Small refactorings

[mseeger]

Some small changes and fixes:

• Move save_hyperparameters to litgpt/parser_config
• Small refactoring in __main__.py
• Clean-up in config.py
• litgpt/generate/sequentially.py: Support better partitioning scheme
• DataModule.connect: Allow for extra **kwargs

The reason for these changes is that I am close to publish a key-value caching and long context inference library which sits on top of LitGPT. These small refactorings here allow me to integrate nicely. For example, I am supplying new scripts in finetune to support fine-tuning with long contexts.

[mseeger]

FAILED tests/test_evaluate.py::test_evaluate_script - ValueError: The repository for EleutherAI/logiqa contains custom code which must be executed to correctly load the dataset. You can inspect the repository content at https://hf.co/datasets/EleutherAI/logiqa.
Please pass the argument `trust_remote_code=True` to allow custom code to be run.
= 1 failed, 1484 passed, 601 skipped, 5 xfailed, 183 warnings, 5 rerun in 1507.57s (0:25:07) =
sys:1: DeprecationWarning: builtin type swigvarlink has no __module__ attribute
Error: Process completed with exit code 1.

[mseeger]

Failure of test not related to my changes

[bhimrazy]

Failure of test not related to my changes

Thanks for the update, @mseeger. Yes, this looks like a new failure on main. We’ll take a look and get it sorted.