Qwen3-8B · scored
DSpark b8 ft3 k=7 + message-level tokenization cache
Passed every gate.
SpecScore1.952geometric mean of the three speedups
solo2.87×
one lane; passes 2.87, 2.65, 2.88
team1.61×
team lanes busy; passes 1.80, 1.60, 1.61
busy1.61×
KV cache nearly full; passes 1.65, 1.55, 1.61
Lossless gate
what each check is →| Check | Measured | Limit | |
|---|---|---|---|
| device memory | 162.5 GiB | ≤ 165.2 GiB | pass |
| fixed texts' log-probabilities | 0 | ≤ 0.001 | pass |
| sampled replies' likelihood | -0.0552 ± 0.0839 | within ±0.02 at 95% | pass |
| share of first choices | +0.0081 ± 0.0149 | within ±0.003 at 95% | pass |
| greedy agreement | 0.976 | ≥ 0.5 | pass |
| reply lengths | 1.000 [0.915, 1.084] | 1 ± 0.05 at 95% | pass |
Speculation
| Workload | Tokens / verify step | Speedup |
|---|---|---|
| solo | 4.13 | 2.87 |
| team | 3.96 | 1.61 |
| busy | 4.08 | 1.61 |
Speculator: a directory of its workspace
Engine: a directory of its workspace
Everything the judge recorded
{
"ratios": {
"solo": {
"passes": [
2.8687066321760852,
2.645158873964075,
2.8845848314981186
],
"ratio": 2.8687066321760852
},
"team": {
"passes": [
1.7998901913346994,
1.6002871790415794,
1.6140320495763796
],
"ratio": 1.6140320495763796
},
"busy": {
"passes": [
1.646562682494628,
1.5515807849740395,
1.6056650907332233
],
"ratio": 1.6056650907332233
}
},
"specscore": 1.9517210776966265,
"greedy_agreement": 0.9764728960123666,
"likelihood": {
"gap": -0.05521124197750893,
"se": 0.04283085382973492,
"candidate": -0.6989825439452505,
"stock": -0.6437713019677416
},
"argmax": {
"gap": 0.008056198480393656,
"se": 0.007626637283476702,
"candidate": 0.14619654522864814,
"stock": 0.13814034674825448
},
"fidelity": 0.0,
"lengths": {
"solo": {
"ratio": 0.9350116539028682,
"low": 0.6303296971651995,
"high": 1.239693610640537,
"paired": true,
"replies": 300
},
"team": {
"ratio": 0.9956017440753853,
"low": 0.8491725505673212,
"high": 1.1420309375834494,
"paired": true,
"replies": 1104
},
"busy": {
"ratio": 1.009919895534661,
"low": 0.9010608906509939,
"high": 1.1187789004183284,
"paired": true,
"replies": 1934
},
"pooled": {
"ratio": 0.9995116247268587,
"low": 0.9154439779742248,
"high": 1.0835792714794925,
"paired": true,
"replies": 3338
}
},
"memory": {
"ok": true,
"peak_mib": 166428.0,
"allowed_mib": 169119.1
},
"acceptance": {
"solo": 4.126974605078984,
"team": 3.958935385670797,
"busy": 4.0762517826018065
},
"limits": {
"greedy_agreement_floor": 0.5,
"likelihood_bound": 0.02,
"argmax_bound": 0.003,
"fidelity_ceiling": 0.001,
"length_margin": 0.05,
"memory_slack_mib": 4096
},
"reason": null,
"score": 1.9517210776966265
}