train_config.json

{
  "data": {
    "input_jsonl_file": "D:/WorkSpace/AnglE_yj/data_preparation/Aspect-Polarity_Pair/output/v2/asap_angle_contextual_ap_data_hybrid.jsonl",
    "output_dir": "checkpoints_learnable/v3_all1"
  },
  "model": {
    "name": "hfl/chinese-roberta-wwm-ext",
    "max_length": 192,
    "pooling_strategy": "cls"
  },
  "training": {
    "batch_size": 32,
    "gradient_accumulation_steps": 4,
    "num_epochs": 5,
    "learning_rate": 2e-05,
    "warmup_steps": 500,
    "save_steps": 1000,
    "logging_steps": 100,
    "fp16": true
  },
  "loss": {
    "cosine_w": 1.0,
    "ibn_w": 1.0,
    "angle_w": 1.0,
    "cosine_tau": 20,
    "ibn_tau": 20,
    "angle_tau": 1
  }
}