{ "cases": [ { "name": "linear_state_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.117, "cosStep": 0.297, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.137, "cosStep": 0.317, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.337, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.397, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "linear_state_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.124, "cosStep": 0.304, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.144, "cosStep": 0.324, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.164, "cosStep": 0.344, "scale": 0.32 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.224, "cosStep": 0.404, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "linear_zero_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.131, "cosStep": 0.311, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.151, "cosStep": 0.331, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.171, "cosStep": 0.351, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "linear_zero_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.138, "cosStep": 0.318, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.158, "cosStep": 0.338, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.178, "cosStep": 0.358, "scale": 0.32 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_state_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.145, "cosStep": 0.325, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.165, "cosStep": 0.345, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.185, "cosStep": 0.365, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.205, "cosStep": 0.385, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.245, "cosStep": 0.425, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_state_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.152, "cosStep": 0.332, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.172, "cosStep": 0.352, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.192, "cosStep": 0.372, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.212, "cosStep": 0.392, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.252, "cosStep": 0.432, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_zero_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.159, "cosStep": 0.339, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.179, "cosStep": 0.359, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.199, "cosStep": 0.379, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.219, "cosStep": 0.399, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_zero_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.166, "cosStep": 0.346, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.186, "cosStep": 0.366, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.206, "cosStep": 0.386, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.226, "cosStep": 0.406, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_qwen_state_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.173, "cosStep": 0.353, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.193, "cosStep": 0.373, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.213, "cosStep": 0.393, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.233, "cosStep": 0.413, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.273, "cosStep": 0.453, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.293, "cosStep": 0.473, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.313, "cosStep": 0.493, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_qwen_state_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.18, "cosStep": 0.36, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.2, "cosStep": 0.38, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.22, "cosStep": 0.4, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.24, "cosStep": 0.42, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.28, "cosStep": 0.46, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.3, "cosStep": 0.48, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.32, "cosStep": 0.5, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_qwen_zero_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.367, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.207, "cosStep": 0.387, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.227, "cosStep": 0.407, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.427, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.487, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.327, "cosStep": 0.507, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_qwen_zero_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.194, "cosStep": 0.374, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.214, "cosStep": 0.394, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.234, "cosStep": 0.414, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.254, "cosStep": 0.434, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.314, "cosStep": 0.494, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.334, "cosStep": 0.514, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "delta_state_ragged", "attrs": { "update_rule": "delta", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.201, "cosStep": 0.381, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.221, "cosStep": 0.401, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.241, "cosStep": 0.421, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.281, "cosStep": 0.461, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.301, "cosStep": 0.481, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "delta_state_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.208, "cosStep": 0.388, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.228, "cosStep": 0.408, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.248, "cosStep": 0.428, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.288, "cosStep": 0.468, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.308, "cosStep": 0.488, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "delta_zero_ragged", "attrs": { "update_rule": "delta", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.215, "cosStep": 0.395, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.235, "cosStep": 0.415, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.255, "cosStep": 0.435, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.295, "cosStep": 0.475, "scale": 0.6 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "delta_zero_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.222, "cosStep": 0.402, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.242, "cosStep": 0.422, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.262, "cosStep": 0.442, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.302, "cosStep": 0.482, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_state_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_state_packed", "attrs": { "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.236, "cosStep": 0.416, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.256, "cosStep": 0.436, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.276, "cosStep": 0.456, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.296, "cosStep": 0.476, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.316, "cosStep": 0.496, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.336, "cosStep": 0.516, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_zero_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.243, "cosStep": 0.423, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.263, "cosStep": 0.443, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.283, "cosStep": 0.463, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.303, "cosStep": 0.483, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.323, "cosStep": 0.503, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_zero_packed", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.25, "cosStep": 0.43, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.27, "cosStep": 0.45, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.29, "cosStep": 0.47, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.31, "cosStep": 0.49, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33, "cosStep": 0.51, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_qwen_state_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.257, "cosStep": 0.437, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.457, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.297, "cosStep": 0.477, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.317, "cosStep": 0.497, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.337, "cosStep": 0.517, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.357, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.377, "cosStep": 0.557, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.397, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_qwen_state_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.264, "cosStep": 0.444, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.284, "cosStep": 0.464, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.304, "cosStep": 0.484, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.324, "cosStep": 0.504, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.344, "cosStep": 0.524, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.364, "cosStep": 0.544, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.384, "cosStep": 0.564, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.404, "cosStep": 0.584, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_qwen_zero_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.271, "cosStep": 0.451, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.291, "cosStep": 0.471, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.311, "cosStep": 0.491, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.331, "cosStep": 0.511, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.351, "cosStep": 0.531, "scale": 0.2, "offset": 0.5 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.391, "cosStep": 0.571, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.411, "cosStep": 0.591, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gated_delta_qwen_zero_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.278, "cosStep": 0.458, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.298, "cosStep": 0.478, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.318, "cosStep": 0.498, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.338, "cosStep": 0.518, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.358, "cosStep": 0.538, "scale": 0.6 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.398, "cosStep": 0.578, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.418, "cosStep": 0.598, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 32] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "scalar_head_dim_gated_delta", "attrs": { "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [5, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.31, "cosStep": 0.49, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [5, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.33, "cosStep": 0.51, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [5, 2, 5], "data": { "kind": "fillFloat32", "sinStep": 0.35, "cosStep": 0.53, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [5, 2], "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.55, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [5, 2], "data": { "kind": "fillFloat32", "sinStep": 0.39, "cosStep": 0.57, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 5, 7], "data": { "kind": "fillFloat32", "sinStep": 0.41, "cosStep": 0.59, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [5, 2, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 5, 7] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "vec4_partial_lane_delta", "attrs": { "update_rule": "delta", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [6, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.35, "cosStep": 0.53, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [6, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.55, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [6, 3, 6], "data": { "kind": "fillFloat32", "sinStep": 0.39, "cosStep": 0.57, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [6, 3], "data": { "kind": "fillFloat32", "sinStep": 0.43, "cosStep": 0.61, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 3, 6, 20], "data": { "kind": "fillFloat32", "sinStep": 0.45, "cosStep": 0.63, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [6, 3, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 3, 6, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "wide_head_dim_gated_delta_qwen", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 128], "data": { "kind": "fillFloat32", "sinStep": 0.39, "cosStep": 0.57, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 128], "data": { "kind": "fillFloat32", "sinStep": 0.41, "cosStep": 0.59, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 2, 16], "data": { "kind": "fillFloat32", "sinStep": 0.43, "cosStep": 0.61, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.45, "cosStep": 0.63, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.47, "cosStep": 0.65, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 16, 128], "data": { "kind": "fillFloat32", "sinStep": 0.49, "cosStep": 0.67, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 2, 16] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 16, 128] } }, "relTolerance": 0.0008, "tolerance": 0.00002 }, { "name": "deep_words_gated_delta", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [3, 1, 256], "data": { "kind": "fillFloat32", "sinStep": 0.43, "cosStep": 0.61, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [3, 1, 256], "data": { "kind": "fillFloat32", "sinStep": 0.45, "cosStep": 0.63, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [3, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.47, "cosStep": 0.65, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.49, "cosStep": 0.67, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 1, 4, 256], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [3, 1, 4] }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 4, 256] } }, "relTolerance": 0.001, "tolerance": 0.00002 }, { "name": "value_tile_tail_gated_delta", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 12], "data": { "kind": "fillFloat32", "sinStep": 0.47, "cosStep": 0.65, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 12], "data": { "kind": "fillFloat32", "sinStep": 0.49, "cosStep": 0.67, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 2, 9], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.55, "cosStep": 0.73, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 9, 12], "data": { "kind": "fillFloat32", "sinStep": 0.57, "cosStep": 0.75, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 2, 9] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 9, 12] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "single_value_channel_qk_norm", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Recurrent_GenericSingleValueChannelWithQkNormalization", "note": "Geometry mirrored. Upstream pins no literal values; it compares against the in-file float64 reference that this package's CPU reference transcribes." }, "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 1, 1], "data": { "kind": "fillFloat32", "sinStep": 0.55, "cosStep": 0.73, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 1], "data": { "kind": "fillFloat32", "sinStep": 0.57, "cosStep": 0.75, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [4, 1], "data": { "kind": "fillFloat32", "sinStep": 0.59, "cosStep": 0.77, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 1, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.61, "cosStep": 0.79, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 1, 1] }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 1, 96] } }, "relTolerance": 0.0008, "tolerance": 0.00002 }, { "name": "identity_head_map_gated_delta", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [6, 4, 8], "data": { "kind": "fillFloat32", "sinStep": 0.55, "cosStep": 0.73, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [6, 4, 8], "data": { "kind": "fillFloat32", "sinStep": 0.57, "cosStep": 0.75, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [6, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.59, "cosStep": 0.77, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 0.61, "cosStep": 0.79, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 0.63, "cosStep": 0.81, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.65, "cosStep": 0.83, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [6, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "block_head_map_gated_delta", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [6, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.59, "cosStep": 0.77, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [6, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.61, "cosStep": 0.79, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [6, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.63, "cosStep": 0.81, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 0.65, "cosStep": 0.83, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [6, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "rank4_gated_delta_qwen_state", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.63, "cosStep": 0.81, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.65, "cosStep": 0.83, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "empty_request_ragged_gated_delta", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Ragged_CompactReplayMatchesSequentialPrefixes", "note": "Zero-length leading request, as the upstream ragged capture case uses. A request with no tokens copies its incoming state through." }, "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [5, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [5, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [5, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 0, 2, 5] } }, "decayT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [3, 4, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [5, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [3, 4, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "uneven_ragged_gated_delta_qwen", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Chunked_RaggedUnequalLengths", "note": "Ragged geometry mirrored at a fixture-sized head count; upstream has no literal expectations." }, "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [20, 1, 16], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [20, 1, 16], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [20, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 4, 15, 20] } }, "decayT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.79, "cosStep": 0.97, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [3, 2, 8, 16], "data": { "kind": "fillFloat32", "sinStep": 0.81, "cosStep": 0.99, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [20, 2, 8] }, "finalStateT": { "dtype": "float32", "shape": [3, 2, 8, 16] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "explicit_scale_gated_delta", "attrs": { "beta_activation": "none", "scale": 0.5 }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 2, 6], "data": { "kind": "fillFloat32", "sinStep": 0.79, "cosStep": 0.97, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.81, "cosStep": 0.99, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 2, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "chunk_size_hint_gated_delta", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Chunked_NarrowChunkForSmallSharedMemory", "note": "Non-default chunk_size witness: the attribute is a scheduling hint and must not change the result." }, "attrs": { "beta_activation": "none", "chunk_size": 32 }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.79, "cosStep": 0.97, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.81, "cosStep": 0.99, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 2, 6], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.87, "cosStep": 1.05, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.89, "cosStep": 1.07, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 2, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "batch3_uniform_gated_delta", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [9, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [9, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [9, 3, 6], "data": { "kind": "fillFloat32", "sinStep": 0.87, "cosStep": 1.05, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [9, 3], "data": { "kind": "fillFloat32", "sinStep": 0.89, "cosStep": 1.07, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [9, 3], "data": { "kind": "fillFloat32", "sinStep": 0.91, "cosStep": 1.09, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [3, 3, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.93, "cosStep": 1.11, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [9, 3, 6] }, "finalStateT": { "dtype": "float32", "shape": [3, 3, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "f16_gated_delta_qwen_state_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [4, 1, 16], "data": { "kind": "fillFloat32", "sinStep": 0.87, "cosStep": 1.05, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [4, 1, 16], "data": { "kind": "fillFloat32", "sinStep": 0.89, "cosStep": 1.07, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [4, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.91, "cosStep": 1.09, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.93, "cosStep": 1.11, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [4, 2], "data": { "kind": "fillFloat32", "sinStep": 0.95, "cosStep": 1.13, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 8, 16], "data": { "kind": "fillFloat32", "sinStep": 0.97, "cosStep": 1.15, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.99, "cosStep": 1.17, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 1.01, "cosStep": 1.19, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [4, 2, 8] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 8, 16] } }, "relTolerances": { "outputT": 0.02, "finalStateT": 0.006 }, "tolerances": { "outputT": 0.002, "finalStateT": 0.0006 } }, { "name": "f16_linear_zero_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float16", "shape": [6, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.91, "cosStep": 1.09, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [6, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.93, "cosStep": 1.11, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [6, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.95, "cosStep": 1.13, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 2, 6] } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [6, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 8] } }, "relTolerances": { "outputT": 0.02, "finalStateT": 0.006 }, "tolerances": { "outputT": 0.002, "finalStateT": 0.0006 } }, { "name": "f16_delta_state_packed", "attrs": { "update_rule": "delta", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [5, 1, 32], "data": { "kind": "fillFloat32", "sinStep": 0.95, "cosStep": 1.13, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [5, 1, 32], "data": { "kind": "fillFloat32", "sinStep": 0.97, "cosStep": 1.15, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [5, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.99, "cosStep": 1.17, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [5, 2], "data": { "kind": "fillFloat32", "sinStep": 1.03, "cosStep": 1.21, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 8, 32], "data": { "kind": "fillFloat32", "sinStep": 1.05, "cosStep": 1.23, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [5, 2, 8] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 8, 32] } }, "relTolerances": { "outputT": 0.02, "finalStateT": 0.006 }, "tolerances": { "outputT": 0.002, "finalStateT": 0.0006 } }, { "name": "f16_wide_head_gated_state_packed", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float16", "shape": [3, 1, 128], "data": { "kind": "fillFloat32", "sinStep": 0.99, "cosStep": 1.17, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [3, 1, 128], "data": { "kind": "fillFloat32", "sinStep": 1.01, "cosStep": 1.19, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [3, 2, 16], "data": { "kind": "fillFloat32", "sinStep": 1.03, "cosStep": 1.21, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [3, 2], "data": { "kind": "fillFloat32", "sinStep": 1.05, "cosStep": 1.23, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 16, 128], "data": { "kind": "fillFloat32", "sinStep": 1.09, "cosStep": 1.27, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [3, 2, 16] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 16, 128] } }, "relTolerances": { "outputT": 0.02, "finalStateT": 0.008 }, "tolerances": { "outputT": 0.002, "finalStateT": 0.0006 } }, { "name": "single_lane_key_gated_delta", "provenance": { "notes": "head_size_qk 8 packs into one lane group word set, so the recurrence renders no cross-lane fold." }, "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [6, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 1.03, "cosStep": 1.21, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [6, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 1.05, "cosStep": 1.23, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [6, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 1.07, "cosStep": 1.25, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 1.09, "cosStep": 1.27, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [6, 4], "data": { "kind": "fillFloat32", "sinStep": 1.11, "cosStep": 1.29, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 1.13, "cosStep": 1.31, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [6, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "single_lane_key_linear_zero", "provenance": { "notes": "The no-fold shape on the rule that reads no gate stream." }, "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 1.07, "cosStep": 1.25, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [4, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 1.09, "cosStep": 1.27, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [4, 2, 6], "data": { "kind": "fillFloat32", "sinStep": 1.11, "cosStep": 1.29, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 4] } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [4, 2, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 6, 8] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "gdn-general-float32-gated_delta-dk7-dv5-h1x2-b1-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [1, 33, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [1, 33, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [1, 33, 2, 5], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "decayT": { "dtype": "float32", "shape": [1, 33, 2], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 33, 2], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [1, 33, 2, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 5, 7] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-delta-dk20-dv6-h1x3-b1-t65-tile2", "attrs": { "update_rule": "delta", "qk_l2_norm": 1, "beta_activation": "sigmoid" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [65, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [65, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [65, 3, 6], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 3, 6, 20], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } }, "betaT": { "dtype": "float32", "shape": [65, 3], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [65, 3, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 3, 6, 20] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-linear-dk32-dv8-h2x6-b2-t17-tile2", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [34, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [34, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [34, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 8, 32], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [34, 6, 8] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 8, 32] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated-dk64-dv16-h2x4-b1-t129-tile2", "attrs": { "update_rule": "gated", "qk_l2_norm": 1, "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [1, 129, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [1, 129, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [1, 129, 4, 16], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 16, 64], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [1, 129, 4, 16] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 16, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk96-dv1-h1x1-b1-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [33, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [33, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [33, 1, 1], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 33] } }, "decayT": { "dtype": "float32", "shape": [33, 1], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [1], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [1], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [33, 1], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [33, 1, 1] }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 1, 96] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv18-h2x6-b1-t65-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [65, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [65, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [65, 6, 18], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 18, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } }, "decayT": { "dtype": "float32", "shape": [65, 6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [65, 6], "data": { "kind": "fillFloat32", "sinStep": 0.367, "cosStep": 0.657, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [65, 6, 18] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 18, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv32-h2x6-b1-t257-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [1, 257, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [1, 257, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [1, 257, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 32, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 257, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 257, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [1, 257, 6, 32] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk256-dv64-h2x4-b2-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 256], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 256] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-linear-dk64-dv9-h1x3-b1-t65-tile4", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float32", "shape": [65, 1, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [65, 1, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [65, 3, 9], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [65, 3, 9] }, "finalStateT": { "dtype": "float32", "shape": [1, 3, 9, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv20-h2x6-b1-t33-tile4", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float32", "shape": [1, 33, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [1, 33, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [1, 33, 6, 20], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 20, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 33, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 33, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [1, 33, 6, 20] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 20, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-delta-dk32-dv12-h2x4-b1-t129-tile4", "attrs": { "update_rule": "delta", "qk_l2_norm": 1, "beta_activation": "sigmoid" }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float32", "shape": [129, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [129, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [129, 4, 12], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 12, 32], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "betaT": { "dtype": "float32", "shape": [129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [129, 4, 12] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 12, 32] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated-dk8-dv4-h1x2-b1-t65-tile2", "attrs": { "update_rule": "gated", "qk_l2_norm": 1, "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [65, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [65, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [65, 2, 4], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 4, 8], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [65, 2], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [65, 2, 4] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 4, 8] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv32-h2x4-b1-t1-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [1, 1, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [1, 1, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [1, 1, 4, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "decayT": { "dtype": "float32", "shape": [1, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [1, 1, 4, 32] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv32-h2x4-b4-t4-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [16, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [16, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [16, 4, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [4, 4, 32, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [5], "data": { "kind": "values", "values": [0, 4, 8, 12, 16] } }, "decayT": { "dtype": "float32", "shape": [16, 4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [16, 4], "data": { "kind": "fillFloat32", "sinStep": 0.367, "cosStep": 0.657, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [16, 4, 32] }, "finalStateT": { "dtype": "float32", "shape": [4, 4, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-linear-dk64-dv64-h2x4-b1-t65-tile2", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [65, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [65, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [65, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 64, 64], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [65, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 64, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float32-gated_delta-dk128-dv64-h2x4-b2-t129-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 129, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [2, 129, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [2, 129, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [2, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [2, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 129, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk7-dv5-h1x2-b1-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 33, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [1, 33, 1, 7], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [1, 33, 2, 5], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "decayT": { "dtype": "float32", "shape": [1, 33, 2], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 33, 2], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 33, 2, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 5, 7] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-delta-dk20-dv6-h1x3-b1-t65-tile2", "attrs": { "update_rule": "delta", "qk_l2_norm": 1, "beta_activation": "sigmoid" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [65, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [65, 1, 20], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [65, 3, 6], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 3, 6, 20], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } }, "betaT": { "dtype": "float32", "shape": [65, 3], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [65, 3, 6] }, "finalStateT": { "dtype": "float32", "shape": [1, 3, 6, 20] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-linear-dk32-dv8-h2x6-b2-t17-tile2", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [34, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [34, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [34, 6, 8], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 8, 32], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [34, 6, 8] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 8, 32] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated-dk64-dv16-h2x4-b1-t129-tile2", "attrs": { "update_rule": "gated", "qk_l2_norm": 1, "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 129, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [1, 129, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [1, 129, 4, 16], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 16, 64], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 129, 4, 16] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 16, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk96-dv1-h1x1-b1-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [33, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [33, 1, 96], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [33, 1, 1], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 33] } }, "decayT": { "dtype": "float32", "shape": [33, 1], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [1], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [1], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [33, 1], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [33, 1, 1] }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 1, 96] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv18-h2x6-b1-t65-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [65, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [65, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [65, 6, 18], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 18, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } }, "decayT": { "dtype": "float32", "shape": [65, 6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [65, 6], "data": { "kind": "fillFloat32", "sinStep": 0.367, "cosStep": 0.657, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [65, 6, 18] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 18, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv32-h2x6-b1-t257-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 257, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [1, 257, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [1, 257, 6, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 32, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 257, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 257, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 257, 6, 32] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk256-dv64-h2x4-b2-t33-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [66, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [66, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 256], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 256] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-linear-dk64-dv9-h1x3-b1-t65-tile4", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float16", "shape": [65, 1, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [65, 1, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [65, 3, 9], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "cuSeqlensT": { "dtype": "int32", "shape": [2], "data": { "kind": "values", "values": [0, 65] } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [65, 3, 9] }, "finalStateT": { "dtype": "float32", "shape": [1, 3, 9, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv20-h2x6-b1-t33-tile4", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 33, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [1, 33, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [1, 33, 6, 20], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 6, 20, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [1, 33, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 33, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 33, 6, 20] }, "finalStateT": { "dtype": "float32", "shape": [1, 6, 20, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-delta-dk32-dv12-h2x4-b1-t129-tile4", "attrs": { "update_rule": "delta", "qk_l2_norm": 1, "beta_activation": "sigmoid" }, "tunables": { "TILE_V": 4 }, "inputs": { "queryT": { "dtype": "float16", "shape": [129, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [129, 2, 32], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [129, 4, 12], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 12, 32], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "betaT": { "dtype": "float32", "shape": [129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [129, 4, 12] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 12, 32] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated-dk8-dv4-h1x2-b1-t65-tile2", "attrs": { "update_rule": "gated", "qk_l2_norm": 1, "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [65, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [65, 1, 8], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [65, 2, 4], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 2, 4, 8], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [65, 2], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [65, 2, 4] }, "finalStateT": { "dtype": "float32", "shape": [1, 2, 4, 8] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv32-h2x4-b1-t1-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 1, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [1, 1, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [1, 1, 4, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "decayT": { "dtype": "float32", "shape": [1, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [1, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 1, 4, 32] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv32-h2x4-b4-t4-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [16, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [16, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [16, 4, 32], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [4, 4, 32, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [5], "data": { "kind": "values", "values": [0, 4, 8, 12, 16] } }, "decayT": { "dtype": "float32", "shape": [16, 4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [16, 4], "data": { "kind": "fillFloat32", "sinStep": 0.367, "cosStep": 0.657, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [16, 4, 32] }, "finalStateT": { "dtype": "float32", "shape": [4, 4, 32, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-linear-dk64-dv64-h2x4-b1-t65-tile2", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [65, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [65, 2, 64], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [65, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 64, 64], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [65, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 64, 64] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-general-float16-gated_delta-dk128-dv64-h2x4-b2-t129-tile2", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 129, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [2, 129, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [2, 129, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [2, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [2, 129, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 129, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk96-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 96], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 96], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 96], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 96] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk96-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 96], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 96], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 96], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 96] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk128-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk128-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 128], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 128], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 128] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk160-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 160], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 160], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 160], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 160] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk160-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 160], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 160], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 160], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 160] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk192-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 192], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 192], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 192], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 192] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk192-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 192], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 192], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 192], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 192] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk224-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 224], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 224], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 224], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 224] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk224-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 224], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 224], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 224], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 224] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk256-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 256], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 256], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 256] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk320-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 320], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 320], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 320], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 320] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk320-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 320], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 320], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 320], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 320] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk384-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 384], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 384], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 384], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 384] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk384-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 384], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 384], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 384], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 384] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk512-t33", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [66, 2, 512], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [66, 2, 512], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [66, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 512], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [66, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [66, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 512] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "gdn-lane-density-f32-dk512-t129", "attrs": { "update_rule": "gated_delta", "qk_l2_norm": 1, "beta_activation": "sigmoid", "gate_activation": "qwen" }, "tunables": { "TILE_V": 2 }, "inputs": { "queryT": { "dtype": "float32", "shape": [258, 2, 512], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float32", "shape": [258, 2, 512], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float32", "shape": [258, 4, 64], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 64, 512], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "decayT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.497, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [258, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [258, 4, 64] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 64, 512] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.0004, "finalStateT": 0.0004 } }, { "name": "power_of_two_lane_cap", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [3, 1, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.43, "cosStep": 0.61, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [3, 1, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.45, "cosStep": 0.63, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [3, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.47, "cosStep": 0.65, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.49, "cosStep": 0.67, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 1, 4, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [3, 1, 4], "tolerance": 1e-7 }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 4, 6144] } }, "relTolerance": 0.001, "tolerance": 0.00002, "tunables": {} }, { "name": "power_of_two_lane_cap_portable_schedule", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [3, 1, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.43, "cosStep": 0.61, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [3, 1, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.45, "cosStep": 0.63, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [3, 1, 4], "data": { "kind": "fillFloat32", "sinStep": 0.47, "cosStep": 0.65, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.49, "cosStep": 0.67, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [3, 1], "data": { "kind": "fillFloat32", "sinStep": 0.51, "cosStep": 0.69, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [1, 1, 4, 6144], "data": { "kind": "fillFloat32", "sinStep": 0.53, "cosStep": 0.71, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [3, 1, 4], "tolerance": 1e-7 }, "finalStateT": { "dtype": "float32", "shape": [1, 1, 4, 6144] } }, "relTolerance": 0.001, "tolerance": 0.00002, "tunables": { "LANE_WORDS": 6, "V_GROUPS": 1 } }, { "name": "paired_portable_ragged_empty_singleton_odd_qk_and_value_tails", "attrs": { "update_rule": "gated_delta", "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [69, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.127, "cosStep": 0.337, "scale": 0.2 } }, "keyT": { "dtype": "float16", "shape": [69, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.377, "scale": 0.2 } }, "valueT": { "dtype": "float16", "shape": [69, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.41700000000000004, "scale": 0.2 } }, "initialStateT": { "dtype": "float32", "shape": [4, 4, 5, 20], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.457, "scale": 0.05 } }, "cuSeqlensT": { "dtype": "int32", "shape": [5], "data": { "kind": "values", "values": [0, 1, 34, 34, 69] } }, "decayT": { "dtype": "float32", "shape": [69, 4], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.577, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.33699999999999997, "cosStep": 0.617, "scale": 0.1 } }, "betaT": { "dtype": "float32", "shape": [69, 4], "data": { "kind": "fillFloat32", "sinStep": 0.367, "cosStep": 0.657, "scale": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [69, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [4, 4, 5, 20] } }, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_delta_zero_ragged_qk_and_value_tails", "attrs": { "update_rule": "delta", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.215, "cosStep": 0.395, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.235, "cosStep": 0.415, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.255, "cosStep": 0.435, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 8, 17] } }, "betaT": { "dtype": "float32", "shape": [17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.295, "cosStep": 0.475, "scale": 0.6 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_delta_zero_packed_qk_and_value_tails", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.222, "cosStep": 0.402, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.242, "cosStep": 0.422, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [1, 17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.262, "cosStep": 0.442, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [1, 17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.302, "cosStep": 0.482, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_gated_delta_state_ragged_qk_and_value_tails", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 8, 17] } }, "decayT": { "dtype": "float32", "shape": [17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 5, 20], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_gated_delta_state_packed_qk_and_value_tails", "attrs": { "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.236, "cosStep": 0.416, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.256, "cosStep": 0.436, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [1, 17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.276, "cosStep": 0.456, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [1, 17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.296, "cosStep": 0.476, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [1, 17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.316, "cosStep": 0.496, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [1, 4, 5, 20], "data": { "kind": "fillFloat32", "sinStep": 0.336, "cosStep": 0.516, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_gated_delta_zero_ragged_qk_and_value_tails", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.243, "cosStep": 0.423, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.263, "cosStep": 0.443, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.283, "cosStep": 0.463, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 8, 17] } }, "decayT": { "dtype": "float32", "shape": [17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.303, "cosStep": 0.483, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.323, "cosStep": 0.503, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "paired_portable_gated_delta_zero_packed_qk_and_value_tails", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.25, "cosStep": 0.43, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [1, 17, 2, 20], "data": { "kind": "fillFloat32", "sinStep": 0.27, "cosStep": 0.45, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [1, 17, 4, 5], "data": { "kind": "fillFloat32", "sinStep": 0.29, "cosStep": 0.47, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [1, 17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.31, "cosStep": 0.49, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [1, 17, 4], "data": { "kind": "fillFloat32", "sinStep": 0.33, "cosStep": 0.51, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [1, 17, 4, 5] }, "finalStateT": { "dtype": "float32", "shape": [1, 4, 5, 20] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tolerances": { "outputT": 0.00002, "finalStateT": 0.00002 }, "relTolerances": { "outputT": 0.002, "finalStateT": 0.0004 } }, { "name": "odd-density-float32-k65-linear_state_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.117, "cosStep": 0.297, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.137, "cosStep": 0.317, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.337, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.397, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k129-linear_state_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.124, "cosStep": 0.304, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.144, "cosStep": 0.324, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.164, "cosStep": 0.344, "scale": 0.32 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.224, "cosStep": 0.404, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k257-linear_zero_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.131, "cosStep": 0.311, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.151, "cosStep": 0.331, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.171, "cosStep": 0.351, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k65-linear_zero_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.138, "cosStep": 0.318, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.158, "cosStep": 0.338, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.178, "cosStep": 0.358, "scale": 0.32 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k129-gated_state_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.145, "cosStep": 0.325, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.165, "cosStep": 0.345, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.185, "cosStep": 0.365, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.205, "cosStep": 0.385, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.245, "cosStep": 0.425, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k257-gated_state_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.152, "cosStep": 0.332, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.172, "cosStep": 0.352, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.192, "cosStep": 0.372, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.212, "cosStep": 0.392, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.252, "cosStep": 0.432, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k65-gated_zero_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.159, "cosStep": 0.339, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.179, "cosStep": 0.359, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.199, "cosStep": 0.379, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.219, "cosStep": 0.399, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k129-gated_zero_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.166, "cosStep": 0.346, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.186, "cosStep": 0.366, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.206, "cosStep": 0.386, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.226, "cosStep": 0.406, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k257-gated_qwen_state_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.173, "cosStep": 0.353, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.193, "cosStep": 0.373, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.213, "cosStep": 0.393, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.233, "cosStep": 0.413, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.273, "cosStep": 0.453, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.293, "cosStep": 0.473, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.313, "cosStep": 0.493, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k65-gated_qwen_state_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.18, "cosStep": 0.36, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.2, "cosStep": 0.38, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.22, "cosStep": 0.4, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.24, "cosStep": 0.42, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.28, "cosStep": 0.46, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.3, "cosStep": 0.48, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.32, "cosStep": 0.5, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k129-gated_qwen_zero_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.367, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.207, "cosStep": 0.387, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.227, "cosStep": 0.407, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.427, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.487, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.327, "cosStep": 0.507, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k257-gated_qwen_zero_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.194, "cosStep": 0.374, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.214, "cosStep": 0.394, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.234, "cosStep": 0.414, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.254, "cosStep": 0.434, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.314, "cosStep": 0.494, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.334, "cosStep": 0.514, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k65-delta_state_ragged", "attrs": { "update_rule": "delta", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.201, "cosStep": 0.381, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.221, "cosStep": 0.401, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.241, "cosStep": 0.421, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.281, "cosStep": 0.461, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.301, "cosStep": 0.481, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k129-delta_state_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.208, "cosStep": 0.388, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.228, "cosStep": 0.408, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.248, "cosStep": 0.428, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.288, "cosStep": 0.468, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.308, "cosStep": 0.488, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k257-delta_zero_ragged", "attrs": { "update_rule": "delta", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.215, "cosStep": 0.395, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.235, "cosStep": 0.415, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.255, "cosStep": 0.435, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.295, "cosStep": 0.475, "scale": 0.6 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k65-delta_zero_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.222, "cosStep": 0.402, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.242, "cosStep": 0.422, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.262, "cosStep": 0.442, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.302, "cosStep": 0.482, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k129-gated_delta_state_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k257-gated_delta_state_packed", "attrs": { "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.236, "cosStep": 0.416, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.256, "cosStep": 0.436, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.276, "cosStep": 0.456, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.296, "cosStep": 0.476, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.316, "cosStep": 0.496, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.336, "cosStep": 0.516, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k65-gated_delta_zero_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.243, "cosStep": 0.423, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.263, "cosStep": 0.443, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.283, "cosStep": 0.463, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.303, "cosStep": 0.483, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.323, "cosStep": 0.503, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k129-gated_delta_zero_packed", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.25, "cosStep": 0.43, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.27, "cosStep": 0.45, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.29, "cosStep": 0.47, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.31, "cosStep": 0.49, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33, "cosStep": 0.51, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k257-gated_delta_qwen_state_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.257, "cosStep": 0.437, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.457, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.297, "cosStep": 0.477, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.317, "cosStep": 0.497, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.337, "cosStep": 0.517, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.357, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.377, "cosStep": 0.557, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.397, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float32-k65-gated_delta_qwen_state_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.264, "cosStep": 0.444, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.284, "cosStep": 0.464, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.304, "cosStep": 0.484, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.324, "cosStep": 0.504, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.344, "cosStep": 0.524, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.364, "cosStep": 0.544, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.384, "cosStep": 0.564, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.404, "cosStep": 0.584, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-k129-gated_delta_qwen_zero_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.271, "cosStep": 0.451, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.291, "cosStep": 0.471, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.311, "cosStep": 0.491, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.331, "cosStep": 0.511, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.351, "cosStep": 0.531, "scale": 0.2, "offset": 0.5 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.391, "cosStep": 0.571, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.411, "cosStep": 0.591, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float32-k257-gated_delta_qwen_zero_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.278, "cosStep": 0.458, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.298, "cosStep": 0.478, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.318, "cosStep": 0.498, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.338, "cosStep": 0.518, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.358, "cosStep": 0.538, "scale": 0.6 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.398, "cosStep": 0.578, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.418, "cosStep": 0.598, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k65-linear_state_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.117, "cosStep": 0.297, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.137, "cosStep": 0.317, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.157, "cosStep": 0.337, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.217, "cosStep": 0.397, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k129-linear_state_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.124, "cosStep": 0.304, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.144, "cosStep": 0.324, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.164, "cosStep": 0.344, "scale": 0.32 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.224, "cosStep": 0.404, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k257-linear_zero_ragged", "attrs": { "update_rule": "linear" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.131, "cosStep": 0.311, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.151, "cosStep": 0.331, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.171, "cosStep": 0.351, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k65-linear_zero_packed", "attrs": { "update_rule": "linear", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.138, "cosStep": 0.318, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.158, "cosStep": 0.338, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.178, "cosStep": 0.358, "scale": 0.32 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k129-gated_state_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.145, "cosStep": 0.325, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.165, "cosStep": 0.345, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.185, "cosStep": 0.365, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.205, "cosStep": 0.385, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.245, "cosStep": 0.425, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k257-gated_state_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.152, "cosStep": 0.332, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.172, "cosStep": 0.352, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.192, "cosStep": 0.372, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.212, "cosStep": 0.392, "scale": 0.25, "offset": -0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.252, "cosStep": 0.432, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k65-gated_zero_ragged", "attrs": { "update_rule": "gated" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.159, "cosStep": 0.339, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.179, "cosStep": 0.359, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.199, "cosStep": 0.379, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.219, "cosStep": 0.399, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k129-gated_zero_packed", "attrs": { "update_rule": "gated", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.166, "cosStep": 0.346, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.186, "cosStep": 0.366, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.206, "cosStep": 0.386, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.226, "cosStep": 0.406, "scale": 0.25, "offset": -0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k257-gated_qwen_state_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.173, "cosStep": 0.353, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.193, "cosStep": 0.373, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.213, "cosStep": 0.393, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.233, "cosStep": 0.413, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.273, "cosStep": 0.453, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.293, "cosStep": 0.473, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.313, "cosStep": 0.493, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k65-gated_qwen_state_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.18, "cosStep": 0.36, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.2, "cosStep": 0.38, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.22, "cosStep": 0.4, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.24, "cosStep": 0.42, "scale": 0.3 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.28, "cosStep": 0.46, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.3, "cosStep": 0.48, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.32, "cosStep": 0.5, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k129-gated_qwen_zero_ragged", "attrs": { "update_rule": "gated", "gate_activation": "qwen" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.187, "cosStep": 0.367, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.207, "cosStep": 0.387, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.227, "cosStep": 0.407, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.247, "cosStep": 0.427, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.307, "cosStep": 0.487, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.327, "cosStep": 0.507, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k257-gated_qwen_zero_packed", "attrs": { "update_rule": "gated", "gate_activation": "qwen", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.194, "cosStep": 0.374, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.214, "cosStep": 0.394, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.234, "cosStep": 0.414, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.254, "cosStep": 0.434, "scale": 0.3 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.314, "cosStep": 0.494, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.334, "cosStep": 0.514, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k65-delta_state_ragged", "attrs": { "update_rule": "delta", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.201, "cosStep": 0.381, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.221, "cosStep": 0.401, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.241, "cosStep": 0.421, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.281, "cosStep": 0.461, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.301, "cosStep": 0.481, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k129-delta_state_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.208, "cosStep": 0.388, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.228, "cosStep": 0.408, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.248, "cosStep": 0.428, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.288, "cosStep": 0.468, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.308, "cosStep": 0.488, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k257-delta_zero_ragged", "attrs": { "update_rule": "delta", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.215, "cosStep": 0.395, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.235, "cosStep": 0.415, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.255, "cosStep": 0.435, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.295, "cosStep": 0.475, "scale": 0.6 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k65-delta_zero_packed", "attrs": { "update_rule": "delta", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.222, "cosStep": 0.402, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.242, "cosStep": 0.422, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.262, "cosStep": 0.442, "scale": 0.32 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.302, "cosStep": 0.482, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k129-gated_delta_state_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k257-gated_delta_state_packed", "attrs": { "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.236, "cosStep": 0.416, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.256, "cosStep": 0.436, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.276, "cosStep": 0.456, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.296, "cosStep": 0.476, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.316, "cosStep": 0.496, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.336, "cosStep": 0.516, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k65-gated_delta_zero_ragged", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.243, "cosStep": 0.423, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.263, "cosStep": 0.443, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.283, "cosStep": 0.463, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.303, "cosStep": 0.483, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.323, "cosStep": 0.503, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k129-gated_delta_zero_packed", "attrs": { "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.25, "cosStep": 0.43, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.27, "cosStep": 0.45, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.29, "cosStep": 0.47, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.31, "cosStep": 0.49, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.33, "cosStep": 0.51, "scale": 0.2, "offset": 0.5 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float16-k257-gated_delta_qwen_state_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.257, "cosStep": 0.437, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.277, "cosStep": 0.457, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.297, "cosStep": 0.477, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.317, "cosStep": 0.497, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.337, "cosStep": 0.517, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.357, "cosStep": 0.537, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.377, "cosStep": 0.557, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.397, "cosStep": 0.577, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 1 } }, { "name": "odd-density-float16-k65-gated_delta_qwen_state_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "none", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.264, "cosStep": 0.444, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [8, 2, 65], "data": { "kind": "fillFloat32", "sinStep": 0.284, "cosStep": 0.464, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [8, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.304, "cosStep": 0.484, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.324, "cosStep": 0.504, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [8, 6], "data": { "kind": "fillFloat32", "sinStep": 0.344, "cosStep": 0.524, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 65], "data": { "kind": "fillFloat32", "sinStep": 0.364, "cosStep": 0.544, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.384, "cosStep": 0.564, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.404, "cosStep": 0.584, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [8, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 65] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-k129-gated_delta_qwen_zero_ragged", "attrs": { "gate_activation": "qwen", "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.271, "cosStep": 0.451, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.291, "cosStep": 0.471, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.311, "cosStep": 0.491, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 3, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.331, "cosStep": 0.511, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.351, "cosStep": 0.531, "scale": 0.2, "offset": 0.5 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.391, "cosStep": 0.571, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.411, "cosStep": 0.591, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 3 } }, { "name": "odd-density-float16-k257-gated_delta_qwen_zero_packed", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.278, "cosStep": 0.458, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.298, "cosStep": 0.478, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.318, "cosStep": 0.498, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.338, "cosStep": 0.518, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.358, "cosStep": 0.538, "scale": 0.6 } }, "aLogT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.398, "cosStep": 0.578, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [6], "data": { "kind": "fillFloat32", "sinStep": 0.418, "cosStep": 0.598, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 4 } }, { "name": "odd-density-float32-k129-rank4_gated_delta_qwen_state", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.63, "cosStep": 0.81, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.65, "cosStep": 0.83, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [2, 4, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [2, 4, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float32-k257-empty_request_ragged_gated_delta", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Ragged_CompactReplayMatchesSequentialPrefixes", "note": "Zero-length leading request, as the upstream ragged capture case uses. A request with no tokens copies its incoming state through." }, "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [5, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [5, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [5, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 0, 2, 5] } }, "decayT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [3, 4, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [5, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [3, 4, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float32-k129-uneven_ragged_gated_delta_qwen", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Chunked_RaggedUnequalLengths", "note": "Ragged geometry mirrored at a fixture-sized head count; upstream has no literal expectations." }, "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float32", "shape": [20, 1, 129], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [20, 1, 129], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [20, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 4, 15, 20] } }, "decayT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.79, "cosStep": 0.97, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [3, 2, 8, 129], "data": { "kind": "fillFloat32", "sinStep": 0.81, "cosStep": 0.99, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [20, 2, 8] }, "finalStateT": { "dtype": "float32", "shape": [3, 2, 8, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float16-k129-rank4_gated_delta_qwen_state", "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.63, "cosStep": 0.81, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [2, 4, 2, 129], "data": { "kind": "fillFloat32", "sinStep": 0.65, "cosStep": 0.83, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [2, 4, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.32 } }, "decayT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [2, 4, 4], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [2, 4, 6, 129], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [4], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [2, 4, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 4, 6, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float16-k257-empty_request_ragged_gated_delta", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Ragged_CompactReplayMatchesSequentialPrefixes", "note": "Zero-length leading request, as the upstream ragged capture case uses. A request with no tokens copies its incoming state through." }, "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [5, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.67, "cosStep": 0.85, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [5, 2, 257], "data": { "kind": "fillFloat32", "sinStep": 0.69, "cosStep": 0.87, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [5, 4, 6], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 0, 2, 5] } }, "decayT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [5, 4], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [3, 4, 6, 257], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [5, 4, 6] }, "finalStateT": { "dtype": "float32", "shape": [3, 4, 6, 257] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float16-k129-uneven_ragged_gated_delta_qwen", "provenance": { "source": "onnxruntime/test/contrib_ops/gated_delta_net_op_test.cc", "test": "GatedDeltaNetTest.Chunked_RaggedUnequalLengths", "note": "Ragged geometry mirrored at a fixture-sized head count; upstream has no literal expectations." }, "attrs": { "gate_activation": "qwen", "beta_activation": "sigmoid", "qk_l2_norm": 1 }, "inputs": { "queryT": { "dtype": "float16", "shape": [20, 1, 129], "data": { "kind": "fillFloat32", "sinStep": 0.71, "cosStep": 0.89, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [20, 1, 129], "data": { "kind": "fillFloat32", "sinStep": 0.73, "cosStep": 0.91, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [20, 2, 8], "data": { "kind": "fillFloat32", "sinStep": 0.75, "cosStep": 0.93, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [4], "data": { "kind": "values", "values": [0, 4, 15, 20] } }, "decayT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.77, "cosStep": 0.95, "scale": 0.3 } }, "betaT": { "dtype": "float32", "shape": [20, 2], "data": { "kind": "fillFloat32", "sinStep": 0.79, "cosStep": 0.97, "scale": 0.6 } }, "initialStateT": { "dtype": "float32", "shape": [3, 2, 8, 129], "data": { "kind": "fillFloat32", "sinStep": 0.81, "cosStep": 0.99, "scale": 0.1 } }, "aLogT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.83, "cosStep": 1.01, "scale": 0.1 } }, "dtBiasT": { "dtype": "float32", "shape": [2], "data": { "kind": "fillFloat32", "sinStep": 0.85, "cosStep": 1.03, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [20, 2, 8] }, "finalStateT": { "dtype": "float32", "shape": [3, 2, 8, 129] } }, "relTolerance": 0.0004, "tolerance": 0.00002 }, { "name": "odd-density-float32-register-boundary-k511", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 511], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 511], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 511], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 511] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float32-register-boundary-k513", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float32", "shape": [7, 2, 513], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float32", "shape": [7, 2, 513], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float32", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 513], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float32", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 513] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-register-boundary-k511", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 511], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 511], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 511], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 511] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } }, { "name": "odd-density-float16-register-boundary-k513", "attrs": { "beta_activation": "none" }, "inputs": { "queryT": { "dtype": "float16", "shape": [7, 2, 513], "data": { "kind": "fillFloat32", "sinStep": 0.229, "cosStep": 0.409, "scale": 0.3 } }, "keyT": { "dtype": "float16", "shape": [7, 2, 513], "data": { "kind": "fillFloat32", "sinStep": 0.249, "cosStep": 0.429, "scale": 0.28 } }, "valueT": { "dtype": "float16", "shape": [7, 6, 6], "data": { "kind": "fillFloat32", "sinStep": 0.269, "cosStep": 0.449, "scale": 0.32 } }, "cuSeqlensT": { "dtype": "int32", "shape": [3], "data": { "kind": "values", "values": [0, 4, 7] } }, "decayT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.289, "cosStep": 0.469, "scale": 0.25, "offset": -0.5 } }, "betaT": { "dtype": "float32", "shape": [7, 6], "data": { "kind": "fillFloat32", "sinStep": 0.309, "cosStep": 0.489, "scale": 0.2, "offset": 0.5 } }, "initialStateT": { "dtype": "float32", "shape": [2, 6, 6, 513], "data": { "kind": "fillFloat32", "sinStep": 0.329, "cosStep": 0.509, "scale": 0.1 } } }, "outputs": { "outputT": { "dtype": "float16", "shape": [7, 6, 6] }, "finalStateT": { "dtype": "float32", "shape": [2, 6, 6, 513] } }, "relTolerance": 0.0004, "tolerance": 0.00002, "tunables": { "TILE_V": 2 } } ] }