Xenova's picture
Xenova HF Staff
sync 6fdf6301e2bb
6111bbd verified
Raw History Blame
350 kB
{
"tunableSpace": {
"WORKGROUP_SIZE": [64, 128, 256],
"CLUSTER_SPLITKV_TILE_Q": [8, 16, 32],
"CLUSTER_SPLITKV_LPQ": [4, 8],
"MATERIALIZED_SGMAT_APPLY_SUBGROUP_TILE_COLS": [32, 64],
"MATERIALIZED_SGMAT_APPLY_SUBGROUP_TILE_COLS_D128": [32, 64]
},
"cases": [
{
"name": "qwen-image21-image-rank3-b1-q4096-k4110-h32-d128",
"preset": "model",
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4096, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4110, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4110, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4096, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "qwen-image21-image-rank3-b1-q4096-k4110-h32-d128-headchunk4",
"preset": "model",
"tunables": { "MATERIALIZED_SGMAT_HEAD_CHUNK": 4 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4096, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4110, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4110, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4096, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "qwen-image21-512-image-rank3-b1-q1024-k1055-h32-d128",
"preset": "model",
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "qwen-image21-512-image-rank3-b1-q1024-k1055-h32-d128-headchunk4",
"preset": "model",
"tunables": { "MATERIALIZED_SGMAT_HEAD_CHUNK": 4 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "qwen-image21-512-image-rank3-b1-q1024-k1055-h32-d128-headchunk8",
"preset": "model",
"tunables": { "MATERIALIZED_SGMAT_HEAD_CHUNK": 8 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "qwen-image21-512-image-rank3-b1-q1024-k1055-h32-d128-headchunk16",
"preset": "model",
"tunables": { "MATERIALIZED_SGMAT_HEAD_CHUNK": 16 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 20261, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20262, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1055, 4096], "dist": "normal", "seed": 20263, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096] } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "4 * dim(shapes.q, 1) * dim(shapes.k, 1) * attrs.q_num_heads * (dim(shapes.q, 2) / attrs.q_num_heads)"
}
]
}
},
{
"name": "rank4_small",
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 32, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float32",
"shape": [1, 4, 32, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float32",
"shape": [1, 4, 32, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 32, 16] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "masked-decode-gqa-32h8kv-d128-kv4096-zero-additive",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 1, "kvSeq": 4096, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 1, 128], "dist": "normal", "seed": 9031, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9032, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9033, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [1, 4096], "data": { "kind": "constant", "value": 0.0 } }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 1, 128] } },
"bench": {
"metrics": [
{
"type": "bandwidth",
"value": "(2 * args.qHeads * args.qSeq * args.headDim + 2 * args.kvHeads * args.kvSeq * args.headDim + args.qSeq * args.kvSeq) * dtypeBytes(args.dtype)"
}
]
}
},
{
"name": "masked-decode-gqa-32h8kv-d128-kv1024-zero-additive",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 1, "kvSeq": 1024, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 1, 128], "dist": "normal", "seed": 9051, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 1024, 128], "dist": "normal", "seed": 9052, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 1024, 128], "dist": "normal", "seed": 9053, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [1, 1024], "data": { "kind": "constant", "value": 0.0 } }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 1, 128] } },
"bench": {
"metrics": [
{
"type": "bandwidth",
"value": "(2 * args.qHeads * args.qSeq * args.headDim + 2 * args.kvHeads * args.kvSeq * args.headDim + args.qSeq * args.kvSeq) * dtypeBytes(args.dtype)"
}
]
}
},
{
"name": "masked-decode-gqa-32h8kv-d128-kv4096-f16-zero-additive",
"preset": "model",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 1, "kvSeq": 4096, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 1, 128], "dist": "normal", "seed": 9061, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9062, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9063, "scale": 0.2 },
"attn_mask": { "dtype": "float16", "shape": [1, 4096], "data": { "kind": "constant", "value": 0.0 } }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 1, 128] } },
"bench": {
"metrics": [
{
"type": "bandwidth",
"value": "(2 * args.qHeads * args.qSeq * args.headDim + 2 * args.kvHeads * args.kvSeq * args.headDim + args.qSeq * args.kvSeq) * dtypeBytes(args.dtype)"
}
]
}
},
{
"name": "masked-shortq8-gqa-32h8kv-d128-kv4096-zero-additive",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 8, "kvSeq": 4096, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 8, 128], "dist": "normal", "seed": 9041, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9042, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9043, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [8, 4096], "data": { "kind": "constant", "value": 0.0 } }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 8, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "masked-shortq16-gqa-32h8kv-d128-kv4096-zero-additive",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 16, "kvSeq": 4096, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 16, 128], "dist": "normal", "seed": 9071, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9072, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 4096, 128], "dist": "normal", "seed": 9073, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [16, 4096], "data": { "kind": "constant", "value": 0.0 } }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 16, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-additive-f32-12h-d64-s512",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 1354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 1355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 1356, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [512, 512], "dist": "normal", "seed": 1357, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 512, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-bool-rank4-f32-12h-d64-s512",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 2354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 2355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 2356, "scale": 0.2 },
"attn_mask": {
"dtype": "bool",
"shape": [1, 1, 512, 512],
"data": { "kind": "cycle", "values": [1, 1, 1, 1, 0] }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 512, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-broadcast-f32-4h-d64-s32",
"preset": "smoke",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 32, "kvSeq": 32, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 356, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [32, 32], "dist": "normal", "seed": 357, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 32, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-singleton-head-broadcast-f32-4h-d64-s32",
"preset": "smoke",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 32, "kvSeq": 32, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 358, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 359, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 32, 64], "dist": "normal", "seed": 360, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [1, 1, 32, 32], "dist": "normal", "seed": 361, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 32, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "llama-decode-gqa-32h8kv-d128-kv2048",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 1, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 1, 128], "dist": "normal", "seed": 330, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 331, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 332, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 1, 128] } },
"bench": {
"metrics": [
{
"type": "bandwidth",
"value": "(2 * args.qHeads * args.qSeq * args.headDim + 2 * args.kvHeads * args.kvSeq * args.headDim) * dtypeBytes(args.dtype)"
}
]
}
},
{
"name": "llama-prefill-causal-32h8kv-d128-s512",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 1, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 512, 128], "dist": "normal", "seed": 333, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 334, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 335, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 512, 128] } },
"bench": {
"metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.headDim * args.qSeq * (args.qSeq + 1) / 2" }]
}
},
{
"name": "llama-prefill-causal-32h8kv-d128-s512-f16",
"preset": "model",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 1, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 128], "dist": "normal", "seed": 336, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 337, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 338, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 128] } },
"bench": {
"metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.headDim * args.qSeq * (args.qSeq + 1) / 2" }]
}
},
{
"name": "flash-prefill-noncausal-32h8kv-d128-s512",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 512, 128], "dist": "normal", "seed": 339, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 340, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 341, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 512, 128] } },
"bench": {
"primary": true,
"metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }]
}
},
{
"name": "flash-prefill-noncausal-32h8kv-d128-s512-f16",
"preset": "model",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 128], "dist": "normal", "seed": 342, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 343, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 344, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-32h8kv-d128-s1024",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 1024, "kvSeq": 1024, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 1024, 128], "dist": "normal", "seed": 345, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 1024, 128], "dist": "normal", "seed": 346, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 1024, 128], "dist": "normal", "seed": 347, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 1024, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-16h16kv-d64-s1024",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 64 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 16, 1024, 64], "dist": "normal", "seed": 348, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 16, 1024, 64], "dist": "normal", "seed": 349, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 16, 1024, 64], "dist": "normal", "seed": 350, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 16, 1024, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-32h8kv-d128-s2048",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 8, "qSeq": 2048, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831845 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 2048, 128], "dist": "normal", "seed": 351, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 352, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 353, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 2048, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-causal-nonsquare-q512-kv2048-8h-d128",
"preset": "model",
"vars": { "dtype": "float32", "heads": 8, "qSeq": 512, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "shape": [1, 8, 512, 128], "dtype": "float32", "dist": "normal", "seed": 401, "scale": 0.3 },
"k": { "shape": [1, 8, 2048, 128], "dtype": "float32", "dist": "normal", "seed": 402, "scale": 0.3 },
"v": { "shape": [1, 8, 2048, 128], "dtype": "float32", "dist": "normal", "seed": 403, "scale": 0.3 }
},
"outputs": { "y": { "shape": [1, 8, 512, 128], "dtype": "float32" } },
"bench": {
"metrics": [{ "type": "gflops", "value": "4 * args.heads * args.headDim * args.qSeq * (args.qSeq + 1) / 2" }]
}
},
{
"name": "flash-prefill-noncausal-12h-d80-s512",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 80 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 348, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 349, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 350, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 512, 80] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "llama-decode-gqa-32h8kv-d128-kv2048-f16",
"preset": "smoke",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 1, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831843 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 1, 128], "dist": "normal", "seed": 4330, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 4331, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 4332, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 1, 128] } },
"bench": {
"metrics": [
{
"type": "bandwidth",
"value": "(2 * args.qHeads * args.qSeq * args.headDim + 2 * args.kvHeads * args.kvSeq * args.headDim) * dtypeBytes(args.dtype)"
}
]
}
},
{
"name": "mask-causal-naivecliff-f32-12h-d64-s512",
"preset": "smoke",
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 5354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 5355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 5356, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [1, 1, 512, 512], "dist": "normal", "seed": 5357, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 512, 64] } },
"bench": {
"metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.headDim * args.qSeq * (args.qSeq + 1) / 2" }]
}
},
{
"name": "shortprompt-naivefloor-f32-4h-d64-s16",
"preset": "smoke",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 16, "kvSeq": 16, "headDim": 64 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 16, 64], "dist": "normal", "seed": 6354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 16, 64], "dist": "normal", "seed": 6355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 16, 64], "dist": "normal", "seed": 6356, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 16, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim72-nonpow2-f32-16h-s1024",
"preset": "smoke",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Non-power-of-two D72 attention exercises the portable LPQ2 flash family; dimensions divisible by 16 retain the materialized GEMM path."
},
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 72 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7354, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7355, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7356, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 16, 1024, 72] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim72-nonpow2-f32-1h-s512",
"preset": "edge",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Measures Attention over a single head with a non-power-of-two head width of 72 and a 512-token sequence."
},
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 512, "kvSeq": 512, "headDim": 72 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1, 512, 72], "dist": "normal", "seed": 7364, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1, 512, 72], "dist": "normal", "seed": 7365, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1, 512, 72], "dist": "normal", "seed": 7366, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 512, 72] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim72-nonpow2-f32-4h-s512",
"preset": "edge",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Four heads at 512 queries (64 query groups) at head dimension 72: midway between the one-head and sixteen-head cases."
},
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 72 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 512, 72], "dist": "normal", "seed": 7374, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 512, 72], "dist": "normal", "seed": 7375, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 512, 72], "dist": "normal", "seed": 7376, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 512, 72] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim80-f32-2h-s512",
"preset": "edge",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Two heads, 512 queries, and head width 80 provide a compact attention case with a non-power-of-two head dimension."
},
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 2, "qSeq": 512, "kvSeq": 512, "headDim": 80 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 7384, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 7385, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 7386, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 2, 512, 80] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim72-nonpow2-f16-16h-s1024",
"preset": "edge",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Half-precision twin of the sixteen-head head-dimension-72 case: the same LPQ2 flash class with half storage, where the materialized GEMM keeps its float32 tiles."
},
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 72 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7394, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7395, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 16, 1024, 72], "dist": "normal", "seed": 7396, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 1024, 72] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-p20-headdim80-f16-12h-s512",
"preset": "edge",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Half-precision twin of the twelve-head head-dimension-80 case: the LPQ4 flash class with half storage."
},
"vars": { "dtype": "float16", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 80 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 7404, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 7405, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 12, 512, 80], "dist": "normal", "seed": 7406, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 12, 512, 80] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "f16-prefill-q32broadcast-8h-d96-s512",
"preset": "smoke",
"vars": { "dtype": "float16", "qHeads": 8, "kvHeads": 8, "qSeq": 512, "kvSeq": 512, "headDim": 96 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8354, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8355, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8356, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 96] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "vhead512-qk64-naive-scalar-doublerecompute-f32-8h-s256",
"preset": "stress",
"vars": {
"dtype": "float32",
"qHeads": 8,
"kvHeads": 8,
"qSeq": 256,
"kvSeq": 256,
"headDim": 64,
"vHeadSize": 512
},
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 8, 256, 64], "dist": "normal", "seed": 8801, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 256, 64], "dist": "normal", "seed": 8802, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 256, 512], "dist": "normal", "seed": 8803, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 256, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "2 * args.qHeads * args.qSeq * args.kvSeq * (args.headDim + args.vHeadSize)" }
]
}
},
{
"name": "vhead320-qk80-f32-12h-s384-cross-attention-projection",
"preset": "stress",
"vars": {
"dtype": "float32",
"qHeads": 12,
"kvHeads": 12,
"qSeq": 384,
"kvSeq": 384,
"headDim": 80,
"vHeadSize": 320
},
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 384, 80], "dist": "normal", "seed": 8804, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 384, 80], "dist": "normal", "seed": 8805, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 384, 320], "dist": "normal", "seed": 8806, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 384, 320], "dist": "empty" } },
"bench": {
"primary": true,
"metrics": [
{ "type": "gflops", "value": "2 * args.qHeads * args.qSeq * args.kvSeq * (args.headDim + args.vHeadSize)" }
]
}
},
{
"name": "rank3-prefill-mha-16h-d64-s1024-barrier-per-key",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "masked-causal-vhead512-forced-naive-scalar-f32-8h-s192",
"preset": "stress",
"vars": {
"dtype": "float32",
"qHeads": 8,
"kvHeads": 8,
"qSeq": 192,
"kvSeq": 192,
"headDim": 64,
"vHeadSize": 512
},
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 8, 192, 64], "dist": "normal", "seed": 8821, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 192, 64], "dist": "normal", "seed": 8822, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 192, 512], "dist": "normal", "seed": 8823, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [1, 1, 192, 192], "dist": "normal", "seed": 8824, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 192, 512], "dist": "empty" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.qHeads * (args.headDim + args.vHeadSize) * args.qSeq * (args.qSeq + 1) / 2"
}
]
}
},
{
"name": "f16-prefill-q31-boundary-gqa-32h8kv-d128-kv2048",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 31, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831843 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 31, 128], "dist": "normal", "seed": 8831, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8832, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8833, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 31, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "f16-prefill-q32-control-gqa-32h8kv-d128-kv2048",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 32, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831843 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 32, 128], "dist": "normal", "seed": 8831, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8832, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8833, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 32, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] },
"provenance": {
"notes": "A query length of 32 fills one complete tile at the same KV length, head counts, and dtype as the q31 boundary shape."
}
},
{
"name": "f16-prefill-q128-control-gqa-32h8kv-d128-kv2048",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 128, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831843 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 128, 128], "dist": "normal", "seed": 8831, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8832, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8833, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 128, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] },
"provenance": {
"notes": "Measures Attention over a grouped-query prefill: 32 query heads over 8 key/value heads, 128 queries against a 2048-token cache, head width 128."
}
},
{
"name": "f16-prefill-q64-control-gqa-32h8kv-d128-kv2048",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 32, "kvHeads": 8, "qSeq": 64, "kvSeq": 2048, "headDim": 128 },
"attrs": { "is_causal": 0, "scale": 0.08838834764831843 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 64, 128], "dist": "normal", "seed": 8831, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8832, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 2048, 128], "dist": "normal", "seed": 8833, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 64, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] },
"provenance": {
"notes": "Measures Attention over a grouped-query prefill: 32 query heads over 8 key/value heads, 64 queries against a 2048-token cache, head width 128."
}
},
{
"name": "prefill-f32-16h-d60-below-flash-floor-s512",
"preset": "stress",
"provenance": {
"source": "hand-authored head-dimension boundary",
"notes": "Realistic hidden-size-960 attention (16 heads x 60) exercises the path below the rank-4 flash headDim>=64 threshold."
},
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 512, "kvSeq": 512, "headDim": 60 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 16, 512, 60], "dist": "normal", "seed": 9171, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 16, 512, 60], "dist": "normal", "seed": 9172, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 16, 512, 60], "dist": "normal", "seed": 9173, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 16, 512, 60], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h4-d64-s128",
"preset": "model",
"provenance": {
"notes": "Noncausal float32 attention with four heads, head dimension 64 and 128-token sequences."
},
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 128, "kvSeq": 128, "headDim": 64 },
"attrs": { "is_causal": 0, "scale": 0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 128, 64], "dist": "normal", "seed": 9175, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 128, 64], "dist": "normal", "seed": 9176, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 128, 64], "dist": "normal", "seed": 9177, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 128, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h4-d256-s128",
"preset": "model",
"provenance": {
"notes": "Noncausal float32 attention with four heads, head dimension 256 and 128-token sequences."
},
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 128, "kvSeq": 128, "headDim": 256 },
"attrs": { "is_causal": 0, "scale": 0.0625 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 128, 256], "dist": "normal", "seed": 9181, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 128, 256], "dist": "normal", "seed": 9182, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 128, 256], "dist": "normal", "seed": 9183, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 128, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "cluster-splitkv-q128-kv8192-h8-d64-f32",
"attrs": { "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 8, 128, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float32",
"shape": [1, 8, 8192, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float32",
"shape": [1, 8, 8192, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 128, 64] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "cluster-splitkv-q256-kv16384-h4-d128-f32",
"attrs": { "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 256, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float32",
"shape": [1, 4, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float32",
"shape": [1, 4, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 256, 128] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "cluster-splitkv-q96-kv32768-h2-d128-f16",
"attrs": { "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 2, 96, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float16",
"shape": [1, 2, 32768, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float16",
"shape": [1, 2, 32768, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 96, 128] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "cluster-splitkv-q128-kv16384-h3-d128-f32",
"attrs": { "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 3, 128, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float32",
"shape": [1, 3, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float32",
"shape": [1, 3, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 3, 128, 128] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "cluster-splitkv-q128-kv16384-h5-d128-f32",
"attrs": { "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 5, 128, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.01, "cosStep": 0.02 }
},
"k": {
"dtype": "float32",
"shape": [1, 5, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.03, "cosStep": 0.04 }
},
"v": {
"dtype": "float32",
"shape": [1, 5, 16384, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.05, "cosStep": 0.06 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 5, 128, 128] } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "rfdetr-backbone-global-6h-d64-s1616-b1",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 6, "kvHeads": 6, "qSeq": 1616, "kvSeq": 1616, "headDim": 64 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 6, 1616, 64], "dist": "normal", "seed": 700, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 6, 1616, 64], "dist": "normal", "seed": 701, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 6, 1616, 64], "dist": "normal", "seed": 702, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 6, 1616, 64] } },
"bench": {
"metrics": [{ "type": "gflops", "value": "1 * 4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }]
},
"tunableSpace": { "MATERIALIZED_SGMAT_PRIVATE_ROW_STATS": [0] }
},
{
"name": "rfdetr-backbone-windowed-6h-d64-s101-b16",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 6, "kvHeads": 6, "qSeq": 101, "kvSeq": 101, "headDim": 64 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [16, 6, 101, 64], "dist": "normal", "seed": 700, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [16, 6, 101, 64], "dist": "normal", "seed": 701, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [16, 6, 101, 64], "dist": "normal", "seed": 702, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [16, 6, 101, 64] } },
"bench": {
"metrics": [{ "type": "gflops", "value": "16 * 4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }]
}
},
{
"name": "depthanything-dinov2-global-6h-d64-s1814-b1",
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 6, 1814, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.011, "cosStep": 0.021 }
},
"k": {
"dtype": "float32",
"shape": [1, 6, 1814, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.023 }
},
"v": {
"dtype": "float32",
"shape": [1, 6, 1814, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.017, "cosStep": 0.029 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 6, 1814, 64] } },
"tunableSpace": { "MATERIALIZED_SGMAT_PRIVATE_ROW_STATS": [0] },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 1) * dim(shapes.q, 2) * dim(shapes.k, 2) * dim(shapes.q, 3)" }
]
}
},
{
"name": "dinov2-encoder-attn-f32-6h-d64-s1814",
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 6, "kvHeads": 6, "qSeq": 1814, "kvSeq": 1814, "headDim": 64 },
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 6, 1814, 64], "dist": "normal", "seed": 9401, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 6, 1814, 64], "dist": "normal", "seed": 9402, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 6, 1814, 64], "dist": "normal", "seed": 9403, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 6, 1814, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-additive-f32-4h-d64-s512",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Long keys (512) with 128 cluster blocks separate key-loop length from total block count in masked d64 attention."
},
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 9461, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 9462, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 9463, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [512, 512], "dist": "normal", "seed": 9464, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 512, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-additive-f32-24h-d64-s256",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Short keys (256) with 384 cluster blocks separate total block count from key-loop length in masked d64 attention."
},
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 24, "kvHeads": 24, "qSeq": 256, "kvSeq": 256, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 24, 256, 64], "dist": "normal", "seed": 9471, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 24, 256, 64], "dist": "normal", "seed": 9472, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 24, 256, 64], "dist": "normal", "seed": 9473, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [256, 256], "dist": "normal", "seed": 9474, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 24, 256, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-additive-f32-8h-d64-s128",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Measures Attention over an 8-head, 128-token sequence with an additive float32 mask, head width 64."
},
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 8, "qSeq": 128, "kvSeq": 128, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9441, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9442, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9443, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [128, 128], "dist": "normal", "seed": 9444, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 128, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "mask-qk-additive-f32-12h-d64-s256",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Measures Attention over a 12-head, 256-token sequence with an additive float32 mask, head width 64."
},
"preset": "model",
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 256, "kvSeq": 256, "headDim": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9451, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9452, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9453, "scale": 0.2 },
"attn_mask": { "dtype": "float32", "shape": [256, 256], "dist": "normal", "seed": 9454, "scale": 0.1 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 256, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h8-d64-s128",
"preset": "model",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Measures Attention over an 8-head, non-causal, 128-token sequence, head width 64."
},
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 8, "qSeq": 128, "kvSeq": 128, "headDim": 64 },
"attrs": { "is_causal": 0, "scale": 0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9401, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9402, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "normal", "seed": 9403, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 128, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h12-d64-s128",
"preset": "model",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Measures Attention over a 12-head, non-causal, 128-token sequence, head width 64."
},
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 128, "kvSeq": 128, "headDim": 64 },
"attrs": { "is_causal": 0, "scale": 0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 128, 64], "dist": "normal", "seed": 9411, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 128, 64], "dist": "normal", "seed": 9412, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 128, 64], "dist": "normal", "seed": 9413, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 128, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h12-d64-s256",
"preset": "model",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "Measures Attention over a 12-head, non-causal, 256-token sequence, head width 64."
},
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 256, "kvSeq": 256, "headDim": 64 },
"attrs": { "is_causal": 0, "scale": 0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9421, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9422, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "normal", "seed": 9423, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 256, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "flash-prefill-noncausal-f32-h12-d64-s512",
"preset": "model",
"provenance": {
"source": "hand-authored attention geometry",
"notes": "A noncausal headDim-64 prefill with 384 cluster blocks isolates unmasked clustered attention at this workgroup count."
},
"vars": { "dtype": "float32", "qHeads": 12, "kvHeads": 12, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"attrs": { "is_causal": 0, "scale": 0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 9431, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 9432, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 9433, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 12, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }] }
},
{
"name": "query_grid_q32_k37_h63_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 4032], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 4032], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 4032], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 4032], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h63_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 4032], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 4032], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 4032], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 4032], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h63_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 6048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 6048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 6048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 6048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h63_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 6048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 6048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 6048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 6048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 6144], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 6144], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 6144], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 6144], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h63_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 8064], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 8064], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 8064], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 8064], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h63_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 63, "kvHeads": 63, "qSeq": 32, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 63, "kv_num_heads": 63 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 8064], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 8064], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 8064], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 8064], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 8192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 8192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 8192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 8192], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q32_k37_h64_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 32, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 32, 8192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 8192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 8192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 8192], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_grid_q512_k509_h4_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 509, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 509, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 509, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 3072], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 3072], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 3072], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 3072], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 3072], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 3072], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 3072], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 3072], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 6144], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 6144], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 6144], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 6144], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b2_q129_k193_h16_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 129, "kvSeq": 193, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 193, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 193, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b3_q65_k23_h16_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 65, "kvSeq": 23, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 65, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 23, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 23, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 65, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q64_k129_h32_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 32, "kvHeads": 32, "qSeq": 64, "kvSeq": 129, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 32, "kv_num_heads": 32 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 64, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 129, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 129, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 8192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 8192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 8192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 8192], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "query_transfer_b1_q33_k257_h64_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 64, "kvHeads": 64, "qSeq": 33, "kvSeq": 257, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 64, "kv_num_heads": 64 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 8192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 257, 8192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 257, 8192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 8192], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q33_k37_h4_d64_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q33_k37_h4_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q33_k37_h4_d256_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 33, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 37, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 37, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q1024_k1024_h16_d128_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q1024_k1024_h16_d256_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q33_k37_h4_d64_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 33, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 37, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 37, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q33_k37_h4_d128_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 33, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 37, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 37, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q33_k37_h4_d256_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 4, "kvHeads": 4, "qSeq": 33, "kvSeq": 37, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 33, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 37, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 37, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q1024_k1024_h16_d64_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 1024, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q1024_k1024_h16_d128_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1024, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 1024, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q1024_k1024_h16_d256_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 1024, "kvSeq": 1024, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 1024, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q257_k263_h16_d64_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 257, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 263, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 263, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 257, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q257_k263_h16_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 257, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 263, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 263, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 257, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q257_k263_h16_d128_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 257, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 263, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 263, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 257, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float32_q257_k263_h16_d160_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 160 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 257, 2560], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 263, 2560], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 263, 2560], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 257, 2560], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q257_k263_h16_d64_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 257, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 263, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 263, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 257, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q257_k263_h16_d96_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 257, 1536], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 263, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 263, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 257, 1536], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q257_k263_h16_d128_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 257, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 263, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 263, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 257, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_float16_q257_k263_h16_d160_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 16, "kvHeads": 16, "qSeq": 257, "kvSeq": 263, "headDim": 160 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 16 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 257, 2560], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 263, 2560], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 263, 2560], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 257, 2560], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b2_float32_q65_k129_h8_d96_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 8, "qSeq": 65, "kvSeq": 129, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 8, "kv_num_heads": 8 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 65, 768], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 129, 768], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 129, 768], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 65, 768], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b3_float32_q129_k67_h4_d96_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 129, "kvSeq": 67, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 129, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 67, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 67, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 129, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b2_float32_q65_k129_h8_d256_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 8, "qSeq": 65, "kvSeq": 129, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 8, "kv_num_heads": 8 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 65, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 129, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 65, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b3_float32_q129_k67_h4_d256_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 129, "kvSeq": 67, "headDim": 256 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 129, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 67, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 67, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 129, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b2_float16_q65_k129_h8_d96_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 8, "kvHeads": 8, "qSeq": 65, "kvSeq": 129, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 8, "kv_num_heads": 8 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 65, 768], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 129, 768], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 129, 768], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 65, 768], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b3_float16_q129_k67_h4_d96_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 4, "kvHeads": 4, "qSeq": 129, "kvSeq": 67, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float16", "shape": [3, 129, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [3, 67, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [3, 67, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [3, 129, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b2_float16_q65_k129_h8_d256_c0",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 8, "kvHeads": 8, "qSeq": 65, "kvSeq": 129, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 8, "kv_num_heads": 8 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 65, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 129, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 129, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 65, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "portable_transfer_b3_float16_q129_k67_h4_d256_c1",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 4, "kvHeads": 4, "qSeq": 129, "kvSeq": 67, "headDim": 256 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float16", "shape": [3, 129, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [3, 67, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [3, 67, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [3, 129, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d31_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 31 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 124], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 124], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 124], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 124], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d31_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 31 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 124], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 124], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 124], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 124], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d32_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 32 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 128], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d32_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 32 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 128], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d33_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 33 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 132], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 132], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 132], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 132], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d33_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 33 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 132], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 132], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 132], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 132], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d63_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 63 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 252], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 252], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 252], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 252], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d63_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 63 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 252], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 252], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 252], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 252], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d64_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d65_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 65 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 260], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 260], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 260], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 260], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d65_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 65 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 260], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 260], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 260], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 260], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d96_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d96_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 384], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 384], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d127_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 127 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 508], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 508], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 508], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 508], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d127_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 127 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 508], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 508], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 508], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 508], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d128_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d128_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d129_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 129 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 516], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 516], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 516], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 516], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d129_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 129 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 516], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 516], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 516], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 516], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d160_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 160 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 640], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 640], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 640], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 640], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d160_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 160 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 640], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 640], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 640], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 640], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d192_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 192 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 768], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 768], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 768], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 768], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d192_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 192 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 768], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 768], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 768], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 768], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d224_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 224 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 896], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 896], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 896], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 896], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d224_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 224 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 896], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 896], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 896], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 896], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d255_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 255 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1020], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1020], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1020], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1020], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d255_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 255 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1020], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1020], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1020], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1020], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d256_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 256 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d257_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1028], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1028], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1028], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1028], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh4_d257_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1028], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1028], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1028], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1028], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k511_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 511, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k511_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 511, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k512_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 512, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k512_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 512, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k513_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 513, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q511_k513_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 511, "kvSeq": 513, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k511_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 511, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k511_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 511, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k512_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q512_k512_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k513_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 513, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q512_k513_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 513, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k511_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 511, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k511_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 511, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k512_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 512, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q513_k512_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 512, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k513_h4_kvh4_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 513, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": { "MATERIALIZED_SUBGROUP_MHA_MIN_HEAD_DIM": [64, 96, 128] }
},
{
"name": "materialized_b1_q513_k513_h4_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 513, "kvSeq": 513, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh1_d64_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 64 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh1_d64_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 64 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h8_kvh2_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 8, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h8_kvh2_d256_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 8, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 256 },
"attrs": { "is_causal": 1, "q_num_heads": 8, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b2_q513_k527_h4_kvh2_d96_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 96 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 513, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b2_q513_k527_h4_kvh2_d96_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 96 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 513, 384], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b3_q512_k512_h2_kvh1_d128_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b3_q512_k512_h2_kvh1_d128_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 512, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [3, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [3, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [3, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [3, 512, 256], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1024_k1024_h1_kvh1_d32_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 1024, "kvSeq": 1024, "headDim": 32 },
"attrs": { "is_causal": 0, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1024_k1024_h1_kvh1_d32_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 1024, "kvSeq": 1024, "headDim": 32 },
"attrs": { "is_causal": 1, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 32], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1024_k1024_h16_kvh4_d256_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 4, "qSeq": 1024, "kvSeq": 1024, "headDim": 256 },
"attrs": { "is_causal": 0, "q_num_heads": 16, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1024_k1024_h16_kvh4_d256_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 16, "kvHeads": 4, "qSeq": 1024, "kvSeq": 1024, "headDim": 256 },
"attrs": { "is_causal": 1, "q_num_heads": 16, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1024, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1024, 4096], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1536_k512_h4_kvh2_d128_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 1536, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1536, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1536, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q1536_k512_h4_kvh2_d128_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 1536, "kvSeq": 512, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 1536, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 1536, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k1536_h4_kvh2_d128_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 512, "kvSeq": 1536, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1536, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1536, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q512_k1536_h4_kvh2_d128_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 512, "kvSeq": 1536, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 1536, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 1536, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c0_s0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c1_s0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c0_s-0.125",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c1_s-0.125",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c0_s0.0625",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0.0625 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c1_s0.0625",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0.0625 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c0_s1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_b1_q513_k527_h4_kvh2_d128_c1_s1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 2, "qSeq": 513, "kvSeq": 527, "headDim": 128 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 512], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d257_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 514], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 514], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d257_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 514], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 514], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d288_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 288 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 576], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 288], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 288], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 576], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d288_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 288 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 576], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 288], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 288], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 576], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d512_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 512 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d512_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 512 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d513_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1026], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d513_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 513 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1026], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d1024_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1024 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d1024_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1024 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1024], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d1025_c0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2050], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2050], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "wide_b1_q513_k527_h2_kvh1_d1025_c1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 2050], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 2050], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d258_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 258 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 774], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 258], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 258], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 774], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d258_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 258 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 774], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 258], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 258], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 774], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d259_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 259 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 777], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 259], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 259], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 777], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d259_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 259 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 777], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 259], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 259], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 777], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d287_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 287 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 861], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 287], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 287], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 861], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d287_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 287 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 861], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 287], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 287], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 861], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d289_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 289 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 867], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 289], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 289], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 867], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d289_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 289 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 867], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 289], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 289], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 867], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d511_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 511 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1533], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 511], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 511], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1533], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d511_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 511 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1533], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 511], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 511], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1533], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d767_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 767 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 2301], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 767], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 767], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 2301], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d767_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 767 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 2301], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 767], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 767], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 2301], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d1023_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 1023 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 3069], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1023], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1023], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 3069], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d1023_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 1023 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 3069], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1023], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1023], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 3069], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d1536_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 1536 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 4608], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 4608], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d1536_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 1536 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 4608], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 1536], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 1536], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 4608], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d2048_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 2048 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k513_h3_kvh1_d2048_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 513, "headDim": 2048 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 6144], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 513, 2048], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 6144], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d257_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 771], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 771], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d257_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 257 },
"attrs": { "is_causal": 0, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 257], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 257], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d257_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 257 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d257_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 257 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 771], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 771], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d257_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 257 },
"attrs": { "is_causal": 1, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 257], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 257], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d257_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 257 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1028], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d513_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 1539], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 1539], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d513_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 513], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 513], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d513_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d513_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 513 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 1539], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 1539], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d513_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 513 },
"attrs": { "is_causal": 1, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 513], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 513], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d513_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 513 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 2052], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d1025_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d1025_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1025], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1025], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d1025_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b2_q512_k527_h3_kvh1_d1025_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 512, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [2, 512, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [2, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [2, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [2, 512, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k512_h1_kvh1_d1025_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 1, "kvHeads": 1, "qSeq": 513, "kvSeq": 512, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 1, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 1025], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 1025], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k512_h4_kvh4_d1025_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 4, "kvHeads": 4, "qSeq": 512, "kvSeq": 512, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 4 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 4100], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q511_k512_h2_kvh1_d513_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 511, "kvSeq": 512, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 511, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 512, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 511, 1026], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q512_k511_h2_kvh1_d513_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 2, "kvHeads": 1, "qSeq": 512, "kvSeq": 511, "headDim": 513 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 512, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 511, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 511, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 512, 1026], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c0_s0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1, "scale": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c1_s0",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1, "scale": 0 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c0_s-0.125",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c1_s-0.125",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c0_s1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 0, "q_num_heads": 3, "kv_num_heads": 1, "scale": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float32_b1_q513_k527_h3_kvh1_d1025_c1_s1",
"preset": "stress",
"vars": { "dtype": "float32", "qHeads": 3, "kvHeads": 1, "qSeq": 513, "kvSeq": 527, "headDim": 1025 },
"attrs": { "is_causal": 1, "q_num_heads": 3, "kv_num_heads": 1, "scale": 1 },
"inputs": {
"q": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float32", "shape": [1, 527, 1025], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 3075], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float16_b1_q512_k512_h2_kvh1_d512_c0_sdefault",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 2, "kvHeads": 1, "qSeq": 512, "kvSeq": 512, "headDim": 512 },
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "general_float16_b1_q512_k512_h2_kvh1_d512_c1_sdefault",
"preset": "stress",
"vars": { "dtype": "float16", "qHeads": 2, "kvHeads": 1, "qSeq": 512, "kvSeq": 512, "headDim": 512 },
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": {
"metrics": [
{ "type": "gflops", "value": "4 * dim(shapes.q, 0) * args.qHeads * args.qSeq * args.kvSeq * args.headDim" }
]
},
"tunableSpace": {}
},
{
"name": "materialized_narrow_r3_d1_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 4],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 2],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 2],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 4] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d1_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 4],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 2],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 2],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 4] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d7_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 28],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 14],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 14],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 28] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d7_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 28],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 14],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 14],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 28] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d8_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 32] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d8_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 16],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 32] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d15_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 60],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 30],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 30],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 60] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d15_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 60],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 30],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 30],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 60] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d16_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d16_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 32],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 64] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d17_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 68],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 34],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 34],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 68] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d17_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 68],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 34],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 34],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 68] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d31_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 124],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 62],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 62],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 124] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d31_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 124],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 62],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 62],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 124] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d32_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r3_d32_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 513, 128],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 527, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 527, 64],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 513, 128] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_narrow_r4_d1_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 1] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_narrow_r4_d1_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 1],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 1] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_narrow_r4_d15_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 15] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_narrow_r4_d15_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 15],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 15] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_narrow_r4_d31_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 31] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_narrow_r4_d31_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 4, 513, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 2, 527, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 2, 527, 31],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 513, 31] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "rank3_cluster_float16_d256_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1024] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d256_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1024] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d288_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1152],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1152] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d288_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1152],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1152] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d480_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1920],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1920] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d480_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 1920],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 1920] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d512_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 2048],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 2048] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d512_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 2048],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 2048] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d544_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 2176],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 2176] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float16_d544_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float16",
"shape": [1, 33, 2176],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float16",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float16",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 33, 2176] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d256_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1024] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d256_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 512],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1024] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d288_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1152],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1152] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d288_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1152],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 576],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1152] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d480_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1920],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1920] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d480_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 1920],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 960],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 1920] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d512_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 2048],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 2048] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d512_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 2048],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 1024],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 2048] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d544_c0",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 0 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 2176],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 2176] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "rank3_cluster_float32_d544_c1",
"attrs": { "q_num_heads": 4, "kv_num_heads": 2, "is_causal": 1 },
"inputs": {
"q": {
"dtype": "float32",
"shape": [1, 33, 2176],
"data": { "kind": "fillFloat32", "sinStep": 0.021, "cosStep": 0.033, "scale": 0.2 }
},
"k": {
"dtype": "float32",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.039, "cosStep": 0.011, "scale": 0.2 }
},
"v": {
"dtype": "float32",
"shape": [1, 47, 1088],
"data": { "kind": "fillFloat32", "sinStep": 0.027, "cosStep": 0.049, "scale": 0.2, "offset": 1.0 }
}
},
"outputs": { "y": { "dtype": "float32", "shape": [1, 33, 2176] } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d1_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 2], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 1], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 1], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 2], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d1_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 2], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 1], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 1], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 2], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d7_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 14], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 7], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 7], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 14], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d7_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 14], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 7], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 7], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 14], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d31_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 62], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 31], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 31], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 62], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d31_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 62], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 31], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 31], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 62], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d32_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 32], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 32], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d32_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 32], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 32], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d64_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d64_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d96_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 192], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d96_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 192], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d127_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 254], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 127], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 127], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 254], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d127_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 254], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 127], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 127], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 254], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d128_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d128_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d256_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d256_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d257_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 514], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 514], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d257_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 514], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 257], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 257], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 514], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d288_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 576], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 288], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 288], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 576], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d288_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 576], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 288], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 288], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 576], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d511_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1022], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 511], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 511], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1022], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d511_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1022], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 511], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 511], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1022], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d512_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d512_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1024], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d513_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1026], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d513_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 1026], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 513], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 513], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 1026], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d31_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 31], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 31], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 31], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 31], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d31_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 31], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 31], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 31], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 31], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d64_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d64_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d256_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r4_b1_q513_k527_h2_kvh1_d256_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 513, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_r3_b2_q512_k512_h4_kvh2_d128_c0_s0",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 512, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b2_q512_k512_h4_kvh2_d128_c0_s-0.125",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 4, "kv_num_heads": 2, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 512, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b2_q512_k512_h4_kvh2_d128_c1_s0",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 512, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b2_q512_k512_h4_kvh2_d128_c1_s-0.125",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 4, "kv_num_heads": 2, "scale": -0.125 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 512, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 512, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q511_k512_h2_kvh1_d512_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 511, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 512, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 511, 1024], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q512_k511_h2_kvh1_d512_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 511, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 511, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 512, 1024], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q2048_k2048_h2_kvh1_d64_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2048, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2048, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2048, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2048, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q2048_k2048_h2_kvh1_d64_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2048, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2048, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2048, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2048, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q2048_k2048_h2_kvh1_d96_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2048, 192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2048, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2048, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2048, 192], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q2048_k2048_h2_kvh1_d96_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2048, 192], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2048, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2048, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2048, 192], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d160_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 320], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 160], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 160], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 320], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d160_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 320], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 160], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 160], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 320], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d192_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 384], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d192_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 384], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 192], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 192], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 384], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d224_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 448], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 224], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 224], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 448], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh1_d224_c1_sdefault",
"preset": "stress",
"attrs": { "is_causal": 1, "q_num_heads": 2, "kv_num_heads": 1 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 448], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 224], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 224], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 448], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh2_d64_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh2_d128_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_r3_b1_q513_k527_h2_kvh2_d256_c0_sdefault",
"preset": "stress",
"attrs": { "is_causal": 0, "q_num_heads": 2, "kv_num_heads": 2 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 513, 512], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 527, 512], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 513, 512], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 1)" }] }
},
{
"name": "materialized_half_rank4_grid_h4_kvh1_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 4, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h8_kvh2_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h16_kvh4_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 16, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 4, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h8_kvh2_d64_q513_k527",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 513, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 527, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 527, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 513, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h4_kvh1_d96_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 4, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 512, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 512, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 4, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h8_kvh2_d96_q513_k527",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 513, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 527, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 527, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 513, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h8_kvh2_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h16_kvh4_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 16, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 4, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 4, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h32_kvh8_d128_q513_k527",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 513, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 513, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h32_kvh8_d128_q512_k527",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 527, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 527, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h16_kvh4_d256_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 16, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 4, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 4, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 512, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_grid_h32_kvh8_d256_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h2_kvh1_d96_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 512, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 512, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h32_kvh8_d96_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h32_kvh8_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 32, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 8, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 8, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h24_kvh6_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 24, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 6, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 6, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 24, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h24_kvh6_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 24, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 6, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 6, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 24, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_boundary_h24_kvh6_d256_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 24, 512, 256], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 6, 512, 256], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 6, 512, 256], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 24, 512, 256], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h12_kvh3_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 12, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 3, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 3, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 12, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h12_kvh3_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 12, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 3, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 3, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 12, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h20_kvh5_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 20, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 5, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 5, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 20, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h20_kvh5_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 20, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 5, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 5, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 20, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b2_h8_kvh2_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 8, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 2, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 2, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 8, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b2_h8_kvh2_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 8, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 2, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 2, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 8, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b2_h16_kvh4_d64_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 16, 512, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 4, 512, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 4, 512, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 16, 512, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b2_h16_kvh4_d96_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 16, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 4, 512, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 4, 512, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 16, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b2_h16_kvh4_d128_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [2, 16, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [2, 4, 512, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [2, 4, 512, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [2, 16, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d64_q768_k768",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 768, 64], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 768, 64], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 768, 64], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 768, 64], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d96_q768_k768",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 768, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 768, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 768, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 768, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d128_q768_k768",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 768, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 768, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 768, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 768, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h16_kvh4_d128_q768_k768",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 16, 768, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 4, 768, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 4, 768, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 768, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d96_q512_k1023",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 1023, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 1023, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d96_q512_k1024",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 1024, 96], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 1024, 96], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 96], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d128_q512_k1023",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 1023, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 1023, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d128_q512_k1024",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 1024, 128], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 1024, 128], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 128], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h2_kvh1_d80_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 512, 80], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 512, 80], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 512, 80], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d80_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 80], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 512, 80], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 80], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h2_kvh1_d112_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 2, 512, 112], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 1, 512, 112], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 1, 512, 112], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 2, 512, 112], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
},
{
"name": "materialized_half_rank4_selection_b1_h8_kvh2_d112_q512_k512",
"preset": "stress",
"attrs": { "is_causal": 0 },
"inputs": {
"q": { "dtype": "float16", "shape": [1, 8, 512, 112], "dist": "normal", "seed": 8811, "scale": 0.2 },
"k": { "dtype": "float16", "shape": [1, 2, 512, 112], "dist": "normal", "seed": 8812, "scale": 0.2 },
"v": { "dtype": "float16", "shape": [1, 2, 512, 112], "dist": "normal", "seed": 8813, "scale": 0.2 }
},
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 512, 112], "dist": "empty" } },
"bench": { "metrics": [{ "type": "gflops", "value": "4 * numel(shapes.q) * dim(shapes.k, 2)" }] }
}
]
}