params-deberta-large.json

{
  "dataset_args": {
    "max_utterances": 20,
    "max_tokens": 512,
    "dataroot": "data/all/fact_full/nlu"
  },
  "task": "linking",
  "model_name_or_path": "microsoft/deberta-v3-large",
  "per_gpu_train_batch_size": 2,
  "per_gpu_eval_batch_size": 2,
  "gradient_accumulation_steps": 1,
  "learning_rate": 2e-07,
  "adam_epsilon": 1e-08,
  "max_grad_norm": 1.0,
  "num_train_epochs": 20,
  "warmup_steps": 0,
  "fp16": "",
  "seed": 47
}