modernBERT-large.yaml

# yaml-language-server: $schema=../../config_schema.json

task: question-answering
base_model: /home/praise/PycharmProjects/encoder_playground/ModernBERT-large

data:
  path: rajpurkar/squad_v2
  train_split: train # this must be either train.csv or train.json
  test_split: validation  # this must be either valid.csv or valid.json

trainer:
  trainer_name: "hf_trainer"
  hf_trainer_args:
    eval_steps: 200
    save_steps: 200
    logging_steps: 200
    optim: adamw_torch
    num_train_epochs: 4
    learning_rate: 1.0E-5
    metric_for_best_model: "f1"
    greater_is_better: true
    per_device_train_batch_size: 8
    per_device_eval_batch_size: 8
    output_dir: /home/praise/PycharmProjects/encoder_playground/runs/ModernBERT-large-squad2-v0.3
    bf16: true
    warmup_ratio: 0.1
    gradient_accumulation_steps: 8

  extra_trainer_args:
    early_stopping_patience: 20
    early_stopping_threshold: 0.001

huggingface_hub_username: "Praise2112"