bug fixes: handle shuffling of rows, columns, and casting of non-stri…

bug fix: undefined variable target_choices

style fix

bug fix: do not cast target_column_unique_values to list, it is alrea…

add support for TargetConfig when serializing

bring up to date

add hyperparameter-tuned version of target quality clf

add a target quality scorer

do not use activation checkpointing when using peft

use default hparams when num_iters is 1

decrease iterations

handle multiclass case

remove thread_count; not a valid kwarg for random search

increase iterations

only use gpu with lager train set

bug fixes; reduce iterations to 64

introduce some changes to reduce runtime and stop early when no impro…

add catboost usage of GPU

update token id in llama 1

add catboost support

update docstring

refactor target selection object into a new object, TargetSelector, a…

correct voab size detection for llama 1 model

typo fix

add support for llama 1 model

bug fix: correct hf saving of model when using FSDP; remove use of sa…

bug fix: convert attention mask to torch tensor before caling .masked…

Merge remote-tracking branch 'origin/main' into llama-3.1-support

move finetune.py into module

implement attention mask inversion as in core transformers library