{
  "repo_id": "FlyinGodzilla/AMOR-GatedDeltaNet-440M",
  "source_filename": "amor_gdn_440m_step_0018255.pt",
  "source_sha256": "ea5c2b88561dd7eb4527ba25c868a7ccea675baa1907e7a5156eb793cd95cb9a",
  "step": 18255,
  "tokens_seen": 8973017088,
  "n_params": 442133056,
  "tensor_dtypes": {
    "torch.float32": 458
  },
  "gate_buffers": {
    "0": {
      "running_threshold": 0.8792613744735718,
      "running_std": 0.043029818683862686
    },
    "1": {
      "running_threshold": 0.854272186756134,
      "running_std": 0.057880498468875885
    },
    "2": {
      "running_threshold": 0.8691568970680237,
      "running_std": 0.04676362872123718
    }
  },
  "optimizer_included": false,
  "precision_changed": false,
  "validation": {
    "exported_tensors_equal": true,
    "strict_model_load": true,
    "generation": true,
    "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,
        4382,
        7855,
        6576,
        430,
        649,
        387,
        1511
      ],
      "published_cli_passed": true,
      "test_scope": "Short CUDA smoke test; not a benchmark rerun",
      "comparison_tolerance": {
        "rtol": 0.02,
        "atol": 0.1
      }
    },
    "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": 0.0,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 1,
              "length": 128,
              "prefill_max_abs_diff": 0.0625,
              "decode_max_abs_diff": 0.125
            },
            {
              "batch": 2,
              "length": 64,
              "prefill_max_abs_diff": 0.0625
            }
          ],
          "canonical_router_exact": null
        },
        {
          "router": true,
          "cases": [
            {
              "batch": 1,
              "length": 16,
              "prefill_max_abs_diff": 0.0625,
              "decode_max_abs_diff": 0.0625
            },
            {
              "batch": 1,
              "length": 128,
              "prefill_max_abs_diff": 0.0,
              "decode_max_abs_diff": 0.125
            },
            {
              "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": "41a61cd94e307aebea14a430d99cb7e95f2bb5e1ddc3b2b1f7ef4beb87ab5081",
    "base_weight_sha256": "d2958bda89a6ecf036d00c545e9b36fb0d8562cb50d41b835397616a3d87530d",
    "precision": "float32",
    "inference_only": true
  },
  "weight_layout": {
    "format": "safetensors-sharded-v1",
    "index": "model.safetensors.index.json",
    "shards": 2,
    "weight_identity_sha256": "d2958bda89a6ecf036d00c545e9b36fb0d8562cb50d41b835397616a3d87530d",
    "original_single_file_sha256": "577c139d54e9e4e353fd729f62e246f0b012d356cd2fcdc650ddb6d1fae9331b",
    "all_tensor_bytes_identical": true
  }
}