SuperposPrompt_Thesis/04_LowerDimPrompt/config.yaml

shared:
  project_name: lowdim_prompts
  use_tqdm: true
  random_seed: 42

default: &default
  model_name: google/t5-large-lm-adapt
  wandb_name: null
  train_batch_size: 32
  valid_batch_size: 32
  num_epochs: 200
  peft_params: null  # no mutation
  hot_modules: null  # fine-tune all
  balancify_train: false
  best_finder:
    save: true
    metric: valid_f1-score-ma
    higher_better: true
  tasks:
  - glue:cola

run_configs:
# - <<: *default
#   wandb_name: n_tokens100_n_comb_tokens512
#   learning_rate: 0.01
#   hot_modules:
#   - sadcl
#   peft_params:
#     kind: comb_prompt
#     n_tokens: 100
#     n_comb_tokens: 512
# - <<: *default
#   wandb_name: n_tokens100_n_comb_tokens2048
#   learning_rate: 0.01
#   hot_modules:
#   - sadcl
#   peft_params:
#     kind: comb_prompt
#     n_tokens: 100
#     n_comb_tokens: 2048
- <<: *default
  wandb_name: large_n_tokens100_64_256
  learning_rate: 0.01
  hot_modules:
  - sadcl
  peft_params:
    kind: lowdim_prompt
    n_tokens: 100
    dims:
    - 64
    - 256
- <<: *default
  wandb_name: large_n_tokens100_256_512
  learning_rate: 0.01
  hot_modules:
  - sadcl
  peft_params:
    kind: lowdim_prompt
    n_tokens: 100
    dims:
    - 256
    - 512