Colab Notebook for ACT - Update training time with A100

[gautz]

What this does

It took me about 11 hours on Google Colab to train ACT with A100 80 GB GDDR for 60k steps (80 credits) .
That make 18 hours (and more than 100 credits) for 100k steps.

How it was tested

see https://api.wandb.ai/links/hentz-robotics/wdsr4k2r
https://huggingface.co/gautz/act_so101_pick_red_18650_drop_black_box_test2_100ep_64batch_60ksteps
https://huggingface.co/datasets/gautz/so101_pick_red_18650_drop_black_box_test2_100ep

How to checkout & try? (for the reviewer)

see https://innovation.iha.unistra.fr/books/robotique-open-source/page/machine-learning-lerobot-avec-so-arm101#bkmrk-google-colab

Basically run notebook https://colab.research.google.com/github/huggingface/notebooks/blob/main/lerobot/training-act.ipynb#scrollTo=NQUk3Y0WwYZ4
with:

!cd lerobot && python src/lerobot/scripts/lerobot_train.py \
  --dataset.repo_id=gautz/so101_pick_red_18650_drop_black_box_test2_100ep \
  --policy.type=act \
  --output_dir=outputs/train/act_so101_pick_red_18650_drop_black_box_test2_100ep_64batch \
  --job_name=act_so101_pick_red_18650_drop_black_box_test2_100ep_64batch \
  --policy.device=cuda \
  --policy.push_to_hub=True \
  --policy.repo_id=gautz/act_so101_pick_red_18650_drop_black_box_test2_100ep_64batch \
  --batch_size=64 \
  --wandb.enable=true