{
  "schema": "kbp-card/v1",
  "blueprint_id": "preview:3ed27b1d08964e89d487415f154d307707d076924bcbfa0e24f039f0f0d36165",
  "lineage": "sha256:7063eae2d813f557cdde5558c2bbb854fc9afe37f43b251dee6049f876cbff45",
  "revision": 0,
  "name": "qwen3.6-35b-a3b-fp8-bf16head",
  "slug": "qwen3.6-35b-a3b-fp8-bf16head--h100-sxm",
  "preview": true,
  "exported_by": "preview: tools/preview/kbp_preview.py over `met circuit plan` (not `met blueprint export`)",
  "model": {
    "repo": "Qwen/Qwen3.6-35B-A3B-FP8",
    "revision": null,
    "arch": "qwen3_6_moe",
    "params": "35B",
    "dtype": "fp8",
    "quantization": "fp8",
    "layers": 40,
    "layer_kinds": {
      "linear_attention": 30,
      "full_attention": 10
    },
    "max_seq_len": 32768,
    "description": "MoE hybrid: 3 GatedDeltaNet layers then 1 gated full-attention layer, 256-expert top-8 MoE with a shared expert, MTP draft head"
  },
  "hardware": {
    "class": "hopper",
    "device": "h100-sxm",
    "name": "NVIDIA H100 SXM5 80GB",
    "ptx_arch": "sm_90a",
    "sm_count": 132,
    "memory_gb": 80,
    "memory_bandwidth_gbps": 3350,
    "world_size": 1,
    "min_visible_mib": null,
    "devices": [
      "h100-sxm"
    ]
  },
  "precision": {
    "tier": "recipe",
    "table": "qwen3.6-35b-a3b-fp8-bf16head",
    "table_sha256": "fc94e079613adb41d7f986f1e4d761a45cb490c87a7697240b56c7c4fd81e276",
    "linear_by_format": [
      {
        "weight": "fp8/block128x128",
        "activation": "bf16",
        "nodes": 180
      },
      {
        "weight": "bf16",
        "activation": "bf16",
        "nodes": 111
      },
      {
        "weight": "fp8/block128x128",
        "activation": "f32",
        "nodes": 80
      }
    ],
    "outputs_by_format": [
      {
        "format": "bf16",
        "nodes": 573
      },
      {
        "format": "f32",
        "nodes": 240
      },
      {
        "format": "f32,i32",
        "nodes": 40
      },
      {
        "format": "f32,f32",
        "nodes": 30
      },
      {
        "format": "bf16,bf16",
        "nodes": 20
      },
      {
        "format": "i32",
        "nodes": 1
      }
    ],
    "kv_cache": "bf16"
  },
  "execution": "executor",
  "kernels": {
    "count": 14,
    "entries": 20,
    "families": 16,
    "residual": 0,
    "launches": 573,
    "groups": {
      "device": 573,
      "host": 2,
      "novel": 0
    },
    "compute": {
      "cuda_core": 261,
      "memory": 212,
      "tensor_core:mma.sync.m16n8k16.bf16": 100
    },
    "kinds": null,
    "bytes_raw": null,
    "bytes_deflated": null,
    "sources": null,
    "list": [
      {
        "id": "causal_conv1d::causal_conv1d_update_l2norm_f32",
        "family": "causal_conv1d_l2norm",
        "groups": 30,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "gated_delta_rule::gated_delta_rule_decode_f32",
        "family": "gdn_recurrence",
        "groups": 30,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "gemv::dense_gemv_bf16",
        "family": "dense_bf16",
        "groups": 41,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "moe_expert_gemv::moe_weighted_sum_blend",
        "family": "moe_blend",
        "groups": 40,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "moe_shared_expert_fused_fp8::moe_expert_gate_up_shared_fp8",
        "family": "moe_fp8_1row",
        "groups": 40,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "moe_shared_expert_fused_fp8::moe_expert_silu_down_shared_fp8",
        "family": "moe_fp8_1row",
        "groups": 40,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "moe_topk::moe_topk_softmax",
        "family": "moe_topk",
        "groups": 40,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "norm::gated_rms_norm_f32_input",
        "family": "gated_rms_norm",
        "groups": 30,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "norm::residual_add_rms_norm",
        "family": "rms_norm",
        "groups": 40,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "norm::rms_norm",
        "family": "rms_norm",
        "groups": 21,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "norm::rms_norm_residual",
        "family": "rms_norm",
        "groups": 1,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "paged_decode::paged_decode_attn",
        "family": "paged_decode_attn",
        "groups": 10,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "reshape_and_cache::reshape_and_cache_flash",
        "family": "kv_write",
        "groups": 10,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "residual_add::bf16_residual_add",
        "family": "residual_add",
        "groups": 1,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "residual_add::sigmoid_gate_mul",
        "family": "sigmoid_gate_mul",
        "groups": 10,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "residual_add_rms_norm_exact::residual_add_rms_norm_exact",
        "family": "rms_norm",
        "groups": 39,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "rope_mrope_interleaved::rope_forward_mrope_interleaved",
        "family": "rope",
        "groups": 10,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "ssm_preprocess::deinterleave_qg",
        "family": "deinterleave_qg",
        "groups": 10,
        "compute": [
          "memory"
        ]
      },
      {
        "id": "ssm_preprocess::dense_gemv_ba_gates",
        "family": "gdn_ba_gates_gemv",
        "groups": 30,
        "compute": [
          "cuda_core"
        ]
      },
      {
        "id": "w8a16_gemm_pipelined_m32::w8a16_gemm_pipelined_m32",
        "family": "w8a16_gemm",
        "groups": 100,
        "compute": [
          "tensor_core:mma.sync.m16n8k16.bf16"
        ]
      }
    ]
  },
  "plans": [
    {
      "mode": "decode",
      "rows": 1,
      "digest": "31d4c6b3314dd12f33e3a75c6d75bca52b0d4efd27afddafef8daa37963e8ad9",
      "groups": 575,
      "launches": 573,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": []
    },
    {
      "mode": "multi_seq",
      "rows": 2,
      "digest": "a29c0dfd865e60df3bcccf2511fba878abc4ba9d57073d169ff0e87fb0798377",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "256cbe7bdc6e416f2e5a7b83d4ad1b4770aba79ea8d04536a1ec0722a32a543a",
          "groups": 575,
          "launches": 733,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 4,
      "digest": "a471c952c48ac6077f18059094f4ebc5ee96ff589a0fc052448c7fb2c91f08c3",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "dad4dfe90213bd1e747c6b3c11931ffcea524da4cff07b5f0bab32aa36c077a0",
          "groups": 575,
          "launches": 973,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 8,
      "digest": "388f30e4e6d4818f30792c5e0c8a5cc3664b1751717dc5ae0e042d4b65f86685",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "8334507bd9ee7d965b726ffd02b598c6086424184dcea5be3956e49784028473",
          "groups": 575,
          "launches": 1453,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 12,
      "digest": "66f7eedd68acab11320c93aedeeb22f908c9ef632b673c9418e5a00911658f53",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "3bcba6625bfeab008a4b69beb6db2335fd63a936c5a0e2be846e23ea9ba2c201",
          "groups": 575,
          "launches": 1933,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 16,
      "digest": "ef971a265d6f7d1ac03539ba695ca36319cba7519db67aff388aee37cd0d8522",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "30b65b151417b98b12a56580434811677cef9d8b1b4ef15f213aa6942059209b",
          "groups": 575,
          "launches": 2413,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 24,
      "digest": "9c497e8477659e2c0c17ddfb3d4051210760a37c44fa5005a7f37b346d6aa62e",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "2f32ed2c7b13afcafaec0bcf0a93efe33690433850ce832f0367ce2f37852436",
          "groups": 575,
          "launches": 3373,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 32,
      "digest": "0cf4c37f9933a00acc8631a4c16a7080f0777e9f8133fdea42229dc2284c3fb6",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "c989b677cf248ec127be60cc0c7a075604d56887187c7cb26916276c9f5a1128",
          "groups": 575,
          "launches": 4333,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 48,
      "digest": "0e7022d3fa53e70872fd7d82f3453dd4aff224751334ca0a669d73328a27ebbc",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "91fdb893da8caaed392db6564f804d9c7a6801798dceba84fe75d15823d9fc73",
          "groups": 575,
          "launches": 6253,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 64,
      "digest": "29a91e56ef7dd4c00c32d1d550f2d75406d02b88ea11407e2f83e6f9a306b076",
      "groups": 575,
      "launches": 613,
      "copies": 0,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "d391f1718b7e012c8c52e633c5b8dae68c130b3dc01fc9ba3fbd9103f67917e1",
          "groups": 575,
          "launches": 8173,
          "copies": 0,
          "edges_fused": 180,
          "edges_materialized": 814
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 96,
      "digest": "f832ef5c75b930c786c59c238225c25fd302ff6d37ea2af1755f8a11ee08d172",
      "groups": 415,
      "launches": 1013,
      "copies": 0,
      "edges_fused": 460,
      "edges_materialized": 534,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "9ec7a71ce145b75ef3459c3c37c4b5957e006c97825f946480a570487703d96d",
          "groups": 415,
          "launches": 12413,
          "copies": 0,
          "edges_fused": 460,
          "edges_materialized": 534
        }
      ]
    },
    {
      "mode": "multi_seq",
      "rows": 128,
      "digest": "d55817abdc1fbc793868bdb91ac5ffc96a518b5d6a18e941bf3969106e89b782",
      "groups": 415,
      "launches": 1013,
      "copies": 0,
      "edges_fused": 460,
      "edges_materialized": 534,
      "routes": [
        {
          "route": "gdn_state_slots_fragmented",
          "when": "when the sequences' GDN state slots are not contiguous in slice order (slots fragment as sequences finish), and the step runs the per-sequence loop; planned as `ssm_batched_recurrent=off` (ml/qwen3_ssm/trait_decode_multi_seq/ssm_batched_recurrent.rs:93-150 (the contiguity check), :396-397 (the per-sequence call))",
          "digest": "c1fc51a03bc6aed5f7d7f4052c698856198a108997a19a78e391d688faead003",
          "groups": 415,
          "launches": 16253,
          "copies": 0,
          "edges_fused": 460,
          "edges_materialized": 534
        }
      ]
    },
    {
      "mode": "verify",
      "rows": 2,
      "digest": "e1078f8739efde34d79b729d2fa190206a26b2f96e4b5435f28a3958b409b6ae",
      "groups": 575,
      "launches": 645,
      "copies": 30,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": []
    },
    {
      "mode": "verify",
      "rows": 3,
      "digest": "76e1d4fda442db7d346a86be561fb3450c3737a710b88e83a49397936f9f4c42",
      "groups": 575,
      "launches": 676,
      "copies": 60,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": []
    },
    {
      "mode": "verify",
      "rows": 4,
      "digest": "0e62ec870943f9188187b1653ec8fe46a728d1e9945c03e81bfb454b48d50f25",
      "groups": 575,
      "launches": 707,
      "copies": 90,
      "edges_fused": 180,
      "edges_materialized": 814,
      "routes": []
    },
    {
      "mode": "draft",
      "rows": 1,
      "digest": "fc0943bc31fa38686371fd31ab4b4cd86e35d8f7904f92560f612755038a4a57",
      "groups": 27,
      "launches": 26,
      "copies": 0,
      "edges_fused": 3,
      "edges_materialized": 33,
      "routes": []
    }
  ],
  "sizes": {
    "body": null,
    "lite_file": null,
    "full_file": null
  },
  "signature": {
    "status": "unverified",
    "suite": "mldsa65+ed25519",
    "kid": null,
    "anchor": null,
    "verified_at": null
  },
  "download": {
    "available": false,
    "note": "No .kbp file exists for a preview: blueprint builds and signing are not available yet."
  },
  "performance": [],
  "evidence": [
    {
      "kind": "circuit_plan",
      "id": "decode:1",
      "digest": "31d4c6b3314dd12f33e3a75c6d75bca52b0d4efd27afddafef8daa37963e8ad9"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:2",
      "digest": "a29c0dfd865e60df3bcccf2511fba878abc4ba9d57073d169ff0e87fb0798377"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:4",
      "digest": "a471c952c48ac6077f18059094f4ebc5ee96ff589a0fc052448c7fb2c91f08c3"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:8",
      "digest": "388f30e4e6d4818f30792c5e0c8a5cc3664b1751717dc5ae0e042d4b65f86685"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:12",
      "digest": "66f7eedd68acab11320c93aedeeb22f908c9ef632b673c9418e5a00911658f53"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:16",
      "digest": "ef971a265d6f7d1ac03539ba695ca36319cba7519db67aff388aee37cd0d8522"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:24",
      "digest": "9c497e8477659e2c0c17ddfb3d4051210760a37c44fa5005a7f37b346d6aa62e"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:32",
      "digest": "0cf4c37f9933a00acc8631a4c16a7080f0777e9f8133fdea42229dc2284c3fb6"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:48",
      "digest": "0e7022d3fa53e70872fd7d82f3453dd4aff224751334ca0a669d73328a27ebbc"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:64",
      "digest": "29a91e56ef7dd4c00c32d1d550f2d75406d02b88ea11407e2f83e6f9a306b076"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:96",
      "digest": "f832ef5c75b930c786c59c238225c25fd302ff6d37ea2af1755f8a11ee08d172"
    },
    {
      "kind": "circuit_plan",
      "id": "multi_seq:128",
      "digest": "d55817abdc1fbc793868bdb91ac5ffc96a518b5d6a18e941bf3969106e89b782"
    },
    {
      "kind": "circuit_plan",
      "id": "verify:2",
      "digest": "e1078f8739efde34d79b729d2fa190206a26b2f96e4b5435f28a3958b409b6ae"
    },
    {
      "kind": "circuit_plan",
      "id": "verify:3",
      "digest": "76e1d4fda442db7d346a86be561fb3450c3737a710b88e83a49397936f9f4c42"
    },
    {
      "kind": "circuit_plan",
      "id": "verify:4",
      "digest": "0e62ec870943f9188187b1653ec8fe46a728d1e9945c03e81bfb454b48d50f25"
    },
    {
      "kind": "circuit_plan",
      "id": "draft:1",
      "digest": "fc0943bc31fa38686371fd31ab4b4cd86e35d8f7904f92560f612755038a4a57"
    }
  ],
  "provenance": {
    "kind": "preview",
    "engine_commit": "b604fd98d239c96a7b9a8d24f6d57fb7ef378611",
    "planner_sha256": "7868f0ba0b76e82e84446d981859af815def9dad984d67fddfb18086153834d6",
    "planner_built_from": "b604fd98d239c96a7b9a8d24f6d57fb7ef378611",
    "planner_header": {
      "recipe": "qwen3.6/qwen3.6-35b-a3b-fp8-bf16head",
      "checkpoint": "Qwen/Qwen3.6-35B-A3B-FP8",
      "target": "hopper/qwen3.6-35b-a3b/nvfp4",
      "settings": "activation_quantization=adaptive decode_split_silu=on gemv_sw=on kv_cache_dtype=bf16 lm_head_dtype=bf16 rms_norm_act_quant=off row_tiers=canonical ssm_ba_gates_hopper=on ssm_batched_recurrent=on ssm_h_dtype=f32 w4a16_tc=on",
      "opt-in levers": "none"
    },
    "commands": [
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode decode --rows 1",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 2",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 4",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 8",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 12",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 16",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 24",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 32",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 48",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 64",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 96",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode multi_seq --rows 128",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode verify --rows 2",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode verify --rows 3",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode verify --rows 4",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format plan --mode draft --rows 1",
      "met circuit plan --root <tree> --checkpoint qwen3.6/qwen3.6-35b-a3b-fp8-bf16head --hardware h100-sxm --precision recipe --format explain --mode decode --rows 1"
    ],
    "files": [
      {
        "path": "kernels/circuits/INSTANCES.toml",
        "sha256": "2267a2a3ac7f6ec18fcce8d003cc82bfe0a19498b0582cbf55853bf508bf881c"
      },
      {
        "path": "recipes/qwen3.6/qwen3.6-35b-a3b-fp8-bf16head.yaml",
        "sha256": "fdc559439b8588d100bea0de2eeaf29cba75445516e9f619b137cb68a65e207a"
      },
      {
        "path": "kernels/hopper/HARDWARE.toml",
        "sha256": "5b07571d17aacdf8ba0dfc25db38b9365a1dc75d2788326193fd1ec7e9455a2c"
      },
      {
        "path": "kernels/DEVICES.toml",
        "sha256": "256d85840f0e77ec84c5fc739f30aacff6bf24775cc8bef344e36647e51e9c25"
      },
      {
        "path": "kernels/circuits/qwen3_6_moe.toml",
        "sha256": "41ef9de162e4e2b73c77e47bacdfc1c4bb3d9cb0683b255826c89f0002cb8de0"
      },
      {
        "path": "kernels/circuits/blocks/qwen3_hybrid.toml",
        "sha256": "cfe66fc69723840ffe0ff9c5c12b91bb51935aef45898d5ae56567b5f2d38ab6"
      },
      {
        "path": "kernels/gb10/common/KERNEL_FAMILIES.toml",
        "sha256": "fa570a7b011e907c0c05ab03da2388b44e99d02ae50278cd7a98b58a1db3ab40"
      },
      {
        "path": "kernels/circuits/precision/qwen3.6-35b-a3b-fp8-bf16head.toml",
        "sha256": "fc94e079613adb41d7f986f1e4d761a45cb490c87a7697240b56c7c4fd81e276"
      }
    ]
  },
  "settings": {
    "activation_quantization": "adaptive",
    "decode_split_silu": "on",
    "gemv_sw": "on",
    "kv_cache_dtype": "bf16",
    "lm_head_dtype": "bf16",
    "rms_norm_act_quant": "off",
    "row_tiers": "canonical",
    "ssm_ba_gates_hopper": "on",
    "ssm_batched_recurrent": "on",
    "ssm_h_dtype": "f32",
    "w4a16_tc": "on"
  },
  "target": "hopper/qwen3.6-35b-a3b/nvfp4"
}