Enable pyTorch-IMage-Models (TIMM) with HPUs

[ZhengHongming888]

What does this PR do?

This PR contains the scripts that showcases how to inference/fine-tune the TIMM models on intel's HPUs with the lazy/graph modes. We support the training for single/multiple HPU cards both two. Currently we support several most downloadable models from Hugging Face as below list.

• timm/resnet50.a1_in1k
• timm/resnet18.a1_in1k
• timm/resnet18.fb_swsl_ig1b_ft_in1k
  ...
• timm/vit_base_patch16_224.augreg2_in21k_ft_in1k
• timm/vgg19.tv_in1k

Here we support the below features:

1. Single-HPU training
   Here we show how to fine-tune the imagenette2-320 dataset and model with timm/resnet50.a1_in1k from Hugging Face.

1. training with hpu lazy mode

python train_hpu_lazy.py \
    --data-dir ./imagenette2-320/ \
    --device 'hpu' \
    --model resnet50.a1_in1k

2. training with hpu graph mode

python train_hpu_graph.py \
    --data-dir ./imagenette2-320/ \
    --device 'hpu' \
    --model resnet50.a1_in1k

2. Multi-HPU training

1. training with hpu lazy mode

torchrun --nnodes 1 --nproc_per_node 2 \
    train_hpu_lazy.py \
    --data-dir ./imagenette2-320/ \
    --device 'hpu' \
    --model resnet50.a1_in1k

2. training with hpu graph mode

torchrun --nnodes 1 --nproc_per_node 2 \
    train_hpu_graph.py \
    --data-dir ./imagenette2-320/ \
    --device 'hpu' \
    --model resnet50.a1_in1k

3. Single HPU inference

1. hpu with graph_mode

python inference.py \
    --data-dir='./download_ds/imagenette2-320' \
    --device='hpu' \
    --model resnet50.a1_in1k \
    --graph_mode

2. hpu with lazy mode

python inference.py \
    --data-dir='./download_ds/imagenette2-320' 
    --device='hpu' \
    --model resnet50.a1_in1k

Welcome for any suggestions/comments. Thanks.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[github-actions]

The code quality check failed, please run make style.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[regisss]

@ZhengHongming888 Can you add a link to the scripts these examples are inspired of please? You can put it in the docstring at the beginning of each file

[ZhengHongming888]

@ZhengHongming888 Can you add a link to the scripts these examples are inspired of please? You can put it in the docstring at the beginning of each file

@regisss updated please check! Thanks.

[regisss]

Finally can you create a test_timm.py file in https://github.com/huggingface/optimum-habana/tree/main/tests where you quickly test the example scripts please?
For example just running a few training/inference steps to make sure the examples work.

[ZhengHongming888]

Finally can you create a test_timm.py file in https://github.com/huggingface/optimum-habana/tree/main/tests where you quickly test the example scripts please? For example just running a few training/inference steps to make sure the examples work.

Thanks @regisss Done please check! Thanks.

[regisss]

LGTM!

[ZhengHongming888]

LGTM!

@regisss @libinta Thanks much for your time on review!!