Brief note: trained locally on Windows. Bits and Bytes and PEFT were not working so I fine tuned a smaller 1b model without quantization. Later I switched to Ubuntu and got qlora and peft working, then fine tuned 7b