'Fetch MMLU' component now has 'special' settings of all_medicine_datasets and all_mmlu_datasets which select more than one dataset from MMLU. It also injects 'dataset' and 'subject' entries into each line
The random and kNN few shot pipelines now support running multiple different guidance programs in parallel
Add a new guidance program for fewshot, which puts the examples as conversation turns, and not in the system prompt
Move some logging messages to 'debug' after filling up local disk on the compute nodes
Improve the scorer, so that it uses the 'dataset' and 'subject' keys
Tidying up things under
azureml. Major changes:all_medicine_datasetsandall_mmlu_datasetswhich select more than one dataset from MMLU. It also injects 'dataset' and 'subject' entries into each line