No lr_scheduler in full_finetune_distributed

[jxmsML]

Learningrate scheduler isn't set up in the full_finetune_distributed training loop, only exists in the lora finetuning.

[ebsmothers]

Hi @jxmsML thanks for pointing this out. We do only support fixed learning rate in the full finetune recipes, but I agree we should support schedulers there like we do in LoRA. There's not really a technical blocker for this, so if you're interested to open a PR for the feature we'd love to have it. Otherwise we hope to add it soon!