enabler: fine-tuning api for models

[kdziedzic68]

Feature description

prepare the structure of ragbits.fine-tune package.

some cases that come to mind:

• distill gpt4o-mini from gpt4 outputs
• distill local models from openai models
• distill llama7b from llama70b

etc.

Motivation

For some case specific applications - it may be not enough for users to properly prompt the model and some fine-tuning may be required to achive the desired performance. Even when the prompting succeds with large and expensive models in the specific case users may want to achieve close quality of results with smaller and cheaper models - so the special focus should be put on distillation tasks here.

Additional context

No response