RongchangLi

> Hi @Lilaoxie, I believe just installing `onnxruntime-gpu` should be enough for GPU-based inference, but I didn't try it on my own for this model. Yeah, I just find it...

> > subprocess.CalledProcessError: Command '['which', 'g++']' returned non-zero exit status 1. > > It seems `g++` is not installed? can you try installing it and then install this package by...

Thanks for your patient explanation. I will try some experiments and I will leave a comment on this issue if get any new findings.

Sorry I did not explain that the ubuntu_train_vpt_vtab.sh is a file equals to slurm_train_vpt_vtab.sh. I tried to set lr=1e-2, weight_decay=1e-4, the result improved to 71.21. If changing the number of...

Paper name/title: C2C: Component-to-Composition Learning for Zero-Shot Compositional Action Recognition Paper link: https://arxiv.org/abs/2407.06113 Code link: https://github.com/RongchangLi/ZSCAR_C2C

The bug is solved by updating the version to v1.0. However, there is another bug: when using _model=clip_model.visual.cpu()_ rather than _model = clip_model.visual.transformer.resblocks.cpu()_. It also falls in an endless loop....