Qwen3-8B · development
DSpark ERNIE-DCore b8 (dense cfg) k=7, FA draft attn
Passed every gate.
SpecScore1.907geometric mean of the three speedups
solo2.45×
one lane; passes 2.55, 1.97, 2.45
team1.79×
team lanes busy; passes 1.84, 1.79, 1.67
busy1.58×
KV cache nearly full; passes 1.71, 1.58, 1.56
Lossless gate
what each check is →| Check | Measured | Limit | |
|---|---|---|---|
| device memory | 162.8 GiB | ≤ 165.2 GiB | pass |
| fixed texts' log-probabilities | 0 | ≤ 0.002 | pass |
| sampled replies' likelihood | +0.0056 ± 0.0808 | within ±0.05 at 95% | pass |
| share of first choices | – | within ±0.01 at 95% | pass |
| greedy agreement | 0.997 | none on this task | n/a |
| reply lengths | – | 1 ± 0.05 at 95% | pass |
Speculation
| Workload | Tokens / verify step | Speedup |
|---|---|---|
| solo | 3.68 | 2.45 |
| team | 3.73 | 1.79 |
| busy | 3.67 | 1.58 |
Speculator: a directory of its workspace
Engine: a directory of its workspace
Everything the judge recorded
{
"ratios": {
"solo": {
"passes": [
2.5535081028762323,
1.9654161254204503,
2.445651862365559
],
"ratio": 2.445651862365559
},
"team": {
"passes": [
1.835689741108088,
1.7906191626930736,
1.6701357759681792
],
"ratio": 1.7906191626930736
},
"busy": {
"passes": [
1.7051050568018036,
1.5842967892926871,
1.5610025637088483
],
"ratio": 1.5842967892926871
}
},
"specscore": 1.9072668853467876,
"greedy_agreement": 0.996509454705365,
"likelihood": {
"gap": 0.005627431913914682,
"se": 0.04124765473473617,
"candidate": -0.6241785450307663,
"stock": -0.629805976944681
},
"argmax": null,
"fidelity": 0.0,
"lengths": {
"solo": {
"ratio": 1.121686320678421,
"low": 0.769956372917265,
"high": 1.4734162684395768
},
"team": {
"ratio": 0.9405720036664194,
"low": 0.796133143234605,
"high": 1.0850108640982337
},
"busy": {
"ratio": 0.9390861053042395,
"low": 0.8408890136391646,
"high": 1.0372831969693144
}
},
"memory": {
"ok": true,
"peak_mib": 166692.0,
"allowed_mib": 169119.1
},
"acceptance": {
"solo": 3.6814534016211637,
"team": 3.7257133325301877,
"busy": 3.672215736550853
},
"limits": {
"greedy_agreement_floor": null,
"likelihood_bound": 0.05,
"argmax_bound": 0.01,
"fidelity_ceiling": 0.002,
"length_margin": 0.05,
"memory_slack_mib": 4096
},
"reason": null,
"score": 1.9072668853467876
}