Qwen3-8B · development
DSpark b8 ft3 (ft2 + run3 chat-heavy), k=7, FA
Passed every gate.
SpecScore2.122geometric mean of the three speedups
solo2.97×
one lane; passes 2.97, 2.65, 3.07
team1.95×
team lanes busy; passes 1.96, 1.95, 1.87
busy1.65×
KV cache nearly full; passes 1.75, 1.56, 1.65
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 | -0.0471 ± 0.0829 | within ±0.02 at 95% | pass |
| share of first choices | +0.0071 ± 0.0152 | within ±0.003 at 95% | pass |
| greedy agreement | 0.982 | ≥ 0.5 | pass |
| reply lengths | 0.989 [0.900, 1.077] | 1 ± 0.05 at 95% | pass |
Speculation
| Workload | Tokens / verify step | Speedup |
|---|---|---|
| solo | 4.55 | 2.97 |
| team | 4.28 | 1.95 |
| busy | 4.23 | 1.65 |
Speculator: a directory of its workspace
Engine: a directory of its workspace
Everything the judge recorded
{
"ratios": {
"solo": {
"passes": [
2.971399812218224,
2.6500367074333013,
3.066804999485311
],
"ratio": 2.971399812218224
},
"team": {
"passes": [
1.9631016810871664,
1.9525548737390028,
1.8747251124833777
],
"ratio": 1.9525548737390028
},
"busy": {
"passes": [
1.747368729826642,
1.5623950040737504,
1.6466529555425884
],
"ratio": 1.6466529555425884
}
},
"specscore": 2.121886448478702,
"greedy_agreement": 0.9818715936945347,
"likelihood": {
"gap": -0.0470792165120244,
"se": 0.04228254298312121,
"candidate": -0.666556876994703,
"stock": -0.6194776604826786
},
"argmax": {
"gap": 0.0071468948478320415,
"se": 0.007747203941236115,
"candidate": 0.1406758018306096,
"stock": 0.13352890698277756
},
"fidelity": 0.0,
"lengths": {
"solo": {
"ratio": 0.9821225682287197,
"low": 0.6375404548986567,
"high": 1.3267046815587826,
"paired": true,
"replies": 245
},
"team": {
"ratio": 1.0657714669608274,
"low": 0.9094146595571372,
"high": 1.2221282743645177,
"paired": true,
"replies": 1115
},
"busy": {
"ratio": 0.9480510225896102,
"low": 0.8361803440552602,
"high": 1.0599217011239601,
"paired": true,
"replies": 1895
},
"pooled": {
"ratio": 0.9886371032816339,
"low": 0.900365241663964,
"high": 1.0769089648993038,
"paired": true,
"replies": 3255
}
},
"memory": {
"ok": true,
"peak_mib": 166428.0,
"allowed_mib": 169119.1
},
"acceptance": {
"solo": 4.549912903658046,
"team": 4.283411012927853,
"busy": 4.226696396146622
},
"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": 2.121886448478702
}