{
  "model_name_or_path": "jhu-clsp/mmBERT-base",
  "rope_scaling_type": "yarn",
  "rope_original_max_position_embeddings": 8192,
  "yarn_beta_fast": 32.0,
  "yarn_beta_slow": 1.0,
  "yarn_extrapolation_factor": 1.0,
  "dataset_path": "/data/modernbert-32k-training/tokenized_dataset_32k",
  "model_max_length": 32768,
  "mlm_probability": 0.3,
  "max_train_samples": null,
  "dataloader_num_workers": 4,
  "output_dir": "/data/modernbert-32k-training/output/mmbert-32k-yarn",
  "per_device_train_batch_size": 1,
  "gradient_accumulation_steps": 16,
  "learning_rate": 1e-05,
  "weight_decay": 0.01,
  "warmup_steps": 100,
  "warmup_ratio": 0.1,
  "num_train_epochs": 1,
  "max_grad_norm": 1.0,
  "lr_scheduler_type": "constant_with_warmup",
  "use_retrieval_masking": false,
  "retrieval_probability": 0.1,
  "min_distance_for_retrieval": 512,
  "use_ewc": false,
  "ewc_lambda": 1000.0,
  "ewc_samples": 200,
  "bf16": true,
  "logging_steps": 10,
  "save_steps": 500,
  "seed": 42
}