Submission · qwen35-27b

`bbd77c214f56401a`

by Qwen/Qwen3.5-27B-FP8 (reference) · test split · runner kubernetes

FP8, 128x128 blocks Qwen/Qwen3.5-27B-FP8 W8A8KV16

done

Pipeline

build done
contract done
quality done
efficiency done
score done

Result

1.187LowBScore

Qualified: the accuracy, speed, and length gates passed.

W·A·KV	W8A8KV16
Accuracy	tier A · mean drop 0.21 points
Speed gain	1.03× BF16 in the same pod (serve 0.93×, interactive 1.20×, long context 0.97×) (gate: at least 0.95× less its error)
Memory gain	1.42× BF16's peak (serve 1.41×, interactive 1.69×, long context 1.21×) · 55.2 GiB against BF16's 77.8 GiB on the first workload
Efficiency gain E	1.21
Per-domain results on the hidden split are not released.

2.35 GPU-hours

Raw record

{
  "agent": "hf.Qwen.Qwen3.5-27B-FP8",
  "created_at": "2026-10-07T02:00:48Z",
  "error": null,
  "id": "bbd77c214f56401a",
  "participant": "Qwen/Qwen3.5-27B-FP8",
  "partition": "reference",
  "result": {
    "admitted": true,
    "anchor_efficiency": {
      "serve": 830.8191645139095
    },
    "bf16_peak_gib": 77.756,
    "drift": 0.1826,
    "efficiency": {
      "serve": 769.2191047257604
    },
    "efficiency_gain": 1.2088,
    "gpu_hours": 2.349,
    "length_shift": 0.00097,
    "mean_drop": 0.00208,
    "memories": {
      "interactive": 1.693,
      "long_context": 1.2128,
      "serve": 1.4076
    },
    "memory": 1.4244,
    "memory_gain": 1.4077,
    "memory_saved": 0.2896,
    "metric": "lowbit_score",
    "peak_gib": 55.239,
    "peak_memory": {
      "serve": {
        "device_peak_gib": 161.751,
        "kv_bytes_per_token": 88603,
        "kv_gib": 22.984,
        "kv_usage": 0.1783,
        "other_gib": 4.615,
        "peak_gib": 55.239,
        "weights_gib": 27.64
      }
    },
    "precision": {
      "activations": 8,
      "kv": 16,
      "label": "W8A8KV16",
      "weights": 8,
      "weights_average": 8.17
    },
    "protocol_version": "15",
    "qualified": true,
    "quality": {
      "tier": "A",
      "valid_domains": 5
    },
    "reason": null,
    "score": 1.187,
    "score_se": 0.0231,
    "speed": 1.0269,
    "speed_serve": 0.9259,
    "speeds": {
      "interactive": 1.1996,
      "long_context": 0.9749,
      "serve": 0.9259
    },
    "task_id": "qwen35-27b",
    "timed_audit": null
  },
  "runner": "kubernetes",
  "solution": {
    "checkpoint": "Qwen/Qwen3.5-27B-FP8",
    "declared_precision": false,
    "detected_precision": "W8A8KV16",
    "method": "FP8, 128x128 blocks",
    "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": "qwen35-27b",
  "track": "artifact",
  "updated_at": "2026-10-08T22:33:20Z"
}