unitxt
unitxt copied to clipboard
IBM
🦄 Unitxt: a python library for getting data fired up and set for training and evaluation
Signed-off-by: elronbandel
These code sections are not being run in preparation tests and de-facto can break without anyone noticing
Key contributions: 1. No need to assign values to output fields in infer/produce functions (only to input fields as the task specify) 2. No need to specify a card unless...
Today, RenameFields uses a map. RenameFields(field_to_field={"labels": "label"}), to simply the experience we should have Rename(field="labels", to_field="label"). We should replace all existing use in documentation of RenameFields with Rename and deprecate...
An inference engine that has in its constructor a LogProbsInferenceEngine and call it once for every option in the "task_data/options". Exactly the way it's being used internally for producing a...
Current dict_utils performs overwrite of dictionaries: ``` if query.strip() == "": # change the whole input dic, as dic indeed matches "" if isinstance(dic, dict): if value is None or...
Instead , replace with load_dataset() which is the recommended way to load datasets.
Today, the when the catalog consistency fails one gets this error message. ``` Comparing generated and old catalog... ***** Directories has differences ****** {'file': '/home/runner/work/unitxt/unitxt/src/unitxt/catalog/metrics/string_containment', 'diff': 'old only'} File "/home/runner/work/unitxt/unitxt/utils/prepare_all_artifacts.py",...