{
  "op": "com.microsoft.MatMulNBitsQkv",
  "fixtureArrays": {
    "norm_gqa_input_vBT": [90, 17, 203, 156, 64, 241, 112, 38, 175, 229, 83, 11, 198, 147, 52, 220, 105]
  },
  "cases": [
    {
      "name": "norm_gqa",
      "provenance": {
        "notes": "Grouped-query shape: Nq is twice Nkv, so the three projections have two different column counts."
      },
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [3, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "norm_rows_past_one_tile",
      "provenance": {
        "notes": "Ten activation rows against an eight-row projection tile: two row groups where the second carries only two real rows, so the store guard has to drop the six that clamped onto the last row."
      },
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [10, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.31, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [10, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [10, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [10, 4], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "skip_gqa",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32, "epsilon": 0.00001 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "skipT": {
          "dtype": "float32",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.11, "scale": 0.5 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [3, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "skipsum_gqa",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "skipT": {
          "dtype": "float32",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.11, "scale": 0.5 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [3, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "residualT": { "dtype": "float32", "shape": [3, 32], "tolerance": 0.000001, "relTolerance": 0.000001 }
      }
    },
    {
      "name": "decode_single_row",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [1, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "skipT": {
          "dtype": "float32",
          "shape": [1, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.11, "scale": 0.5 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [1, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [1, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [1, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "residualT": { "dtype": "float32", "shape": [1, 32], "tolerance": 0.000001, "relTolerance": 0.000001 }
      }
    },
    {
      "name": "equal_nq_nkv",
      "provenance": {
        "notes": "Multi-head shape where Nq equals Nkv; the tile grid still lays Q, K and V out in sequence."
      },
      "attrs": { "K": 32, "Nq": 8, "Nkv": 8, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [2, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [2, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [2, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [2, 8], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "tiles_not_aligned",
      "provenance": {
        "notes": "Neither column count is a multiple of the column tile, so both projections have a partial last tile."
      },
      "attrs": { "K": 32, "Nq": 5, "Nkv": 3, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [2, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [5, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [5, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [3, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [3, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [3, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [3, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [2, 5], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [2, 3], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [2, 3], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "tail_block_partial_k",
      "provenance": { "notes": "K is not a multiple of block_size, so the final block is partly out of range." },
      "attrs": { "K": 20, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [2, 20],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [20],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [2, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [2, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [2, 4], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "rank3_activation",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [1, 4, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [1, 4, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [1, 4, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [1, 4, 4], "tolerance": 0.0001, "relTolerance": 0.0001 }
      }
    },
    {
      "name": "two_quant_blocks",
      "provenance": { "notes": "K=64 covers two of the provider's supported 32-element quantization blocks." },
      "attrs": { "K": 64, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float32",
          "shape": [2, 64],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "skipT": {
          "dtype": "float32",
          "shape": [2, 64],
          "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.11, "scale": 0.5 }
        },
        "normScaleT": {
          "dtype": "float32",
          "shape": [64],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 2, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float32",
          "shape": [8, 2],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 2, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float32",
          "shape": [4, 2],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 2, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float32",
          "shape": [4, 2],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float32", "shape": [2, 8], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "kT": { "dtype": "float32", "shape": [2, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "vT": { "dtype": "float32", "shape": [2, 4], "tolerance": 0.0001, "relTolerance": 0.0001 },
        "residualT": { "dtype": "float32", "shape": [2, 64], "tolerance": 0.000001, "relTolerance": 0.000001 }
      }
    },
    {
      "name": "f16_skipsum",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float16",
          "shape": [2, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "skipT": {
          "dtype": "float16",
          "shape": [2, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.11, "scale": 0.5 }
        },
        "normScaleT": {
          "dtype": "float16",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float16",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float16",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float16",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float16", "shape": [2, 8], "tolerance": 0.002, "relTolerance": 0.01 },
        "kT": { "dtype": "float16", "shape": [2, 4], "tolerance": 0.002, "relTolerance": 0.01 },
        "vT": { "dtype": "float16", "shape": [2, 4], "tolerance": 0.002, "relTolerance": 0.01 },
        "residualT": { "dtype": "float16", "shape": [2, 32], "tolerance": 0.002, "relTolerance": 0.002 }
      }
    },
    {
      "name": "f16_norm",
      "attrs": { "K": 32, "Nq": 8, "Nkv": 4, "block_size": 32 },
      "inputs": {
        "aT": {
          "dtype": "float16",
          "shape": [3, 32],
          "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29, "scale": 0.8 }
        },
        "normScaleT": {
          "dtype": "float16",
          "shape": [32],
          "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 0.4, "offset": 1.0 }
        },
        "qBT": {
          "dtype": "uint8",
          "shape": [8, 1, 16],
          "data": { "kind": "cycle", "values": [27, 180, 75, 226, 33, 150, 201, 108, 57, 246, 129, 66, 195] }
        },
        "qScalesT": {
          "dtype": "float16",
          "shape": [8, 1],
          "data": { "kind": "cycle", "values": [0.04, 0.055, 0.05, 0.065, 0.06, 0.075, 0.07] }
        },
        "kBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": [211, 44, 137, 98, 165, 20, 233, 121, 78, 190, 15, 252, 87, 143, 61] }
        },
        "kScalesT": {
          "dtype": "float16",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.045, 0.03, 0.07, 0.05, 0.08, 0.035] }
        },
        "vBT": {
          "dtype": "uint8",
          "shape": [4, 1, 16],
          "data": { "kind": "cycle", "values": { "$ref": "#/fixtureArrays/norm_gqa_input_vBT" } }
        },
        "vScalesT": {
          "dtype": "float16",
          "shape": [4, 1],
          "data": { "kind": "cycle", "values": [0.052, 0.038, 0.061, 0.029, 0.073] }
        }
      },
      "outputs": {
        "qT": { "dtype": "float16", "shape": [3, 8], "tolerance": 0.002, "relTolerance": 0.01 },
        "kT": { "dtype": "float16", "shape": [3, 4], "tolerance": 0.002, "relTolerance": 0.01 },
        "vT": { "dtype": "float16", "shape": [3, 4], "tolerance": 0.002, "relTolerance": 0.01 }
      }
    }
  ]
}