Acc. to https://huggingface.co/blog/dvgodoy/fine-tuning-llm-hugging-face

bitsandbytes had to be bumped to 0.45.2 to avoid errors in Colab env

this is what has been working for me

!pip install transformers==4.46.2 peft==0.13.2 accelerate==1.1.1 trl==0.12.1 bitsandbytes==0.45.2 datasets==3.1.0 huggingface-hub==0.26.2 safetensors==0.4.5 -q