FIX: Make CorDA example work

[5eqn]

Previously when changing load_dataset logic in #2231 , I've forgotten traindataset = [] for dataset codefeedback and wizardlm, so the example doesn't work. In this PR I've fixed this bug, and tested that the example works for all datasets.

[BenjaminBossan]

Thanks for catching this. I wonder if it would be easier if we move this line:

peft/examples/corda_finetuning/datautils.py

Line 199 in 6d458b3

traindataset = []

to the beginning of the function body. That way, we can ensure that traindataset is always initialized and don't need to call it in each if-else branch. WDYT?

[5eqn]

Thanks for catching this. I wonder if it would be easier if we move this line:

peft/examples/corda_finetuning/datautils.py

Line 199 in 6d458b3

traindataset = []

to the beginning of the function body. That way, we can ensure that traindataset is always initialized and don't need to call it in each if-else branch. WDYT?

Thanks for your suggestion! After this move, if someone were to add a dataset, it will indeed be less error-prone. I've updated the code and it's ready for review.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[BenjaminBossan]

Thanks for the fix of the CorDA script. The failing tests are unrelated and flaky, so this is good to be merged.