feat: Silence certain warnings

[willmj]

Description of the change

This PR is to silence certain warnings, preliminarily silencing future warnings if the new Training Arguments flag warnings is set to False (which it is by default). Imported warnings into sft_trainer.py Changed the maximum number of class parameters to 8 from 7 in .pylintrc.

With this preliminary change (only silencing FutureWarnings) when running tox -e py: Before: 153 passed, 14 skipped, 249 warnings in 76.34s (0:01:16) After: 153 passed, 14 skipped, 207 warnings in 69.31s (0:01:09)

Questions for reviewers

• Should I use a pre-existing flag such as log_level to do this?
• Is where I put the silencing code where it should be?
• If we want to add this new flag:
  • Is it okay to change pylintrc max from 7 to 8 or do we want to maintain 7 as a maximum?
  • Should we allow direct control over the type of warnings that are silenced? If so,

How to verify the PR

In root of fms-hf-tuning

python3 tuning/sft_trainer.py \                                        
--model_name_or_path Maykeye/TinyLLama-v0 \
--training_data_path tests/data/twitter_complaints_small.jsonl \
--output_dir outputs/lora-tuning \
--num_train_epochs 5 \
--per_device_train_batch_size 4 \
--gradient_accumulation_steps 4 \
--learning_rate 1e-5 \
--response_template "\n### Label:" \
--dataset_text_field "output" \
--use_flash_attn false \
--torch_dtype "float32" \
--peft_method "lora" \
--r 8 \
--lora_dropout 0.05 \
--lora_alpha 16  \
--warnings True

OR

--warnings False

You can also run tox -e py before and after changes.

Was the PR tested

• [ ] I have added >=1 unit test(s) for every new method I have added.
• [x] I have ensured all unit tests pass