{
  "run_name": "olmo3_1b_ddp",
  "world_size": 4,
  "per_rank_ga": 4,
  "global_tokens_per_update": 131072,
  "train_config_sha256": "bca830d8e52c95a0febdc5be3e017fd53cc9a7a889f7d8bd6051b78a42e3e08d",
  "data_manifest_sha256": "8f2266cc633cb5b103be3856bad7eb8437ed499f7be6ce1acd174f1eb4dc84d2",
  "code_commit": "72ef872e2e8e61938262f4fbccdadf6caf6ac1cf",
  "env": {
    "python": "3.11.15",
    "platform": "Linux-6.12.0-160000.34-default-x86_64-with-glibc2.40",
    "torch": "2.11.0+cu130",
    "numpy": "2.3.5",
    "cuda_available": true,
    "cuda_version": "13.0",
    "cudnn_version": 91900,
    "device_name": "NVIDIA RTX A6000",
    "env_vars": {
      "PYTHONHASHSEED": "0",
      "CUBLAS_WORKSPACE_CONFIG": ":4096:8",
      "TOKENIZERS_PARALLELISM": "false"
    }
  },
  "config": {
    "run_name": "olmo3_1b_ddp",
    "seed": 1337,
    "data_dir": "data/olmo3_prefix_hfstream_1b",
    "train_tokens": 1073741824,
    "stored_tokens_required": 1073741825,
    "block_size": 1024,
    "batch_size": 8,
    "global_tokens_per_update": 131072,
    "max_updates": 8192,
    "vocab_size": 100352,
    "n_layer": 12,
    "n_head": 12,
    "n_embd": 768,
    "mlp_ratio": 4,
    "dropout": 0.0,
    "bias": false,
    "weight_decay": 0.1,
    "learning_rate": 0.0003,
    "beta1": 0.9,
    "beta2": 0.95,
    "grad_clip": 1.0,
    "warmup_updates": 500,
    "min_lr": 3e-05,
    "device": "cuda",
    "dtype": "float32",
    "num_workers": 0,
    "checkpoint_interval_updates": 256,
    "log_interval_updates": 1
  }
}