config_template_mimir.yaml

# Global configuration
global:
  target_model: "EleutherAI/pythia-6.9b-deduped"
  datasets:
    - mimir_name: "arxiv"
      split: "ngram_13_0.8"
    - mimir_name: "pile_cc"
      split: "ngram_13_0.8"
  batch_size: 32
  device: "cuda:0"
  fpr_thresholds:
    - 0.1
    - 0.01
  n_bootstrap_samples: 10

bag_of_words:
  module: bag_of_words
  test_size: 0.2
  min_df: 0.05
  n_estimators: 100
  max_depth: 2
  min_samples_leaf: 5
  seed: 42

loss:
  module: loss

zlib:
  module: zlib

lowercase:
  module: lowercase
  batch_size: 32

ratio:
  module: ratio
  reference_model_path: ""
  reference_tokenizer_path: ""
  batch_size: 32
  device: "cuda:0"

neighborhood:
  module: neighborhood
  batch_size: 32
  mlm_model: 'roberta-base'
  n_neighbors: 50
  top_k: 10
  is_scale_embeds: true
  device: 'cuda:0'

samia:
  module: samia
  rouge_version: paper
  n: 1
  zlib: false
  batch_size: 16
  prefix_ratio: 0.5
  max_length: 1024
  n_candidates: 10
  temperature: 1.0
  top_k: 50
  top_p: 1
 
pac_10:
  module: pac
  k_min: 0.3
  k_max: 0.05
  alpha: 0.3
  num_augmentations: 10
 
surp_40_2:
  module: surp
  k: 40
  max_entropy: 2.0
  batch_size: 32

minkprob:
  module: minkprob
  k: 20
  batch_size: 32

minkplusplus:
  module: minkplusplus
  k: 20
  batch_size: 32