| { |
| "op": "com.microsoft.LinearAttention", |
| "tunableSpace": { |
| "dvGroups": [2, 4, 8], |
| "tileV": [4, 8, 16], |
| "gatedTileV": [2, 4, 8], |
| "chunkSize": [16, 32], |
| "chunkTileV": [16, 32] |
| }, |
| "cases": [ |
| { |
| "name": "linear-attention-f32-zero-32x4x16x16", |
| "preset": "smoke", |
| "vars": { "batch": 1, "seq": 32, "qHeads": 4, "kvHeads": 2, "dk": 16, "dv": 16 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 2, "update_rule": "linear", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [1, 32, 64], "dtype": "float32", "dist": "normal", "seed": 201, "scale": 0.2 }, |
| "keyT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 202, "scale": 0.2 }, |
| "valueT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 203, "scale": 0.2 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 32, 64], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 2, 16, 16], "dtype": "float32" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { "type": "gflops", "value": "2 * args.batch * args.seq * (args.qHeads + args.kvHeads) * args.dk * args.dv" } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-f32-state-32x4x16x16", |
| "preset": "smoke", |
| "vars": { "batch": 1, "seq": 32, "qHeads": 4, "kvHeads": 2, "dk": 16, "dv": 16 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 2, "update_rule": "gated_delta", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [1, 32, 64], "dtype": "float32", "dist": "normal", "seed": 202, "scale": 0.2 }, |
| "keyT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 203, "scale": 0.2 }, |
| "valueT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 204, "scale": 0.2 }, |
| "pastStateT": { "shape": [1, 2, 16, 16], "dtype": "float32", "dist": "normal", "seed": 205, "scale": 0.1 }, |
| "decayT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 206, "scale": 0.1 }, |
| "betaT": { "shape": [1, 32, 2], "dtype": "float32", "dist": "normal", "seed": 207, "scale": 0.1 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 32, 64], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 2, 16, 16], "dtype": "float32" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-f32-linear-state-32x4x16x16", |
| "preset": "smoke", |
| "vars": { "batch": 1, "seq": 32, "qHeads": 4, "kvHeads": 2, "dk": 16, "dv": 16 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 2, "update_rule": "linear", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [1, 32, 64], "dtype": "float32", "dist": "normal", "seed": 208, "scale": 0.2 }, |
| "keyT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 209, "scale": 0.2 }, |
| "valueT": { "shape": [1, 32, 32], "dtype": "float32", "dist": "normal", "seed": 210, "scale": 0.2 }, |
| "pastStateT": { "shape": [1, 2, 16, 16], "dtype": "float32", "dist": "normal", "seed": 211, "scale": 0.1 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 32, 64], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 2, 16, 16], "dtype": "float32" } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * args.batch * args.seq * (args.qHeads + args.kvHeads) * args.dk * args.dv" } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-linear-state-scalar-f16-seq1536-pathology", |
| "preset": "stress", |
| "provenance": { |
| "source": "authored for branch coverage", |
| "notes": "Long-sequence supplied-state case at head_dim_k 16. It exercises the recurrent small-dk route's serial token recurrence and distinguishes it from the chunked prefill decomposition." |
| }, |
| "vars": { "batch": 4, "seq": 1536, "qHeads": 4, "kvHeads": 2, "dk": 16, "dv": 16 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 2, "update_rule": "linear", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [4, 1536, 64], "dtype": "float16", "dist": "normal", "seed": 314, "scale": 0.2 }, |
| "keyT": { "shape": [4, 1536, 32], "dtype": "float16", "dist": "normal", "seed": 315, "scale": 0.2 }, |
| "valueT": { "shape": [4, 1536, 32], "dtype": "float16", "dist": "normal", "seed": 316, "scale": 0.2 }, |
| "pastStateT": { "shape": [4, 2, 16, 16], "dtype": "float16", "dist": "normal", "seed": 317, "scale": 0.1 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [4, 1536, 64], "dtype": "float16" }, |
| "presentStateT": { "shape": [4, 2, 16, 16], "dtype": "float16" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { "type": "gflops", "value": "2 * args.batch * args.seq * (args.qHeads + args.kvHeads) * args.dk * args.dv" } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-gated_delta-scalar-headdimk6-seq1536-stress", |
| "preset": "stress", |
| "vars": { "batch": 8, "seq": 1536, "qHeads": 4, "kvHeads": 4, "dk": 6, "dv": 12 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 4, "update_rule": "gated_delta", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [8, 1536, 24], "dtype": "float32", "dist": "normal", "seed": 301, "scale": 0.2 }, |
| "keyT": { "shape": [8, 1536, 24], "dtype": "float32", "dist": "normal", "seed": 302, "scale": 0.2 }, |
| "valueT": { "shape": [8, 1536, 48], "dtype": "float32", "dist": "normal", "seed": 303, "scale": 0.2 }, |
| "pastStateT": { "shape": [8, 4, 6, 12], "dtype": "float32", "dist": "normal", "seed": 304, "scale": 0.1 }, |
| "decayT": { "shape": [8, 1536, 4], "dtype": "float32", "dist": "normal", "seed": 305, "scale": 0.1 }, |
| "betaT": { "shape": [8, 1536, 4], "dtype": "float32", "dist": "normal", "seed": 306, "scale": 0.1 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [8, 1536, 48], "dtype": "float32" }, |
| "presentStateT": { "shape": [8, 4, 6, 12], "dtype": "float32" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-linear-scalar-f16-seq1536-stress", |
| "preset": "stress", |
| "vars": { "batch": 4, "seq": 1536, "qHeads": 4, "kvHeads": 2, "dk": 16, "dv": 16 }, |
| "attrs": { "q_num_heads": 4, "kv_num_heads": 2, "update_rule": "linear", "scale": 0.25 }, |
| "inputs": { |
| "queryT": { "shape": [4, 1536, 64], "dtype": "float16", "dist": "normal", "seed": 311, "scale": 0.2 }, |
| "keyT": { "shape": [4, 1536, 32], "dtype": "float16", "dist": "normal", "seed": 312, "scale": 0.2 }, |
| "valueT": { "shape": [4, 1536, 32], "dtype": "float16", "dist": "normal", "seed": 313, "scale": 0.2 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [4, 1536, 64], "dtype": "float16" }, |
| "presentStateT": { "shape": [4, 2, 16, 16], "dtype": "float16" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { "type": "gflops", "value": "2 * args.batch * args.seq * (args.qHeads + args.kvHeads) * args.dk * args.dv" } |
| ] |
| }, |
| "provenance": { |
| "source": "authored for branch coverage", |
| "notes": "Zero-state sibling of the supplied-state long-sequence case, covering the same branch and shape without an entry state." |
| } |
| }, |
| { |
| "name": "linear-attention-gated-delta-f32-bonsai-m16-h48-kv16-dk128-dv128", |
| "preset": "stress", |
| "vars": { "batch": 1, "seq": 16, "qHeads": 48, "kvHeads": 16, "dk": 128, "dv": 128 }, |
| "attrs": { "q_num_heads": 48, "kv_num_heads": 16, "update_rule": "gated_delta", "scale": 0.08838834764831845 }, |
| "inputs": { |
| "queryT": { "shape": [1, 16, 6144], "dtype": "float32", "dist": "normal", "seed": 320, "scale": 0.05 }, |
| "keyT": { "shape": [1, 16, 2048], "dtype": "float32", "dist": "normal", "seed": 321, "scale": 0.05 }, |
| "valueT": { "shape": [1, 16, 2048], "dtype": "float32", "dist": "normal", "seed": 322, "scale": 0.05 }, |
| "pastStateT": { "shape": [1, 16, 128, 128], "dtype": "float32", "dist": "normal", "seed": 323, "scale": 0.02 }, |
| "decayT": { "shape": [1, 16, 16], "dtype": "float32", "dist": "normal", "seed": 324, "scale": 0.08 }, |
| "betaT": { "shape": [1, 16, 16], "dtype": "float32", "dist": "normal", "seed": 325, "scale": 0.08 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 16, 6144], "dtype": "float32", "dist": "empty" }, |
| "presentStateT": { "shape": [1, 16, 128, 128], "dtype": "float32", "dist": "empty" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-gated-delta-f16-bonsai-m16-h48-kv16-dk128-dv128", |
| "preset": "stress", |
| "vars": { "batch": 1, "seq": 16, "qHeads": 48, "kvHeads": 16, "dk": 128, "dv": 128 }, |
| "attrs": { "q_num_heads": 48, "kv_num_heads": 16, "update_rule": "gated_delta", "scale": 0.08838834764831845 }, |
| "inputs": { |
| "queryT": { "shape": [1, 16, 6144], "dtype": "float16", "dist": "normal", "seed": 326, "scale": 0.05 }, |
| "keyT": { "shape": [1, 16, 2048], "dtype": "float16", "dist": "normal", "seed": 327, "scale": 0.05 }, |
| "valueT": { "shape": [1, 16, 2048], "dtype": "float16", "dist": "normal", "seed": 328, "scale": 0.05 }, |
| "pastStateT": { "shape": [1, 16, 128, 128], "dtype": "float16", "dist": "normal", "seed": 329, "scale": 0.02 }, |
| "decayT": { "shape": [1, 16, 16], "dtype": "float16", "dist": "normal", "seed": 330, "scale": 0.08 }, |
| "betaT": { "shape": [1, 16, 16], "dtype": "float16", "dist": "normal", "seed": 331, "scale": 0.08 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 16, 6144], "dtype": "float16", "dist": "empty" }, |
| "presentStateT": { "shape": [1, 16, 128, 128], "dtype": "float16", "dist": "empty" } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-qwen3next-decode-s1", |
| "preset": "model", |
| "provenance": { |
| "notes": "Qwen3-Next class defaults (linear_num_value_heads 32, linear_num_key_heads 16, linear_key_head_dim 128, linear_value_head_dim 128) at a decode step, where the recurrence carries the whole cost." |
| }, |
| "vars": { "batch": 1, "seq": 1, "qHeads": 32, "kvHeads": 16, "dk": 128, "dv": 128 }, |
| "attrs": { |
| "q_num_heads": 32, |
| "kv_num_heads": 16, |
| "update_rule": "gated_delta", |
| "scale": 0.08838834764831843, |
| "chunk_size": 64 |
| }, |
| "inputs": { |
| "queryT": { "shape": [1, 1, 4096], "dtype": "float32", "dist": "normal", "seed": 8100, "scale": 0.3 }, |
| "keyT": { "shape": [1, 1, 2048], "dtype": "float32", "dist": "normal", "seed": 8101, "scale": 0.3 }, |
| "valueT": { "shape": [1, 1, 2048], "dtype": "float32", "dist": "normal", "seed": 8102, "scale": 0.3 }, |
| "pastStateT": { "shape": [1, 16, 128, 128], "dtype": "float32", "dist": "normal", "seed": 8103, "scale": 0.1 }, |
| "decayT": { "shape": [1, 1, 2048], "dtype": "float32", "dist": "uniform", "seed": 8104, "min": 0.9, "max": 1 }, |
| "betaT": { "shape": [1, 1, 16], "dtype": "float32", "dist": "uniform", "seed": 8105, "min": 0.1, "max": 0.9 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 1, 4096], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 16, 128, 128], "dtype": "float32" } |
| }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-qwen3next-prefill-s512", |
| "preset": "model", |
| "provenance": { "notes": "Qwen3-Next class defaults over a 512-token prefill chunk." }, |
| "vars": { "batch": 1, "seq": 512, "qHeads": 32, "kvHeads": 16, "dk": 128, "dv": 128 }, |
| "attrs": { |
| "q_num_heads": 32, |
| "kv_num_heads": 16, |
| "update_rule": "gated_delta", |
| "scale": 0.08838834764831843, |
| "chunk_size": 64 |
| }, |
| "inputs": { |
| "queryT": { "shape": [1, 512, 4096], "dtype": "float32", "dist": "normal", "seed": 8200, "scale": 0.3 }, |
| "keyT": { "shape": [1, 512, 2048], "dtype": "float32", "dist": "normal", "seed": 8201, "scale": 0.3 }, |
| "valueT": { "shape": [1, 512, 2048], "dtype": "float32", "dist": "normal", "seed": 8202, "scale": 0.3 }, |
| "pastStateT": { "shape": [1, 16, 128, 128], "dtype": "float32", "dist": "normal", "seed": 8203, "scale": 0.1 }, |
| "decayT": { "shape": [1, 512, 2048], "dtype": "float32", "dist": "uniform", "seed": 8204, "min": 0.9, "max": 1 }, |
| "betaT": { "shape": [1, 512, 16], "dtype": "float32", "dist": "uniform", "seed": 8205, "min": 0.1, "max": 0.9 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 512, 4096], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 16, 128, 128], "dtype": "float32" } |
| }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "linear-attention-qwen3next-prefill-s2048", |
| "preset": "model", |
| "provenance": { |
| "notes": "Qwen3-Next class defaults over a 2048-token prefill chunk, eight chunk_size 64 blocks per workgroup pass." |
| }, |
| "vars": { "batch": 1, "seq": 2048, "qHeads": 32, "kvHeads": 16, "dk": 128, "dv": 128 }, |
| "attrs": { |
| "q_num_heads": 32, |
| "kv_num_heads": 16, |
| "update_rule": "gated_delta", |
| "scale": 0.08838834764831843, |
| "chunk_size": 64 |
| }, |
| "inputs": { |
| "queryT": { "shape": [1, 2048, 4096], "dtype": "float32", "dist": "normal", "seed": 8300, "scale": 0.3 }, |
| "keyT": { "shape": [1, 2048, 2048], "dtype": "float32", "dist": "normal", "seed": 8301, "scale": 0.3 }, |
| "valueT": { "shape": [1, 2048, 2048], "dtype": "float32", "dist": "normal", "seed": 8302, "scale": 0.3 }, |
| "pastStateT": { "shape": [1, 16, 128, 128], "dtype": "float32", "dist": "normal", "seed": 8303, "scale": 0.1 }, |
| "decayT": { |
| "shape": [1, 2048, 2048], |
| "dtype": "float32", |
| "dist": "uniform", |
| "seed": 8304, |
| "min": 0.9, |
| "max": 1 |
| }, |
| "betaT": { "shape": [1, 2048, 16], "dtype": "float32", "dist": "uniform", "seed": 8305, "min": 0.1, "max": 0.9 } |
| }, |
| "outputs": { |
| "outputT": { "shape": [1, 2048, 4096], "dtype": "float32" }, |
| "presentStateT": { "shape": [1, 16, 128, 128], "dtype": "float32" } |
| }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.seq * (args.qHeads + 2 * args.kvHeads) * args.dk * args.dv" |
| } |
| ] |
| } |
| } |
| ] |
| } |
|
|
