reviewed