Qwen3-8B · development
DSpark ERNIE b8 fine-tuned on target replies, k=7
Passed every gate.
SpecScore2.029geometric mean of the three speedups
solo2.70×
one lane; passes 2.65, 2.70, 2.90
team1.84×
team lanes busy; passes 1.84, 1.74, 1.89
busy1.69×
KV cache nearly full; passes 1.69, 1.53, 1.70
Lossless gate
what each check is →| Check | Measured | Limit | |
|---|---|---|---|
| device memory | 162.5 GiB | ≤ 165.2 GiB | pass |
| fixed texts' log-probabilities | 0 | ≤ 0.002 | pass |
| sampled replies' likelihood | +0.0397 ± 0.0895 | within ±0.05 at 95% | pass |
| share of first choices | -0.0109 ± 0.0165 | within ±0.01 at 95% | pass |
| greedy agreement | 1.000 | none on this task | n/a |
| reply lengths | – | 1 ± 0.05 at 95% | pass |
Speculation
| Workload | Tokens / verify step | Speedup |
|---|---|---|
| solo | 4.26 | 2.70 |
| team | 4.22 | 1.84 |
| busy | 4.14 | 1.69 |
Speculator: a directory of its workspace
Engine: a directory of its workspace
Everything the judge recorded
{
"ratios": {
"solo": {
"passes": [
2.6487824680151304,
2.6963078384143935,
2.901212274236556
],
"ratio": 2.6963078384143935
},
"team": {
"passes": [
1.83515481100013,
1.738231269906612,
1.8878673175107075
],
"ratio": 1.83515481100013
},
"busy": {
"passes": [
1.6892475257641855,
1.527788383305857,
1.7042368439276694
],
"ratio": 1.6892475257641855
}
},
"specscore": 2.0294506299005044,
"greedy_agreement": 1.0,
"likelihood": {
"gap": 0.039693153544942406,
"se": 0.045655831540944734,
"candidate": -0.6351922517678941,
"stock": -0.6748854053128365
},
"argmax": {
"gap": -0.01094099291389719,
"se": 0.008436217367788285,
"candidate": 0.13396954750160842,
"stock": 0.1449105404155056
},
"fidelity": 0.0,
"lengths": {
"solo": {
"ratio": 0.9369137400340901,
"low": 0.6640412281351796,
"high": 1.2097862519330007
},
"team": {
"ratio": 0.9789848344908098,
"low": 0.8268363939422396,
"high": 1.1311332750393799
},
"busy": {
"ratio": 0.9500869057567334,
"low": 0.850067378589731,
"high": 1.0501064329237355
}
},
"memory": {
"ok": true,
"peak_mib": 166430.0,
"allowed_mib": 169119.1
},
"acceptance": {
"solo": 4.259804693652543,
"team": 4.220432463631797,
"busy": 4.138552013059812
},
"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": 2.0294506299005044
}