{
  "repo_id": "FlyinGodzilla/AMOR-Mamba2-180M",
  "source_filename": "amor_180m_step_0007503.pt",
  "source_sha256": "8edd3ba619712905e8a00aaa7e6b97de6cc96d6b8a11ec2e24b5107e102b08bd",
  "step": 7503,
  "tokens_seen": 3687948288,
  "n_params": 184396896,
  "tensor_dtypes": {
    "torch.float32": 182
  },
  "gate_buffers": {
    "0": {
      "running_threshold": 0.7798937559127808,
      "running_std": 0.08388833701610565
    },
    "1": {
      "running_threshold": 0.6911748051643372,
      "running_std": 0.12514136731624603
    },
    "2": {
      "running_threshold": 0.7131258845329285,
      "running_std": 0.11415675282478333
    }
  },
  "optimizer_included": false,
  "precision_changed": false,
  "validation": {
    "exported_tensors_equal": true,
    "strict_model_load": true,
    "generation": true,
    "device": "cpu",
    "dtype": "float32",
    "torch": "2.6.0",
    "transformers": "4.51.3",
    "tokenizer_revision": "5590cc63198b0f111f4289daf006da0a12044173",
    "prompt": "The Eiffel Tower is",
    "generated_token_ids": [
      264,
      220,
      16,
      11
    ],
    "prefill_max_abs_diff": 0.0,
    "decode_max_abs_diff": 1.71661376953125e-05,
    "seconds": 22.35716986656189,
    "canonical_forward_exact_match": true,
    "canonical_forward_max_abs_diff": 0.0,
    "cuda": {
      "versions": {
        "torch": "2.6.0+cu124",
        "transformers": "5.0.0",
        "mamba-ssm": "2.3.0",
        "causal-conv1d": "1.6.0",
        "flash-linear-attention": "0.4.2",
        "triton": "3.2.0",
        "safetensors": "0.7.0"
      },
      "tokenizer_revision": "5590cc63198b0f111f4289daf006da0a12044173",
      "strict_model_load": true,
      "generation": true,
      "device": "NVIDIA H100 NVL",
      "torch": "2.6.0+cu124",
      "autocast": "bfloat16",
      "prefill_max_abs_diff": 0.0,
      "decode_max_abs_diff": 0.0625,
      "generated_token_ids": [
        264,
        1949,
        323,
        1825,
        2592,
        2447,
        430,
        6276
      ],
      "published_cli_passed": true,
      "test_scope": "Short CUDA smoke test; not a benchmark rerun",
      "comparison_tolerance": {
        "rtol": 0.02,
        "atol": 0.1
      },
      "tested_decoder_sha256": "4b48db8e5fb67318244c533356903f40703156cdca24ccf4fd49086e72b16db7"
    },
    "router_release": {
      "exact_router_export": true,
      "base_model_association": true,
      "default_forward_max_abs_diff": 0.0,
      "modes": [
        {
          "router": false,
          "cases": [
            {
              "batch": 1,
              "length": 16,
              "prefill_max_abs_diff": 6.103515625e-05,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 1,
              "length": 128,
              "prefill_max_abs_diff": 0.0,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 2,
              "length": 64,
              "prefill_max_abs_diff": 0.0
            }
          ],
          "canonical_router_exact": null
        },
        {
          "router": true,
          "cases": [
            {
              "batch": 1,
              "length": 16,
              "prefill_max_abs_diff": 6.103515625e-05,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 1,
              "length": 128,
              "prefill_max_abs_diff": 0.0,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 2,
              "length": 64,
              "prefill_max_abs_diff": 0.0
            }
          ],
          "canonical_router_exact": true
        }
      ],
      "frozen_thresholds_preserved": true,
      "training_router_rejected": true,
      "gpu": "NVIDIA H100 NVL",
      "torch": "2.6.0+cu124",
      "generation_cli_modes_passed": [
        "entropy",
        "router"
      ]
    },
    "current_release": {
      "strict_model_load": true,
      "entropy_and_router_generation_passed": true,
      "unit_tests_passed": 17,
      "device": "NVIDIA H100 NVL"
    }
  },
  "implementation": "AMOR shared release: native entropy gate and optional fitted inference router",
  "router": {
    "variant": "distill",
    "hidden_dim": 512,
    "activation": "silu",
    "weights": "router.safetensors",
    "config": "router_config.json",
    "sha256": "5cdb826237fb235af5a63a9d56cdcc63bdeab1b4cc1027970420f5aabadb8c48",
    "base_weight_sha256": "9857e8219052f0272aaf6a9801029dd8e1f51b668cbad3e1aadfc3a109b4aacb",
    "precision": "float32",
    "inference_only": true
  }
}