donut-booking-gradio / config /train_cord.yaml
uartimcs's picture
Upload folder using huggingface_hub
1cc5dab verified
raw
history blame
940 Bytes
resume_from_checkpoint_path: null # only used for resume_from_checkpoint option in PL
result_path: "./result"
pretrained_model_name_or_path: "naver-clova-ix/donut-base" # loading a pre-trained model (from moldehub or path)
dataset_name_or_paths: ["naver-clova-ix/cord-v2"] # loading datasets (from moldehub or path)
sort_json_key: False # cord dataset is preprocessed, and publicly available at https://huggingface.co/datasets/naver-clova-ix/cord-v2
train_batch_sizes: [8]
val_batch_sizes: [1]
input_size: [1280, 960] # when the input resolution differs from the pre-training setting, some weights will be newly initialized (but the model training would be okay)
max_length: 768
align_long_axis: False
num_nodes: 1
seed: 2022
lr: 3e-5
warmup_steps: 300 # 800/8*30/10, 10%
num_training_samples_per_epoch: 800
max_epochs: 30
max_steps: -1
num_workers: 8
val_check_interval: 1.0
check_val_every_n_epoch: 3
gradient_clip_val: 1.0
verbose: True