Transformers issued an important change for the OPT models breaking their compatibility with all older versions. In order for people to be able to use all models on the menu they need 4.20.1 so this is now forced in the dependencies making the update easier.
Transformers 4.20 now requires accelerate to be installed for some of the features we use in KoboldAI. This is now a required dependency for updated users.
Makes transformers 4.20 mandatory in the dependency lists, not because the old versions are no longer supported but because it contains fixes that benefit our users and this makes it easier for them to update to it. If you stick to an older version the OPT and XGLM workarounds we have in place will remain functional, but you miss on the enhancements newer transformers versions bring.
Huggingface's repo is further behind than conda-forge so we will no longer offer it in the installer. The more is loaded from conda-forge the better. The same transformers package will still be installed, but a newer one from conda-forge is now guaranteed.
Allows model creators to customize the welcome message using Markdown and Limited HTML
Existing United users need to run install_requirements..bat again, you can leave the existing dependencies intact.
Updates dependencies, play.sh didn't work properly so removing that for now since manually running aiserver.py is superior on Linux until I can get conda to init inside the script
Seperate file so people can easily go back to the legacy implementation based on finetune (Recommended until Huggingface's compatibility is improved) . You can install and use both.
Allows anyone to easily create a ROCm compatible conda environment. Currently set to the newer transformers, you can edit the github link if you want the finetune one.
Finetune's fork has unofficial support which we supported, but this is not compatible with models designed for the official version. In this update we let models decide which transformers backend to use, and fall back to Neo if they don't choose any. We also add the 6B to the menu and for the time being switch to the github version of transformers to be ahead of the waiting time. (Hopefully we can switch back to the conda version before merging upstream).
Looks like cudatoolkit is now shipping 11.3, but pytorch has no version for this resulting in an installation of the CPU only version. This is going to lead to people unable to get their GPU running, so for now we force the recommended 11.1 version. I also don't see any harm in allowing Python 3.9 so thats now the default as well to prevent future issues.
First step towards native Colab support, built in Cloudflare tunnels easily allows players to play KoboldAI on another device. This mode also removes buttons that would get you stuck if you have no local PC access.
Big overhaul of the installer, partially based on the #53 commit from LexSong.
The following is new :
- Conda has been replaced with MicroMamba, allows the dependencies to automatically download the best version of Python and prevents all the issues with people failing to download conda.
- The installer now has more options so you can choose not to delete the existing files, and it has new optional virtual K: drive support to bypass all the pathing issues people are having (Sorry Windows 7 users, its still not compatible even now).
- Docker support for Linux systems has been added including ROCm support.
- Environment files are now used to more easily keep everything on track, and to allow conda users to manually create environments across all operating systems (ROCm is an outlier in this because i have to use AMD's Pytorch docker for now it was to much hassle getting their Conda to use the environment file to add it to this commit).
- Play.bat has been changed to allow the virtual drive support, everything should still be compatible with old installations as I kept all the paths intact.
Henk
henk717
Ben Fox
Gnome Ann