Submission · qwen3-8b
eff88ec942264431
by Own engine: FP8 + FP8 KV, 16K prefill chunks (reference) · test split · runner kubernetes
FP8-chunked-prefill Qwen3-8B-FP8-Dynamic
W8A8KV8
Pipeline
- build done
- contract done
- quality done
- efficiency done
- score done
Result
1.456LowBScore
Qualified: the accuracy, speed, and length gates passed.
| W·A·KV | W8A8KV8 declared · the service reads W8A8KV8 |
| Accuracy | tier A · mean drop 0.59 points |
| Drift from BF16 | KL 0.010 nats per token · generated length +1.8% · D = 0.55 points |
| Speed gain | 1.35× BF16 in the same pod (serve 1.44×, interactive 1.23×, long context 1.40×) (gate: at least 0.95× less its error) |
| Memory gain | 1.78× BF16's peak (serve 1.85×, interactive 1.58×, long context 1.94×) · 30.1 GiB against BF16's 55.6 GiB on the first workload |
| Efficiency gain E | 1.54 |
| Per-domain results on the hidden split are not released. | |
Engine: 8.8 GiB of weights held, 8.8 GiB stored.
0.78 GPU-hours
Raw record
{
"agent": "zoo-engine-fp8-chunked",
"created_at": "2026-10-11T20:06:33Z",
"error": null,
"id": "eff88ec942264431",
"participant": "Own engine: FP8 + FP8 KV, 16K prefill chunks",
"partition": "reference",
"result": {
"admitted": true,
"anchor_efficiency": {
"interactive": 246.32308683114326,
"long_context": 730.9919209844271,
"serve": 1937.1458343328025
},
"anchor_peak_memory": {
"interactive": {
"device_peak_gib": 163.132,
"kv_bytes_per_token": 147464,
"kv_gib": 0.176,
"kv_usage": 0.0012,
"other_gib": 2.145,
"peak_gib": 17.591,
"weights_gib": 15.27
},
"long_context": {
"device_peak_gib": 163.132,
"kv_bytes_per_token": 146884,
"kv_gib": 142.32,
"kv_usage": 0.9901,
"other_gib": 2.145,
"peak_gib": 159.736,
"weights_gib": 15.27
},
"serve": {
"device_peak_gib": 163.132,
"kv_bytes_per_token": 147210,
"kv_gib": 38.186,
"kv_usage": 0.2657,
"other_gib": 2.145,
"peak_gib": 55.602,
"weights_gib": 15.27
}
},
"bf16_peak_gib": 55.602,
"drift": 0.5509,
"efficiency": {
"interactive": 302.5793654997613,
"long_context": 1020.5239679733002,
"serve": 2783.0896029523274
},
"efficiency_gain": 1.5388,
"engine": {
"audit": {
"interactive": {
"bf16_mean": -0.13170158582772096,
"bf16_top1": 0.9895301533100402,
"empty": 1,
"empty_alone": 0,
"empty_reference": 1,
"engine_mean": -0.33488874823383913,
"engine_top1": 0.8934413514207463,
"gap": 0.20318716240611817,
"gap_max": 2.259063767962636,
"gap_median": 0.12190573442048194,
"length_ratio": 1.0,
"length_ratio_min": 0.0,
"picked": 24,
"run": "interactive-r0.json",
"samples": 23,
"status": "audited",
"top1_gap": 0.09608880188929392
},
"long_context": {
"bf16_mean": -0.3193891973994322,
"bf16_top1": 0.9852744424349366,
"empty": 0,
"empty_alone": 0,
"empty_reference": 0,
"engine_mean": -0.22300654186599272,
"engine_top1": 0.9493727651074984,
"gap": -0.09638265553343953,
"gap_max": 0.35711406119412015,
"gap_median": 0.04173304980499706,
"length_ratio": 1.0,
"length_ratio_min": 0.5078125,
"picked": 24,
"run": "long_context-r0.json",
"samples": 24,
"status": "audited",
"top1_gap": 0.03590167732743821
},
"memory_capped": {
"status": "missing"
},
"offline": {
"status": "missing"
},
"online_slo": {
"status": "missing"
},
"serve": {
"bf16_mean": -0.06854064463553336,
"bf16_top1": 0.9947859789301092,
"empty": 0,
"empty_alone": 0,
"empty_reference": 0,
"engine_mean": -0.09481864496485903,
"engine_top1": 0.9783919085971547,
"gap": 0.02627800032932568,
"gap_max": 0.3413345401085057,
"gap_median": 0.021810810235200748,
"length_ratio": 1.0,
"length_ratio_min": 0.501953125,
"picked": 24,
"run": "serve-r0.json",
"samples": 24,
"status": "audited",
"top1_gap": 0.01639407033295448
}
},
"budget": {
"interactive@1.02": {
"budget_gib": 11.325,
"factor": 1.02,
"full_running": 1.0,
"healthy": true,
"running": 1.0,
"value": 305.7984912860148
},
"long_context@1.02": {
"budget_gib": 83.987,
"factor": 1.02,
"full_running": 64.0,
"healthy": true,
"running": 64.0,
"value": 1024.3082042840674
},
"serve@1.02": {
"budget_gib": 30.7,
"factor": 1.02,
"full_running": 32.0,
"healthy": true,
"running": 32.0,
"value": 2774.9187179136993
}
},
"startup": {
"cuda_graph_gib": 0.44,
"engine_init_seconds": 12.35,
"kv_bytes_per_token": 73727.87291882833,
"kv_cache_dtype": "fp8",
"kv_cache_gib": 150.14,
"kv_cache_tokens": 2186576.0,
"load_seconds": 3.352658,
"max_concurrency": {
"sequences": 66.73,
"tokens_per_request": 32768
},
"speculative_config": "None",
"vllm_version": "0.1.dev18498+g7a33ba95f",
"weights_gib": 8.84
},
"stored_gib": 8.79
},
"gpu_hours": 0.779,
"kl": {
"identity": {
"kl": 0.0,
"kl_median": 0.0,
"kl_per_token": 0.0,
"kl_se": 0.0,
"sequences": 256,
"tokens": 62076
},
"kl": 0.009696001195112797,
"kl_median": 0.00793668609329845,
"kl_per_token": 0.009684813823354317,
"kl_se": 0.0007364314878597962,
"ppl": {
"hidden": {
"dnll": 0.0009363764522941531,
"dnll_se": 0.0005956011568420956,
"ppl": 15.633382484622631,
"ppl_bf16": 15.618750604937217,
"tokens": 262016,
"windows": 128
},
"wikitext2": {
"dnll": 0.0011002849716787954,
"dnll_se": 0.0006998532920999665,
"ppl": 9.732609386380801,
"ppl_bf16": 9.721906631657593,
"tokens": 298862,
"windows": 146
}
},
"sequences": 256,
"tokens": 62197
},
"length_shift": 0.01762,
"mean_drop": 0.00587,
"memories": {
"interactive": 1.5843,
"long_context": 1.94,
"serve": 1.8474
},
"memory": 1.784,
"memory_saved": 0.4587,
"metric": "lowbit_score",
"peak_gib": 30.098,
"peak_memory": {
"interactive": {
"device_peak_gib": 164.366,
"kv_bytes_per_token": 73728,
"kv_gib": 0.088,
"kv_usage": 0.0006,
"other_gib": 2.175,
"peak_gib": 11.103,
"weights_gib": 8.84
},
"long_context": {
"device_peak_gib": 164.366,
"kv_bytes_per_token": 73612,
"kv_gib": 71.325,
"kv_usage": 0.4751,
"other_gib": 2.175,
"peak_gib": 82.34,
"weights_gib": 8.84
},
"serve": {
"device_peak_gib": 164.366,
"kv_bytes_per_token": 73563,
"kv_gib": 19.082,
"kv_usage": 0.1271,
"other_gib": 2.175,
"peak_gib": 30.098,
"weights_gib": 8.84
}
},
"precision": {
"activations": 8,
"kv": 8,
"label": "W8A8KV8",
"weights": 8,
"weights_average": 8.0
},
"qualified": true,
"quality": {
"tier": "A",
"valid_domains": 5
},
"reason": null,
"score": 1.4563,
"score_se": 0.0394,
"speed": 1.3506,
"speed_serve": 1.4367,
"speeds": {
"interactive": 1.2284,
"long_context": 1.3961,
"serve": 1.4367
},
"task_id": "qwen3-8b",
"timed_audit": {
"interactive": {
"bf16_mean": -0.13170158582772096,
"bf16_top1": 0.9895301533100402,
"empty": 1,
"empty_alone": 0,
"empty_reference": 1,
"engine_mean": -0.33488874823383913,
"engine_top1": 0.8934413514207463,
"gap": 0.20318716240611817,
"gap_max": 2.259063767962636,
"gap_median": 0.12190573442048194,
"length_ratio": 1.0,
"length_ratio_min": 0.0,
"picked": 24,
"run": "interactive-r0.json",
"samples": 23,
"status": "audited",
"top1_gap": 0.09608880188929392
},
"long_context": {
"bf16_mean": -0.3193891973994322,
"bf16_top1": 0.9852744424349366,
"empty": 0,
"empty_alone": 0,
"empty_reference": 0,
"engine_mean": -0.22300654186599272,
"engine_top1": 0.9493727651074984,
"gap": -0.09638265553343953,
"gap_max": 0.35711406119412015,
"gap_median": 0.04173304980499706,
"length_ratio": 1.0,
"length_ratio_min": 0.5078125,
"picked": 24,
"run": "long_context-r0.json",
"samples": 24,
"status": "audited",
"top1_gap": 0.03590167732743821
},
"memory_capped": {
"status": "missing"
},
"offline": {
"status": "missing"
},
"online_slo": {
"status": "missing"
},
"serve": {
"bf16_mean": -0.06854064463553336,
"bf16_top1": 0.9947859789301092,
"empty": 0,
"empty_alone": 0,
"empty_reference": 0,
"engine_mean": -0.09481864496485903,
"engine_top1": 0.9783919085971547,
"gap": 0.02627800032932568,
"gap_max": 0.3413345401085057,
"gap_median": 0.021810810235200748,
"length_ratio": 1.0,
"length_ratio_min": 0.501953125,
"picked": 24,
"run": "serve-r0.json",
"samples": 24,
"status": "audited",
"top1_gap": 0.01639407033295448
}
}
},
"runner": "kubernetes",
"solution": {
"checkpoint": "Qwen3-8B-FP8-Dynamic",
"declared_precision": true,
"detected_precision": "W8A8KV8",
"method": "FP8-chunked-prefill",
"precision": "W8A8KV8"
},
"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": "engine",
"updated_at": "2026-10-11T21:00:22Z"
}