Missing: General-purpose pre-trained model

[sahoo-rk]

Hi @cactuskid
Nice to see the development of the FT2. I am trying to give a try. Particularly, I am interested in using your pre-trained model to reconstruct a phylogeny as mentioned over the 'Basic Workflow'. However, the model 'mergeddecoder_foldtree2_test' and associated files (e.g., mergeddecoder_foldtree2_test.pt, mergeddecoder_foldtree2_test_pair_counts.pkl) are currently missing, as I tried on 12 Jan 2026.

Any lead in this aspect would be helpful. Thanks.

[cactuskid]

hi. thanks for the comment. I haven't updated these files in the git yet for the latest trained models. I was just working on the main branch since i didnt think anyone was using this project yet and a few things are broken... im going to update the models in a few days and make sure everything runs end to end. then ill switch to a cleaner dev strategy. ill let you know on here and update the readme with the latest instructions on running the pipeline.

[sahoo-rk]

Thanks for the reply. I look forward to the updates.

[JorgensenD]

Hi David,
I have just been checking out the package and got stuck in the same spot. I just wanted to check the model is still missing rather than I have some other cryptic error!
Thanks,
David

[cactuskid]

hey. sorry for taking so long to respond. i overhauled a few things on how the models are trained and saved etc. I've been working this project solo and didn't really count on users taking much of an interest until I wrote the preprint haha. Anyway, I've updated the main branch with a 'production model' trained with the latest iteration of the structural graph objects and that's compatible with the new classes of encoder and decoder. I'll leave the main branch more or less stable/working from now on and try to dev new features on other branches. the current model is a 30 character one which should be in models/production. I added some lines to the readme with a tldr on how to run the basic foldtree2 treebuilder.

[cactuskid]

give it a try and let me know if the models are working for you or anything is missing... and also if it's performing well for you.