base SFT for the model, use the down-stream -winton model
SFT: https://wandb.ai/new-eden/AFM-SFT/runs/u8fj6r6o?nw=nwuserdeltavector
KTO: https://wandb.ai/new-eden/AFM-SFT/runs/fgkl4ijs?nw=nwuserdeltavector (Early stopped at ckpt 100~)
base SFT for the model, use the down-stream -winton model
SFT: https://wandb.ai/new-eden/AFM-SFT/runs/u8fj6r6o?nw=nwuserdeltavector
KTO: https://wandb.ai/new-eden/AFM-SFT/runs/fgkl4ijs?nw=nwuserdeltavector (Early stopped at ckpt 100~)