[CAP-9] Supervised fine-tuning

[chiffonng]

Adapted from Unsloth notebook for finetuning Gemma-2-9B and this tutorial

• PEFT documentation
• Wandb to track finetuning
• Hands-on Large Language Model Chapter 12

Hyperparameter search will be done in CAP-33, but some ideas to start is

• Initial values for LoRA config here
• Use ray tune for hyperparameter search and integration with wandb

_{From SyncLinear.com | CAP-9}

[chiffonng]

Safe to close