{
  "schema_version": 1,
  "reference": "transformers.FunAsrNanoEncoderStem/FunAsrNanoEncoderLayer",
  "transformers_commit": "48e7f65fb274172e15aa88875d780c67c37606c7",
  "modeling_sha256": "70f8230fd5a75e8a119cea1c8dc919c3f51d8aa7c5d8f1c23c5402b856ebebfb",
  "torch_version": "2.11.0+cu130",
  "data_file": "sanm_reference.bin",
  "data_format": "little-endian-float32",
  "data_sha256": "2149540b2a724efa8e3242933071163ae97fab5d85331a35b119330de090255e",
  "config": {
    "input_size": 12,
    "model_size": 8,
    "num_heads": 2,
    "ffn_size": 16,
    "fsmn_kernel_size": 3,
    "layer_norm_eps": 1e-05,
    "frames": 5
  },
  "blocks": {
    "projection": {
      "input": {
        "shape": [
          1,
          5,
          12
        ],
        "offset_f32": 0,
        "count": 60
      },
      "output": {
        "shape": [
          1,
          5,
          8
        ],
        "offset_f32": 60,
        "count": 40
      },
      "checkpoints": {
        "self_attn_layer_norm": {
          "shape": [
            1,
            5,
            12
          ],
          "offset_f32": 100,
          "count": 60
        }
      },
      "weights": {
        "fc1.bias": {
          "shape": [
            16
          ],
          "offset_f32": 160,
          "count": 16
        },
        "fc1.weight": {
          "shape": [
            16,
            8
          ],
          "offset_f32": 176,
          "count": 128
        },
        "fc2.bias": {
          "shape": [
            8
          ],
          "offset_f32": 304,
          "count": 8
        },
        "fc2.weight": {
          "shape": [
            8,
            16
          ],
          "offset_f32": 312,
          "count": 128
        },
        "feedforward_sequential_memory.conv.weight": {
          "shape": [
            8,
            1,
            3
          ],
          "offset_f32": 440,
          "count": 24
        },
        "final_layer_norm.bias": {
          "shape": [
            8
          ],
          "offset_f32": 464,
          "count": 8
        },
        "final_layer_norm.weight": {
          "shape": [
            8
          ],
          "offset_f32": 472,
          "count": 8
        },
        "self_attn.k_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 480,
          "count": 8
        },
        "self_attn.k_proj.weight": {
          "shape": [
            8,
            12
          ],
          "offset_f32": 488,
          "count": 96
        },
        "self_attn.out_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 584,
          "count": 8
        },
        "self_attn.out_proj.weight": {
          "shape": [
            8,
            8
          ],
          "offset_f32": 592,
          "count": 64
        },
        "self_attn.q_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 656,
          "count": 8
        },
        "self_attn.q_proj.weight": {
          "shape": [
            8,
            12
          ],
          "offset_f32": 664,
          "count": 96
        },
        "self_attn.v_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 760,
          "count": 8
        },
        "self_attn.v_proj.weight": {
          "shape": [
            8,
            12
          ],
          "offset_f32": 768,
          "count": 96
        },
        "self_attn_layer_norm.bias": {
          "shape": [
            12
          ],
          "offset_f32": 864,
          "count": 12
        },
        "self_attn_layer_norm.weight": {
          "shape": [
            12
          ],
          "offset_f32": 876,
          "count": 12
        }
      }
    },
    "residual": {
      "input": {
        "shape": [
          1,
          5,
          8
        ],
        "offset_f32": 888,
        "count": 40
      },
      "output": {
        "shape": [
          1,
          5,
          8
        ],
        "offset_f32": 928,
        "count": 40
      },
      "checkpoints": {
        "self_attn_layer_norm": {
          "shape": [
            1,
            5,
            8
          ],
          "offset_f32": 968,
          "count": 40
        }
      },
      "weights": {
        "fc1.bias": {
          "shape": [
            16
          ],
          "offset_f32": 1008,
          "count": 16
        },
        "fc1.weight": {
          "shape": [
            16,
            8
          ],
          "offset_f32": 1024,
          "count": 128
        },
        "fc2.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1152,
          "count": 8
        },
        "fc2.weight": {
          "shape": [
            8,
            16
          ],
          "offset_f32": 1160,
          "count": 128
        },
        "feedforward_sequential_memory.conv.weight": {
          "shape": [
            8,
            1,
            3
          ],
          "offset_f32": 1288,
          "count": 24
        },
        "final_layer_norm.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1312,
          "count": 8
        },
        "final_layer_norm.weight": {
          "shape": [
            8
          ],
          "offset_f32": 1320,
          "count": 8
        },
        "self_attn.k_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1328,
          "count": 8
        },
        "self_attn.k_proj.weight": {
          "shape": [
            8,
            8
          ],
          "offset_f32": 1336,
          "count": 64
        },
        "self_attn.out_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1400,
          "count": 8
        },
        "self_attn.out_proj.weight": {
          "shape": [
            8,
            8
          ],
          "offset_f32": 1408,
          "count": 64
        },
        "self_attn.q_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1472,
          "count": 8
        },
        "self_attn.q_proj.weight": {
          "shape": [
            8,
            8
          ],
          "offset_f32": 1480,
          "count": 64
        },
        "self_attn.v_proj.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1544,
          "count": 8
        },
        "self_attn.v_proj.weight": {
          "shape": [
            8,
            8
          ],
          "offset_f32": 1552,
          "count": 64
        },
        "self_attn_layer_norm.bias": {
          "shape": [
            8
          ],
          "offset_f32": 1616,
          "count": 8
        },
        "self_attn_layer_norm.weight": {
          "shape": [
            8
          ],
          "offset_f32": 1624,
          "count": 8
        }
      }
    }
  }
}