feat: add adapter for fms-hf-tuning

[YashasviChaurasia]

Addresses Issue: #2
This PR introduces the FMS-HF Adapter, that converts model, compute, and data configs to fms-hf-tuning ready training configs.

The FMSAdapter extends the VanillaAdapter to convert the processed Intermediate Representation (IR) into runnable fms-hf training configurations.

• Write clean data_config.yaml derived directly from the IR’s data_preprocessor section.
• Handles nested FSDP configs — flattens only the relevant fsdp_config keys into proper CLI arguments, assuming only fsdp config as nested config in dist_config
• Skips redundant fields (e.g., training_data_path) that are already included in dataset section in the data_config.
• Ensures raw Jinja templates in YAML are preserved exactly, avoiding unwanted formatting or escaping.

[YashasviChaurasia]

Sister PRs:

• update: ensure knowledge_base *.csv are installed #5
• fix: remove multiple bugs with QA dataset configs #6
• fix: remove multiple bugs with Chat dataset configs #7

[YashasviChaurasia]

Will soon add Fixes for Error Handling with Datasets and ChatData..
and will then push Wrapper code on fms-hf-tuning once all of these are reviewed and merged!

[YashasviChaurasia]

The PRs for Chat and QA dataset bug have been added separately so that they can be merged earlier incase Adapter PR takes longer review cycle.. Those PRs [#7 , #6 ] are independent of FMSAdapter fix and are required for proper functioning for recommender..

[kmehant]

Can you paste the corresponding PR and design on how you are planning to use this adapter with fms-hf-tuning?

[YashasviChaurasia]

Attached the PR for fms-recommender.py script which used FMSAdapter : foundation-model-stack/fms-hf-tuning#644

[kmehant]

why is that PR foundation-model-stack/fms-hf-tuning#644 in draft?

[YashasviChaurasia]

Its open for review now !
Would require to update dependency to main branch from my custom branch when FMSAdapter is merged to main

[kmehant]

Once the fms-hf-tuning PR is merged. Please add one example on e2e usage using fms-hf-tuning in this repo and then refer that in README.md of this repo.