Trying to fine tune this with LoRA
Did a run the other night, and then FOOLISHLY stopped it to try unsloth, this this corrupted my local env, and have spent the last two days battling to get it back.
Does anyone have a config which works?
bitsandbytes==0.45.3
peft==0.14.0
sentencepiece==0.2.0
tokenizers==0.13.3
torch==2.5.1+cu121
torchvision==0.20.1+cu121
transformers==4.26.1
trl==0.4.2
PyTorch: 2.5.1+cu121
CUDA available: True
CUDA version: 12.1
Transformers: 4.26.1
PEFT not installed
BitsAndBytes: 0.45.3
to be clear, im not sure any of these work
@abcampbell Apologies on the Unsloth env issue - what specifically broke? I can help debug it if that works.
Note it's best to update transformers - 4.26 is way too old - at least 4.46 or so is recommended.
On env issues, it's best to make a new virtual env and install packages