config.yaml

dataset:
  name: wikidata5m_v3
  v1: false
  is_legacy: false
model:
  name: t5-small
  tokenizer_type: t5
  max_input_length: 512
  max_output_length: 40
context:
  use: true
  max_size: 100
  shuffle: true
descriptions:
  use: true
train:
  batch_size: 128
  max_epochs: 100
  drop_subject: 0.0
  num_workers: 10
  precision: 16
  accelerator: auto
  devices: auto
  strategy: ddp_find_unused_parameters_false
eval:
  num_predictions: 200
  max_length: 40
  batch_size: 1
  few_shot:
    use: false
    num_shots: 10
valid:
  every: 1
  tiny: true
checkpoint:
  keep_top_k: 3
resume_from: ''
wandb:
  use: true
  project_name: kgt5_context_${dataset.name}
  run_name: v1=${dataset.v1}_desc=${descriptions.use}_bs=${train.batch_size}