-
revert model downloader version
revert downloader version to last compatible file structure
-
switch to archived link
-
update repo
-
update zip files
-
change deprecated api code
update_repo_visibility is deprecated -
commit message change
-
set correct repo yaml metadata
instead of showing up as finetune, show up as quantized -
potential support for blackwell
don't have 50 series so can't test for sure, but there's no reason it shouldn't work
-
upgrade to new default pytorch version and instructions
-
6hb default again
much bigger vram usage in newer models, not really worth it much anymore.
also removed fast safetensors because deprecated
-
prepare for exllamav3
