training_config.json

{
  "config": {
    "model": {
      "dtype": "f32",
      "experts_per_tok": 1,
      "hidden_size": 128,
      "max_seq_len": 64,
      "name": "nano-start",
      "num_experts": 2,
      "num_heads": 4,
      "num_layers": 4,
      "vocab_size": 100315
    },
    "trainer": {
      "batch_size": 4,
      "effective_batch_size": 8,
      "gradient_accumulation": 2,
      "learning_rate": 0.002,
      "max_steps": 0,
      "num_epochs": 20,
      "seq_len": 64,
      "total_steps": 260
    }
  },
  "dataset_size": 6379,
  "device": "Cuda(CudaDevice(DeviceId(1)))",
  "error": null,
  "run_dir": "./runs/20251205_144741",
  "status": "completed"
}