Submission · qwen3-8b

`fee712e078904618`

by Unsloth NF4 (reference) · test split · runner kubernetes

NF4, Unsloth unsloth/Qwen3-8B-bnb-4bit W4A16KV16

done

Pipeline

build done
contract done
quality done
efficiency done
score done

Result

pending LowBScore

accuracy gate: the worst domain's 95% upper bound on the drop is 7.02 points, over 5; speed or memory on the other workloads not measured

W·A·KV	W4A16KV16 declared · the service reads W4A16KV16
Accuracy	fail · mean drop 2.14 points
Drift from BF16	KL 0.036 nats per token
Speed gain	0.63× BF16 in the same pod (serve 0.63×) (gate: at least 0.95× less its error)
Memory gain	— BF16's peak (serve 1.21×) · 46.0 GiB against BF16's 55.6 GiB on the first workload
Per-domain results on the hidden split are not released.

0.54 GPU-hours

Raw record

{
  "agent": "zoo-pub-unsloth-nf4",
  "created_at": "2026-10-08T05:59:14Z",
  "error": null,
  "id": "fee712e078904618",
  "participant": "Unsloth NF4",
  "partition": "reference",
  "result": {
    "admitted": false,
    "anchor_efficiency": {
      "serve": 1978.854511081479
    },
    "bf16_peak_gib": 55.624,
    "efficiency": {
      "serve": 1246.0105004107304
    },
    "gpu_hours": 0.536,
    "kl": {
      "kl": 0.03582149655823167,
      "kl_median": 0.031063964891860682,
      "kl_per_token": 0.03560865118834652,
      "kl_se": 0.0016561104713529225,
      "sequences": 256,
      "tokens": 62197
    },
    "mean_drop": 0.02142,
    "memories": {
      "interactive": null,
      "long_context": null,
      "serve": 1.2105
    },
    "memory": null,
    "memory_saved": 0.1739,
    "metric": "lowbit_score",
    "peak_gib": 45.953,
    "peak_memory": {
      "serve": {
        "device_peak_gib": 163.898,
        "kv_bytes_per_token": 147294,
        "kv_gib": 38.208,
        "kv_usage": 0.2491,
        "other_gib": 1.775,
        "peak_gib": 45.953,
        "weights_gib": 5.97
      }
    },
    "precision": {
      "activations": 16,
      "kv": 16,
      "label": "W4A16KV16",
      "weights": 4,
      "weights_average": 4.0
    },
    "qualified": false,
    "quality": {
      "tier": "fail",
      "valid_domains": 5
    },
    "reason": "accuracy gate: the worst domain\u0027s 95% upper bound on the drop is 7.02 points, over 5; speed or memory on the other workloads not measured",
    "score": null,
    "score_se": null,
    "speed": null,
    "speed_serve": 0.6297,
    "speeds": {
      "interactive": null,
      "long_context": null,
      "serve": 0.6297
    },
    "task_id": "qwen3-8b",
    "timed_audit": null
  },
  "runner": "kubernetes",
  "solution": {
    "checkpoint": "unsloth/Qwen3-8B-bnb-4bit",
    "declared_precision": true,
    "detected_precision": "W4A16KV16",
    "method": "NF4, Unsloth",
    "precision": "W4A16KV16"
  },
  "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": "artifact",
  "updated_at": "2026-10-08T22:33:20Z"
}