transformer_image_fewshot.yaml

project: HINT_image_fewshot
program: train.py
method: grid
metric:
  name: test/result_acc/avg
  goal: maximize
parameters:
  save_model:
    value: True
  fewshot:
    # value: 'a'
    values: ['x', 'y', 'a', 'b', 'c', 'd']
  # max_op_train:
  #   values: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
  main_dataset_ratio:
    value: 0.5
  input:
    value: image
  resume:
    value: models/image.TRAN.relative_universal/model_100000.p
  perception_pretrain:
    value: ""
    # values: ["", "imagenet"]
  model:
    value: TRAN.relative_universal
    # values: [TRAN.relative, TRAN.opennmt]
    # values: [TRAN.relative_universal, TRAN.relative, TRAN.opennmt]
  result_encoding:
    value: decimal
    # values: [decimal, sin, binary]
  batch_size:
    value: 128
  lr:
    value: 0.00001
    # values: [0.001, 0.0001, 0.00001]
  iterations:
    value: 1000
  iterations_eval:
    value: 100
  early_stop:
    value: 50
  dropout:
    value: 0.1
  nhead:
    value: 8
    # values: [4, 8]
  # layers:
    # value: 6
    # values: [3, 6, 9]
  enc_layers:
    value: 6
    # values: [3, 6, 9, 12]
  dec_layers:
    value: 1
    # values: [1, 3, 6]
  hid_dim:
    value: 512
    # values: [128, 256, 512]
  max_rel_pos:
    value: 15
  seed:
    values: [1,2]