Submission · qwen35-27b
871c569e02094a38
by NVFP4 W4A4 + FP8 KV (reference) · test split · runner replay
NVFP4 + FP8 KV Qwen3.5-27B-NVFP4
W4A4KV8
Pipeline
- build done
- contract done
- quality done
- efficiency done
- score done
Result
1.594LowBScore
Qualified: the accuracy, speed, and length gates passed.
| W·A·KV | W4A4KV8 |
| Accuracy | tier A · mean drop 0.62 points |
| Speed gain | 1.61× BF16 in the same pod (serve 1.66×, interactive 1.44×, long context 1.75×) (gate: at least 0.95× less its error) |
| Memory gain | 1.83× BF16's peak (serve 1.81×, interactive 1.88×, long context 1.78×) · 42.2 GiB against BF16's 76.6 GiB on the first workload |
| Efficiency gain E | 1.71 |
| Per-domain results on the hidden split are not released. | |
Raw record
{
"agent": "zoo-nvfp4-w4a4-kvfp8",
"created_at": "2026-10-07T07:22:39Z",
"error": null,
"id": "871c569e02094a38",
"participant": "NVFP4 W4A4 + FP8 KV",
"partition": "reference",
"result": {
"admitted": true,
"bf16_peak_gib": 76.59,
"drift": 0.6729,
"efficiency_gain": 1.7054,
"length_shift": 0.01022,
"mean_drop": 0.00623,
"memories": {
"interactive": 1.8816,
"long_context": 1.7827,
"serve": 1.8142
},
"memory": 1.8257,
"memory_saved": 0.4488,
"metric": "lowbit_score",
"peak_gib": 42.216,
"precision": {
"activations": 4,
"kv": 8,
"label": "W4A4KV8",
"weights": 4
},
"qualified": true,
"quality": {
"tier": "A",
"valid_domains": 5
},
"reason": null,
"score": 1.5944,
"score_se": 0.0365,
"speed": 1.6093,
"speed_serve": 1.6553,
"speeds": {
"interactive": 1.4406,
"long_context": 1.7479,
"serve": 1.6553
},
"task_id": "qwen35-27b"
},
"runner": "replay",
"solution": {
"checkpoint": "Qwen3.5-27B-NVFP4",
"declared_precision": false,
"detected_precision": "W4A4KV8",
"method": "NVFP4 + FP8 KV",
"precision": "W4A4KV8"
},
"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": "qwen35-27b",
"track": "reference",
"updated_at": "2026-10-09T04:57:10Z"
}