transformers: Errors while training apple/mobilevit-xx-small on image-classification example with and without deepspeed

System Info

• transformers installed from source
• python 3.8
• ZeRO-Stage-1

Who can help?

@amyeroberts @NielsRogge @JingyaHuang

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, …)
• My own task or dataset (give details below)

Reproduction

python -m torch.distributed.launch --nproc_per_node=8 ~/transformers/examples/pytorch/image-classification/run_image_classification.py --model_name_or_path apple/mobilevit-xx-small --dataset_name beans --overwrite_output_dir --output_dir ./outputs/ --remove_unused_columns False --do_train --do_eval --learning_rate 2e-5 --num_train_epochs 50 --per_device_train_batch_size 8 --per_device_eval_batch_size 8 --logging_strategy steps --logging_steps 10 --evaluation_strategy epoch --seed 1337 --fp16 True --report_to none --ignore_mismatched_sizes True

AttributeError: ‘MobileViTImageProcessor’ object has no attribute ‘image_mean’

python -m torch.distributed.launch --nproc_per_node=8 ~/transformers/examples/pytorch/image-classification/run_image_classification.py --model_name_or_path apple/mobilevit-xx-small --dataset_name beans --overwrite_output_dir --output_dir ./outputs/ --remove_unused_columns False --do_train --do_eval --learning_rate 2e-5 --num_train_epochs 50 --per_device_train_batch_size 8 --per_device_eval_batch_size 8 --logging_strategy steps --logging_steps 10 --evaluation_strategy epoch --seed 1337 --fp16 True --report_to none --ignore_mismatched_sizes True --deepspeed ~/zero_stage_1.json

AttributeError: ‘MobileViTConfig’ object has no attribute ‘hidden_size’

Expected behavior

I expect both examples to train with the deepspeed-enabled run completing faster than baseline. Currently, both scenarios error out. Thank you in advance for the assistance.