{
  "arms": [
    "public_native",
    "public_native_runtime",
    "optimizer_plus"
  ],
  "committed_steps": 5000,
  "data_artifact_sha256": "f2ef758ab83beb7b925ab5996e3a76aaf7db8fc467521ec16ae4d9734e05b98d",
  "evaluation": {
    "endpoint_examples": 256,
    "online_examples": 64,
    "retention_examples": 128
  },
  "evaluation_every_steps": 250,
  "families": {
    "adamw": {
      "learning_rates": {
        "high": 0.0002,
        "low": 5e-05,
        "medium": 0.0001
      },
      "settings": {
        "betas": [
          0.9,
          0.95
        ],
        "epsilon": 1e-08,
        "weight_decay": 0.01
      }
    },
    "lion": {
      "learning_rates": {
        "high": 5e-05,
        "low": 3e-06,
        "medium": 1e-05
      },
      "settings": {
        "betas": [
          0.9,
          0.99
        ],
        "weight_decay": 0.01
      }
    },
    "muon": {
      "learning_rates": {
        "high": 0.0001,
        "low": 3e-05,
        "medium": 6e-05
      },
      "settings": {
        "adamw_fallback_lr": 3e-05,
        "adjust_lr_fn": "match_rms_adamw",
        "momentum": 0.95,
        "nesterov": true,
        "ns_steps": 5,
        "weight_decay": 0.01
      }
    }
  },
  "model": {
    "adaptation": "LoRA",
    "base_dtype": "bfloat16",
    "id": "Qwen/Qwen2.5-3B",
    "lora_alpha": 32,
    "lora_dropout": 0.0,
    "lora_rank": 16,
    "micro_batch_size": 2,
    "revision": "3aab1f1954e9cc14eb9509a215f9e5ca08227a9b",
    "sequence_length": 512
  },
  "objective": "continued pretraining",
  "proprietary_boundary": "Controller rules, actuator coefficients, runtime implementation, and product code are excluded.",
  "seeds": [
    2203,
    2213
  ],
  "study": "OptimizerPlus-RC1.02-Matrix54-3B-CPT-v1"
}