GitLab

Fix for 4bit without compress_statistics.

chore: update dev setup

Avoid double-quantizing when calling `cuda()`

Update env_vars.py to include GOOGLE_VM_CONFIG_LOCK_FILE

Update cuda_install.sh

Add option to download and install Cuda 12.3

See this comment: https://github.com/TimDettmers/bitsandbytes/issues/620#issuecomment-1666014197

adding this variable to be ignored allows the bitsandbytes import and setup to proceed on a GCP VM in a jupyter notebook.

fixes to recent PR #753

Fix typehint error

Save and load  in NF4 / FP4 formats

Update utils.py to remove dupe `replace_linear`