Qwen3-8B · development
DSpark b8 ft2 (ft1 + run2 self-distill), k=7, FA
Passed every gate.
SpecScore2.071geometric mean of the three speedups
solo2.73×
one lane; passes 2.82, 2.73, 2.38
team1.94×
team lanes busy; passes 1.86, 1.97, 1.94
busy1.68×
KV cache nearly full; passes 1.61, 1.68, 1.73
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 | -0.0442 ± 0.0822 | within ±0.02 at 95% | pass |
| share of first choices | +0.0055 ± 0.0153 | within ±0.003 at 95% | pass |
| greedy agreement | 0.996 | ≥ 0.5 | pass |
| reply lengths | 1.058 [0.966, 1.150] | 1 ± 0.05 at 95% | pass |
Speculation
| Workload | Tokens / verify step | Speedup |
|---|---|---|
| solo | 4.19 | 2.73 |
| team | 4.20 | 1.94 |
| busy | 4.34 | 1.68 |
Speculator: a directory of its workspace
Engine: a directory of its workspace
Everything the judge recorded
{
"ratios": {
"solo": {
"passes": [
2.8240053608654043,
2.727968910559463,
2.3771374070635387
],
"ratio": 2.727968910559463
},
"team": {
"passes": [
1.8619811529185843,
1.966516712732128,
1.9399263147654409
],
"ratio": 1.9399263147654409
},
"busy": {
"passes": [
1.6113085932011413,
1.678673793290558,
1.727718287936316
],
"ratio": 1.678673793290558
}
},
"specscore": 2.071080535465982,
"greedy_agreement": 0.9961477272727273,
"likelihood": {
"gap": -0.044225745478564615,
"se": 0.0419174549821264,
"candidate": -0.6641571470941986,
"stock": -0.619931401615634
},
"argmax": {
"gap": 0.005542405434844827,
"se": 0.00778507507444489,
"candidate": 0.13978203901667985,
"stock": 0.13423963358183502
},
"fidelity": 0.0,
"lengths": {
"solo": {
"ratio": 1.0239878185970932,
"low": 0.711721100174213,
"high": 1.3362545370199734,
"paired": true,
"replies": 266
},
"team": {
"ratio": 1.1122862697762883,
"low": 0.9472976459325513,
"high": 1.2772748936200253,
"paired": true,
"replies": 1172
},
"busy": {
"ratio": 1.0335241572436282,
"low": 0.9155253597261215,
"high": 1.151522954761135,
"paired": true,
"replies": 1930
},
"pooled": {
"ratio": 1.0577582821647866,
"low": 0.965923106511207,
"high": 1.149593457818366,
"paired": true,
"replies": 3368
}
},
"memory": {
"ok": true,
"peak_mib": 166430.0,
"allowed_mib": 169119.1
},
"acceptance": {
"solo": 4.1905526449214126,
"team": 4.200886427716586,
"busy": 4.343736311870346
},
"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": 2.071080535465982
}