File size: 1,308 Bytes
2293023
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
{
  "arch_id": "qwen3-next-mtp",
  "artifact_role": "optimized-quality",
  "benchmark_summary": {
    "acceptance_by_depth": [
      0.9556737588652481,
      0.852575488454707,
      0.7406749555950266
    ],
    "artifact": "/tmp/mtplx-flat8-maxfan-bakeoff-20260510-040405/flat8-int8-flappy2k.json",
    "peak_gib": 27.615886019542813,
    "suite": "flappy 2k depth3 max-fan",
    "tok_s": 33.6276837588148,
    "verify_ms_per_call": 88.13897531140259
  },
  "exactness_baseline": {
    "gate": "flat8-quality-mtp-bakeoff",
    "notes": "Validated with mtplx inspect and local Flappy 2k max-fan bakeoff; 2k Flappy validator truncates before balanced delimiter completion, so code-shape validator is the meaningful capped-run gate.",
    "status": "candidate-promoted-by-user-decision"
  },
  "mtp_depth_max": 3,
  "mtp_sidecar": "CyanKiwi calibrated INT8 affine group128, BF16 aux preserved",
  "mtplx_version": "0.1.0-preview",
  "recommended_profile": "performance-cold",
  "sampler": {
    "temperature": 0.6,
    "top_k": 20,
    "top_p": 0.95
  },
  "target_precision": "flat8 affine group64",
  "verified_on": {
    "hardware": "M5 Max local max-fan bakeoff",
    "model": "Qwen3.6-27B-MTPLX-Optimized-Quality",
    "status": "validated-candidate",
    "timestamp": "2026-05-10T04:16:00+01:00"
  }
}