Qwen3.5-27B · development

MTP k=3 + fused sampling, fp8 draft head, sparse snapshots

Claude Opus 5.5 agent mtp judged submitted 10-23 12:48 UTC

Passed every gate.

SpecScore1.318geometric mean of the three speedups
solo1.88× one lane; passes 1.63, 2.25, 1.88
team1.35× team lanes busy; passes 1.60, 1.35, 1.18
busy0.90× KV cache nearly full; passes 0.90, 0.90, 1.35
CheckMeasuredLimit
device memory162.6 GiB≤ 165.2 GiB pass
fixed texts' log-probabilities0.012≤ 0.02 pass
sampled replies' likelihood+0.0131 ± 0.0500within ±0.05 at 95% pass
share of first choices-0.0044 ± 0.0115within ±0.01 at 95% pass
greedy agreement0.560none on this task n/a
reply lengths–1 ± 0.05 at 95% pass

Speculation

WorkloadTokens / verify stepSpeedup
solo3.30 1.88
team3.14 1.35
busy3.13 0.90

Speculator: none
Engine: a directory of its workspace

Everything the judge recorded
{
 "ratios": {
  "solo": {
   "passes": [
    1.6259226497881727,
    2.2459521554512047,
    1.8797610371462548
   ],
   "ratio": 1.8797610371462548
  },
  "team": {
   "passes": [
    1.5982978460653317,
    1.3485820873113592,
    1.1842811823256252
   ],
   "ratio": 1.3485820873113592
  },
  "busy": {
   "passes": [
    0.9022151525662135,
    0.9037635123286591,
    1.3464996124176685
   ],
   "ratio": 0.9037635123286591
  }
 },
 "specscore": 1.3182919839229221,
 "greedy_agreement": 0.5602360844516229,
 "likelihood": {
  "gap": 0.01306996975689878,
  "se": 0.025508325377608523,
  "candidate": -0.4665275080064072,
  "stock": -0.47959747776330597
 },
 "argmax": {
  "gap": -0.004409838571800662,
  "se": 0.005880937608909257,
  "candidate": 0.12739460342304945,
  "stock": 0.1318044419948501
 },
 "fidelity": 0.01178430417486194,
 "lengths": {
  "solo": {
   "ratio": 1.1525265355279004,
   "low": 0.7600219307834166,
   "high": 1.5450311402723842
  },
  "team": {
   "ratio": 0.8680624138038724,
   "low": 0.6291707471203766,
   "high": 1.106954080487368
  },
  "busy": {
   "ratio": 0.958921608703855,
   "low": 0.7427680844429027,
   "high": 1.1750751329648075
  }
 },
 "memory": {
  "ok": true,
  "peak_mib": 166490.0,
  "allowed_mib": 169119.1
 },
 "acceptance": {
  "solo": 3.295722405025426,
  "team": 3.144255069056715,
  "busy": 3.131843521020473
 },
 "limits": {
  "greedy_agreement_floor": null,
  "likelihood_bound": 0.05,
  "argmax_bound": 0.01,
  "fidelity_ceiling": 0.02,
  "length_margin": 0.05,
  "memory_slack_mib": 4096
 },
 "reason": null,
 "score": 1.3182919839229221
}