{
  "teacher": "deepseek-ai/DeepSeek-V2-Lite",
  "dataset": "HuggingFaceFW/fineweb-edu",
  "dataset_config": "sample-10BT",
  "max_steps": 200,
  "temperature": 1.0,
  "teacher_top_k": null,
  "kl_direction": "forward",
  "intermediate_loss": false,
  "intermediate_loss_layers": null,
  "learning_rate": 0.0001,
  "per_device_batch_size": 4,
  "gradient_accumulation_steps": 24,
  "max_length": 4096,
  "final_loss": 9.81964067697525,
  "global_step": 200,
  "student_init": "SF pruned/merged dense student, see https://github.com/krafton-ai/moe-to-dense"
}