training_config.json

{
  "model_variant": "0.6B",
  "train_file": "data/train_queries.csv",
  "output_dir": "models/qwen-finetuned",
  "run_name": "20251001-014002",
  "epochs": 5,
  "batch_size": 64,
  "learning_rate": 2e-05,
  "warmup_ratio": 0.1,
  "validation_split": 0.2,
  "eval_steps": 0,
  "checkpoint_steps": 0,
  "seed": 2025,
  "mixed_precision": false,
  "push_to_hub": true,
  "hub_model_id": "JacobLinCool/Qwen3-Embedding-0.6B-GIR-2",
  "hub_token": null,
  "hub_private_repo": false,
  "save_total_limit": 2,
  "logging_steps": null,
  "gradient_accumulation_steps": 1,
  "gradient_checkpointing": true,
  "optim": "paged_adamw_8bit"
}