tianyic / only_train_once

OTOv1-v3, NeurIPS, ICLR, TMLR, DNN Training, Compression, Structured Pruning, Erasing Operators, CNN, Diffusion, LLM

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

pruning transformer-based model

huanghmingyue opened this issue · comments

Hi. Thank you for your contribution.
Can OTO be used for pruning Transformers? I want to apply it to a DETR-like object detection model(DINO:https://github.com/IDEA-Research/DINO). Is this feasible?
I tried
oto = OTO(model, example_inputs)
and got this error:

image
image
image
image
image

@huanghmingyue I think you are using a very old version of pytorch. Try upgrading python and pytorch.

@huanghmingyue I agreed with @iamanigeeit that you need to consider use a newer pytorch for getting tracing graph. Transformer is supported such as bert and LLM. But some new transformers may require implement some unseen operators.