# Requirements for model training - pinned versions for compatibility
torch
transformers==4.44.2
datasets
peft
bitsandbytes
trl==0.8.6
accelerate
scipy
hf_transfer
rich