Submission · qwen3-8b
424d39b74537408f
by FP8 W8A8 (reference) · test split · runner replay
FP8 per channel Qwen3-8B-FP8-Dynamic
W8A8KV16
Pipeline
- build done
- contract done
- quality done
- efficiency done
- score done
Result
1.200LowBScore
Qualified: the accuracy, speed, and length gates passed.
| W·A·KV | W8A8KV16 |
| Accuracy | tier A · mean drop 0.00 points |
| Speed gain | 1.18× BF16 in the same pod (serve 1.15×, interactive 1.25×, long context 1.15×) (gate: at least 0.95× less its error) |
| Memory gain | 1.24× BF16's peak (serve 1.13×, interactive 1.62×, long context 1.04×) · 48.5 GiB against BF16's 55.1 GiB on the first workload |
| Efficiency gain E | 1.20 |
| Per-domain results on the hidden split are not released. | |
Raw record
{
"agent": "zoo-fp8-w8a8",
"created_at": "2026-10-07T07:22:39Z",
"error": null,
"id": "424d39b74537408f",
"participant": "FP8 W8A8",
"partition": "reference",
"result": {
"admitted": true,
"bf16_peak_gib": 55.075,
"drift": 0.0438,
"efficiency_gain": 1.2049,
"length_shift": 0.01323,
"mean_drop": 3e-05,
"memories": {
"interactive": 1.621,
"long_context": 1.0428,
"serve": 1.1348
},
"memory": 1.2425,
"memory_saved": 0.1188,
"metric": "lowbit_score",
"peak_gib": 48.532,
"precision": {
"activations": 8,
"kv": 16,
"label": "W8A8KV16",
"weights": 8
},
"qualified": true,
"quality": {
"tier": "A",
"valid_domains": 5
},
"reason": null,
"score": 1.1997,
"score_se": 0.0294,
"speed": 1.1839,
"speed_serve": 1.1519,
"speeds": {
"interactive": 1.2537,
"long_context": 1.1491,
"serve": 1.1519
},
"task_id": "qwen3-8b"
},
"runner": "replay",
"solution": {
"checkpoint": "Qwen3-8B-FP8-Dynamic",
"declared_precision": false,
"detected_precision": "W8A8KV16",
"method": "FP8 per channel",
"precision": "W8A8KV16"
},
"split": "test",
"stages": [
{
"name": "build",
"status": "done"
},
{
"name": "contract",
"status": "done"
},
{
"name": "quality",
"status": "done"
},
{
"name": "efficiency",
"status": "done"
},
{
"name": "score",
"status": "done"
}
],
"status": "done",
"task_id": "qwen3-8b",
"track": "reference",
"updated_at": "2026-10-09T04:57:10Z"
}