OneTrainer/modules/util/quantization_util.py explicitly imports bitsandbytes, no checks, no requirements, no nothing.
This will fail on startup on pretty much anything that's not Windows running CUDA.
What did you expect would happen?
All other code seems to only require bnb as needed, not as an explicit, static import.
This seems to have been done as part of recently introduced Flux support.
What happened?
OneTrainer/modules/util/quantization_util.py explicitly imports bitsandbytes, no checks, no requirements, no nothing. This will fail on startup on pretty much anything that's not Windows running CUDA.
What did you expect would happen?
All other code seems to only require bnb as needed, not as an explicit, static import. This seems to have been done as part of recently introduced Flux support.
Relevant log output
No response
Output of
pip freeze
No response