mode: training
job_name: bart_large
job_dir: /poster/jobs/bart_large
seed: 1
cuda_visible_gpu: '0'
ckpt_dir: null
checkpoint_name: best
debug: false
use_wandb: false
wandb_project_name: SciPosterLayout
write_preidiction_epoch: 10
backbone:
  _target_: trainer.models.transformer_utils.TransformerEncoder
  encoder_layer:
    _target_: trainer.models.transformer_utils.Block
    d_model: 512
    nhead: 8
    dim_feedforward: 2048
    dropout: 0.1
    batch_first: true
    norm_first: true
  num_layers: 12
dataset:
  _target_: trainer.datasets.posterGlobal.posterGlobalDataset
  _partial_: true
  dir: /poster/data
  max_seq_length: 10
model:
  _target_: trainer.models.bart.BART
  _partial_: true
extra_model_params:
  use_size_position_decoders: false
  category_loss: true
  multi_decoder: false
data:
  batch_size: 16
  num_workers: 4
  num_bin_bboxes: 32
  bbox_quantization: linear
  shared_bbox_vocab: xywh
  pad_until_max: true
  special_tokens:
  - pad
  - bos
  - eos
  - mask
  transforms: []
  var_order: c-w-h-x-y
training:
  epochs: 3000
  grad_norm_clip: 1.0
  weight_decay: 0.1
  loss_plot_iter_interval: 50
optimizer:
  _target_: torch.optim.AdamW
  _partial_: true
  lr: 0.001
  betas:
  - 0.9
  - 0.98
scheduler:
  _target_: trainer.helpers.scheduler.VoidScheduler
  _partial_: true
sampling:
  name: random
  temperature: 1.0