{
  "arch_id": "qwen3-next-mtp",
  "artifact_role": "forge-local",
  "base_trunk": "/Users/youssof/.mtplx/models/Qwen--Qwen3.8-27B",
  "exactness_baseline": {},
  "forge_provenance": {
    "forge_inputs": {
      "mtp_source_path": "/Users/youssof/.mtplx/models/Qwen--Qwen3.8-27B",
      "trunk_path": "/Users/youssof/.mtplx/models/Qwen3.8-27B-MTPLX-Optimized-Speed"
    },
    "forge_recipe": {
      "body_bits": 4,
      "body_group_size": 32,
      "body_mode": "affine",
      "module_overrides": [
        {
          "bits": 8,
          "group_size": 64,
          "suffix": "embed_tokens"
        },
        {
          "bits": 8,
          "group_size": 64,
          "suffix": "lm_head"
        },
        {
          "bits": 8,
          "group_size": 64,
          "suffix": "linear_attn.out_proj"
        },
        {
          "bits": 8,
          "group_size": 64,
          "layers": [
            56,
            57,
            58,
            59,
            60,
            61,
            62,
            63
          ],
          "suffix": "mlp.gate_proj"
        },
        {
          "bits": 8,
          "group_size": 64,
          "layers": [
            56,
            57,
            58,
            59,
            60,
            61,
            62,
            63
          ],
          "suffix": "mlp.up_proj"
        },
        {
          "bits": 8,
          "group_size": 64,
          "layers": [
            56,
            57,
            58,
            59,
            60,
            61,
            62,
            63
          ],
          "suffix": "mlp.down_proj"
        }
      ],
      "mtp_policy": "keep_bf16"
    },
    "forged_at": "2026-08-14T11:28:40-07:00",
    "forged_locally": true,
    "mtp_contract": {
      "base_hidden_variant": "post_norm",
      "concat_order": "embedding_hidden",
      "hidden_variant": "post_norm",
      "mtp_position_mode": "local",
      "mtp_quant_group_size": 64,
      "mtp_quant_mode": "affine"
    },
    "mtplx_version": "2.6.0",
    "published_to_hf": null,
    "recommended_profile_note": "Restamped sustained->turbo 2026-08-14 23:50 (mirrors the OQ restamp): the sustained recommendation came from the forge-verify rows; every gated dropday arm, the CLI turbo-default resolution for this public id, and the installed-app receipt (55.5 tok/s, draft 1.0) all served turbo. Served behavior unchanged: under turbo the hidden profile-scoped contract fell back to draft=target sampler (1.0/0.95/20), identical to the recommended_draft_sampler stamp now visible.",
    "source_format": "bf16_native",
    "source_repo": "/Users/youssof/.mtplx/models/Qwen--Qwen3.8-27B",
    "source_sha": null,
    "head_quantization": {
      "bits": 4,
      "group_size": 64,
      "mode": "affine",
      "policy": "all",
      "quantized_at": "2026-08-20T04:49:57Z",
      "source_sidecar_bytes": 849400388,
      "quantized_sidecar_bytes": 238934226,
      "tool": "scripts/build_qwen38_q4head_sidecar.py",
      "note": "Structural head quantization of the released sidecar; no calibration, no training. Trunk weights unchanged."
    }
  },
  "mtp_contract": {
    "base_hidden_variant": "post_norm",
    "concat_order": "embedding_hidden",
    "hidden_variant": "post_norm",
    "mtp_position_mode": "local",
    "mtp_quant_group_size": 64,
    "mtp_quant_mode": "affine"
  },
  "mtp_depth_default": 3,
  "mtp_depth_max": 3,
  "mtp_sidecar": "int4-g64-prequantized",
  "mtplx_version": "2.9.0",
  "precision_policy": {
    "intended_default_for": [
      "m1",
      "m2"
    ],
    "note": "FP16 precision sibling of Qwen3.8-27B-MTPLX-Optimized-Speed: byte-identical INT packs, bf16 floats cast to fp16 for M1/M2 Macs. Same tuned depth/draft-sampler stamps as the parent; not a separate speed claim.",
    "routing": "mtplx start auto-selects this artifact on M1/M2 Apple Silicon",
    "variant": "fp16"
  },
  "precision_variant": "fp16",
  "public_model_id": "mtplx-qwen38-27b-optimized-speed-fp16",
  "recommended_draft_sampler": {
    "temperature": 1.0,
    "top_k": 20,
    "top_p": 0.95
  },
  "recommended_profile": "turbo",
  "release_validation": {
    "dropday_draft_sampler_abba": {
      "basis": "dropday uncapped ABBA campaign (clean-receipts.jsonl, temp 1.0 xhigh/medium, die-gated, max fans verified)",
      "decode_tps_blended_draft_0_6": [
        30.4,
        33.6
      ],
      "decode_tps_blended_draft_1_0": [
        35.1,
        37.3
      ],
      "stamped_at": "2026-08-14T21:45:00-07:00",
      "verdict": "draft_temperature_1_0 (+13% blended)",
      "workload": "simple flappy xhigh uncapped (20k-44k tokens, EOS stop)"
    }
  },
  "sampler": {
    "temperature": 0.6,
    "top_k": 20,
    "top_p": 0.95
  },
  "speed_evidence": {
    "depth": 3,
    "tok_s": [
      45.38375130593394
    ],
    "acceptance_by_depth": [
      0.9384057971014492,
      0.855072463768116,
      0.7427536231884058
    ],
    "greedy_diagnostic": {
      "tok_s": 19.60451265560769
    },
    "forge_verify_rows": [
      {
        "acceptance_by_position": [],
        "depth": 0,
        "finish_reasons": {
          "stop": 1
        },
        "hit_token_budget": false,
        "hit_token_budget_count": 0,
        "multiplier_vs_ar": 1.0,
        "quality_passed": true,
        "tok_s": 19.60451265560769,
        "verify_time_s": 30.370817083050497
      },
      {
        "acceptance_by_position": [
          0.9712460063897763
        ],
        "depth": 1,
        "finish_reasons": {
          "stop": 1
        },
        "hit_token_budget": false,
        "hit_token_budget_count": 0,
        "multiplier_vs_ar": 1.5051947774660717,
        "quality_passed": true,
        "tok_s": 29.5086100639882,
        "verify_time_s": 20.34781050006859
      },
      {
        "acceptance_by_position": [
          0.9889624724061811,
          0.9448123620309051
        ],
        "depth": 2,
        "finish_reasons": {
          "stop": 1
        },
        "hit_token_budget": false,
        "hit_token_budget_count": 0,
        "multiplier_vs_ar": 1.976632257031939,
        "quality_passed": true,
        "tok_s": 38.75091209846504,
        "verify_time_s": 34.64071964554023
      },
      {
        "acceptance_by_position": [
          0.9384057971014492,
          0.855072463768116,
          0.7427536231884058
        ],
        "depth": 3,
        "finish_reasons": {
          "stop": 1
        },
        "hit_token_budget": false,
        "hit_token_budget_count": 0,
        "multiplier_vs_ar": 2.3149645238925305,
        "quality_passed": true,
        "tok_s": 45.38375130593394,
        "verify_time_s": 21.260114512755536
      }
    ],
    "quality_rejected": [],
    "acceptance_collapsed": [],
    "failure_reasons": [],
    "verdict": "mtp_depth_wins",
    "artifact_fingerprint": "sha256:069c2de291fd15b130383119b13f60c45e0f78481a180cf51887ae87c3986b12"
  },
  "verified_on": {
    "timestamp": "2026-08-19T23:10:32-07:00",
    "hardware": "macOS-26.3.1-arm64-arm-64bit-Mach-O",
    "machine_arch": "arm64",
    "macos": "26.3.1",
    "model": "Qwen3.8-27B-MTPLX-Optimized-Speed-FP16"
  }
}