Submission · qwen35-27b

`5b503eb089664578`

by NVFP4 on the agent's coverage (reference) · test split · runner kubernetes

NVFP4 on the agent's coverage Qwen3.5-27B-NVFP4-GDN-calKV W4A4KV8

done

Pipeline

build done
contract done
quality done
efficiency done
score done

Result

0.148LowBScore

accuracy gate: the worst domain's 95% upper bound on the drop is 6.74 points, over 5

Entered as an engine that built its own checkpoint before the task fixed one; scored as the stock engine with the stock flags serves that checkpoint.

W·A·KV	W4A4KV8 declared · the service reads W4A4KV8
Accuracy	fail · mean drop 3.16 points
Drift from BF16	KL 0.085 nats per token · generated length +1.4% · D = 3.26 points
Speed gain	1.93× BF16 in the same pod (serve 1.99×, interactive 1.66×, long context 2.18×) (gate: at least 0.95× less its error)
Memory gain	2.20× BF16's peak (serve 2.23×, interactive 2.43×, long context 1.96×) · 34.8 GiB against BF16's 77.8 GiB on the first workload
Efficiency gain E	2.05
Per-domain results on the hidden split are not released.

1.16 GPU-hours

Raw record

{
  "agent": "zoo-agentcov-nvfp4",
  "created_at": "2026-10-07T22:45:06Z",
  "error": null,
  "id": "5b503eb089664578",
  "participant": "NVFP4 on the agent\u0027s coverage",
  "partition": "reference",
  "result": {
    "admitted": false,
    "anchor_efficiency": {
      "serve": 841.73351775042
    },
    "bf16_peak_gib": 77.756,
    "drift": 3.2611,
    "efficiency": {
      "serve": 1653.2572378628213
    },
    "efficiency_gain": 2.0458,
    "gpu_hours": 1.161,
    "kl": {
      "kl": 0.08474156537564312,
      "kl_median": 0.08156433427859498,
      "kl_per_token": 0.08461917847900788,
      "kl_se": 0.0029504475451079333,
      "sequences": 256,
      "tokens": 63668
    },
    "length_shift": 0.01442,
    "mean_drop": 0.03164,
    "memories": {
      "interactive": 2.4271,
      "long_context": 1.963,
      "serve": 2.2321
    },
    "memory": 2.1991,
    "memory_saved": 0.552,
    "metric": "lowbit_score",
    "peak_gib": 34.836,
    "peak_memory": {
      "serve": {
        "device_peak_gib": 163.359,
        "kv_bytes_per_token": 53164,
        "kv_gib": 13.791,
        "kv_usage": 0.0984,
        "other_gib": 3.125,
        "peak_gib": 34.836,
        "weights_gib": 17.92
      }
    },
    "precision": {
      "activations": 4,
      "kv": 8,
      "label": "W4A4KV8",
      "weights": 4,
      "weights_average": 4.65
    },
    "qualified": false,
    "quality": {
      "tier": "fail",
      "valid_domains": 5
    },
    "reason": "accuracy gate: the worst domain\u0027s 95% upper bound on the drop is 6.74 points, over 5",
    "score": 0.1477,
    "score_se": 0.0043,
    "served_as": "checkpoint",
    "speed": 1.9307,
    "speed_serve": 1.9882,
    "speeds": {
      "interactive": 1.6642,
      "long_context": 2.1751,
      "serve": 1.9882
    },
    "task_id": "qwen35-27b",
    "timed_audit": null
  },
  "runner": "kubernetes",
  "solution": {
    "checkpoint": "Qwen3.5-27B-NVFP4-GDN-calKV",
    "declared_precision": true,
    "detected_precision": "W4A4KV8",
    "method": "NVFP4 on the agent\u0027s coverage",
    "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": "artifact",
  "updated_at": "2026-10-08T22:40:30Z"
}