{ "schema_version": "xencore-benchmark-publication-api/1", "dataset_name": "XenCore Nexus Intel Arc Pro B70 benchmark results", "canonical_url": "https://xencorenexus.com/api/benchs.json", "publication_status": "results_available", "validation_policy": { "runner_self_validation_accepted": false, "required": [ "measurement_artifact", "artifact_sha256", "independent_validator_report" ], "missing_values": "remain_null_or_absent" }, "total_benchmarks": 56, "benchmarks": [ { "id": "Qwen2.5-0.5B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-0.5B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.5B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 0.93, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 222.69, "aggregateDecodeTokS": 16919.64, "perRequestDecodeTokS": 222.69, "ttftMs": 36.7, "acceptanceRate": null, "energyJoules": 139.364, "avgPowerWatts": 77.77, "joulesPerToken": 1.0888, "vramUsedGiB": 30.13, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 222.69, "perRequestTokS": 222.69, "ttftMs": 36.7, "energyJoules": 139.364, "avgPowerWatts": 77.77, "joulesPerToken": 1.0888 }, { "concurrency": 2, "aggregateThroughputTokS": 338.93, "perRequestTokS": 227.08, "ttftMs": 25.32, "energyJoules": 119.246, "avgPowerWatts": 88.23, "joulesPerToken": 0.4658 }, { "concurrency": 4, "aggregateThroughputTokS": 842.19, "perRequestTokS": 226.39, "ttftMs": 23.9, "energyJoules": 124.362, "avgPowerWatts": 90.85, "joulesPerToken": 0.2429 }, { "concurrency": 8, "aggregateThroughputTokS": 1534.62, "perRequestTokS": 220.21, "ttftMs": 797.85, "energyJoules": 165.716, "avgPowerWatts": 77.11, "joulesPerToken": 0.1618 }, { "concurrency": 16, "aggregateThroughputTokS": 1963.71, "perRequestTokS": 223.69, "ttftMs": 30.6, "energyJoules": 128.25, "avgPowerWatts": 94.02, "joulesPerToken": 0.0626 }, { "concurrency": 32, "aggregateThroughputTokS": 5222.97, "perRequestTokS": 216.68, "ttftMs": 49.01, "energyJoules": 145.819, "avgPowerWatts": 100.73, "joulesPerToken": 0.0356 }, { "concurrency": 64, "aggregateThroughputTokS": 8083.01, "perRequestTokS": 166.46, "ttftMs": 80.17, "energyJoules": 192.351, "avgPowerWatts": 117.05, "joulesPerToken": 0.0235 }, { "concurrency": 128, "aggregateThroughputTokS": 12430.21, "perRequestTokS": 149.04, "ttftMs": 144.11, "energyJoules": 253.094, "avgPowerWatts": 137.84, "joulesPerToken": 0.0154 }, { "concurrency": 256, "aggregateThroughputTokS": 16838.65, "perRequestTokS": 98.74, "ttftMs": 275.7, "energyJoules": 411.036, "avgPowerWatts": 168.81, "joulesPerToken": 0.0125 }, { "concurrency": 512, "aggregateThroughputTokS": 16919.64, "perRequestTokS": 89.62, "ttftMs": 961.91, "energyJoules": 768.511, "avgPowerWatts": 196.17, "joulesPerToken": 0.0117 }, { "concurrency": 1024, "aggregateThroughputTokS": 16824.95, "perRequestTokS": 99.35, "ttftMs": 2399.33, "energyJoules": 1478.168, "avgPowerWatts": 214.94, "joulesPerToken": 0.0113 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:47:50.404620+00:00", "notes": null, "run_key": "Qwen2.5-0.5B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-0.5B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-0.6B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-0.6B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "0.6B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 1.41, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 159.07, "aggregateDecodeTokS": 11544.51, "perRequestDecodeTokS": 159.07, "ttftMs": 38.46, "acceptanceRate": null, "energyJoules": 156.642, "avgPowerWatts": 88.68, "joulesPerToken": 1.2238, "vramUsedGiB": 30.08, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 159.07, "perRequestTokS": 159.07, "ttftMs": 38.46, "energyJoules": 156.642, "avgPowerWatts": 88.68, "joulesPerToken": 1.2238 }, { "concurrency": 2, "aggregateThroughputTokS": 352.38, "perRequestTokS": 176.93, "ttftMs": 28.89, "energyJoules": 149.718, "avgPowerWatts": 90.88, "joulesPerToken": 0.5848 }, { "concurrency": 4, "aggregateThroughputTokS": 692.73, "perRequestTokS": 174.4, "ttftMs": 25.08, "energyJoules": 156.351, "avgPowerWatts": 92.92, "joulesPerToken": 0.3054 }, { "concurrency": 8, "aggregateThroughputTokS": 1398.58, "perRequestTokS": 177.17, "ttftMs": 918.59, "energyJoules": 202.233, "avgPowerWatts": 79.48, "joulesPerToken": 0.1975 }, { "concurrency": 16, "aggregateThroughputTokS": 2806.7, "perRequestTokS": 180.01, "ttftMs": 45.1, "energyJoules": 168.345, "avgPowerWatts": 102.13, "joulesPerToken": 0.0822 }, { "concurrency": 32, "aggregateThroughputTokS": 4949.91, "perRequestTokS": 161.18, "ttftMs": 55.86, "energyJoules": 198.055, "avgPowerWatts": 115.12, "joulesPerToken": 0.0484 }, { "concurrency": 64, "aggregateThroughputTokS": 7738.2, "perRequestTokS": 130.32, "ttftMs": 90.43, "energyJoules": 275.292, "avgPowerWatts": 136.06, "joulesPerToken": 0.0336 }, { "concurrency": 128, "aggregateThroughputTokS": 9989.23, "perRequestTokS": 89.59, "ttftMs": 222.38, "energyJoules": 421.248, "avgPowerWatts": 161.5, "joulesPerToken": 0.0257 }, { "concurrency": 256, "aggregateThroughputTokS": 11544.51, "perRequestTokS": 52.11, "ttftMs": 337.21, "energyJoules": 735.577, "avgPowerWatts": 192.72, "joulesPerToken": 0.0224 }, { "concurrency": 512, "aggregateThroughputTokS": 11532.87, "perRequestTokS": 49.92, "ttftMs": 1760.61, "energyJoules": 1418.566, "avgPowerWatts": 213.51, "joulesPerToken": 0.0216 }, { "concurrency": 1024, "aggregateThroughputTokS": 11042.61, "perRequestTokS": 47.65, "ttftMs": 4781.26, "energyJoules": 2884.097, "avgPowerWatts": 227.17, "joulesPerToken": 0.022 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:48:00.393168+00:00", "notes": null, "run_key": "Qwen3-0.6B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3-0.6B-BF16-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-0.6B-FP8-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-0.6B-FP8", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.6B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 1, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 163.99, "aggregateDecodeTokS": 10436.13, "perRequestDecodeTokS": 163.99, "ttftMs": 30.24, "acceptanceRate": null, "energyJoules": 168.692, "avgPowerWatts": 95.59, "joulesPerToken": 1.3179, "vramUsedGiB": 30.07, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 163.99, "perRequestTokS": 163.99, "ttftMs": 30.24, "energyJoules": 168.692, "avgPowerWatts": 95.59, "joulesPerToken": 1.3179 }, { "concurrency": 2, "aggregateThroughputTokS": 303.14, "perRequestTokS": 152.16, "ttftMs": 22.11, "energyJoules": 176.87, "avgPowerWatts": 97.31, "joulesPerToken": 0.6909 }, { "concurrency": 4, "aggregateThroughputTokS": 628.38, "perRequestTokS": 158.1, "ttftMs": 23.12, "energyJoules": 174.947, "avgPowerWatts": 100.83, "joulesPerToken": 0.3417 }, { "concurrency": 8, "aggregateThroughputTokS": 1217.6, "perRequestTokS": 154.2, "ttftMs": 61.41, "energyJoules": 190.003, "avgPowerWatts": 105.16, "joulesPerToken": 0.1855 }, { "concurrency": 16, "aggregateThroughputTokS": 2384.89, "perRequestTokS": 154.34, "ttftMs": 48.08, "energyJoules": 202.609, "avgPowerWatts": 113.29, "joulesPerToken": 0.0989 }, { "concurrency": 32, "aggregateThroughputTokS": 4716.34, "perRequestTokS": 159.49, "ttftMs": 83.03, "energyJoules": 232.406, "avgPowerWatts": 128.8, "joulesPerToken": 0.0567 }, { "concurrency": 64, "aggregateThroughputTokS": 7264.04, "perRequestTokS": 128.92, "ttftMs": 147.28, "energyJoules": 307.932, "avgPowerWatts": 146.94, "joulesPerToken": 0.0376 }, { "concurrency": 128, "aggregateThroughputTokS": 9067.55, "perRequestTokS": 82.36, "ttftMs": 255.41, "energyJoules": 496.967, "avgPowerWatts": 185.29, "joulesPerToken": 0.0303 }, { "concurrency": 256, "aggregateThroughputTokS": 10436.13, "perRequestTokS": 49.82, "ttftMs": 521.78, "energyJoules": 904.742, "avgPowerWatts": 221.84, "joulesPerToken": 0.0276 }, { "concurrency": 512, "aggregateThroughputTokS": 10416.43, "perRequestTokS": 46.37, "ttftMs": 2086.34, "energyJoules": 1786.405, "avgPowerWatts": 246.09, "joulesPerToken": 0.0273 }, { "concurrency": 1024, "aggregateThroughputTokS": 10068.47, "perRequestTokS": 43.88, "ttftMs": 5364.81, "energyJoules": 3604.403, "avgPowerWatts": 259.45, "joulesPerToken": 0.0275 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:48:05.393673+00:00", "notes": null, "run_key": "Qwen3-0.6B-FP8-vllm-xpu-tp1-baseline", "run_id": "Qwen3-0.6B-FP8-vllm-xpu-tp1-baseline" }, { "id": "Qwen3.5-0.8B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.5-0.8B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "0.8B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 3.3, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 213.77, "aggregateDecodeTokS": 6854.81, "perRequestDecodeTokS": 213.77, "ttftMs": 36.39, "acceptanceRate": null, "energyJoules": 161.46, "avgPowerWatts": 87.52, "joulesPerToken": 1.2614, "vramUsedGiB": 28.5, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 213.77, "perRequestTokS": 213.77, "ttftMs": 36.39, "energyJoules": 161.46, "avgPowerWatts": 87.52, "joulesPerToken": 1.2614 }, { "concurrency": 2, "aggregateThroughputTokS": 405.34, "perRequestTokS": 203.59, "ttftMs": 34.69, "energyJoules": 146.979, "avgPowerWatts": 94.18, "joulesPerToken": 0.5741 }, { "concurrency": 4, "aggregateThroughputTokS": 838.09, "perRequestTokS": 211.15, "ttftMs": 43.64, "energyJoules": 156.87, "avgPowerWatts": 95.95, "joulesPerToken": 0.3064 }, { "concurrency": 8, "aggregateThroughputTokS": 1562.85, "perRequestTokS": 206.43, "ttftMs": 47.55, "energyJoules": 162.874, "avgPowerWatts": 105.21, "joulesPerToken": 0.1591 }, { "concurrency": 16, "aggregateThroughputTokS": 2735.44, "perRequestTokS": 177.08, "ttftMs": 48.28, "energyJoules": 194.885, "avgPowerWatts": 116.02, "joulesPerToken": 0.0952 }, { "concurrency": 32, "aggregateThroughputTokS": 3835.41, "perRequestTokS": 138.48, "ttftMs": 72.24, "energyJoules": 251.239, "avgPowerWatts": 132.4, "joulesPerToken": 0.0613 }, { "concurrency": 64, "aggregateThroughputTokS": 5306.38, "perRequestTokS": 93.62, "ttftMs": 120.86, "energyJoules": 389.791, "avgPowerWatts": 164.72, "joulesPerToken": 0.0476 }, { "concurrency": 128, "aggregateThroughputTokS": 6473.34, "perRequestTokS": 57.79, "ttftMs": 246.3, "energyJoules": 652.525, "avgPowerWatts": 195.32, "joulesPerToken": 0.0398 }, { "concurrency": 256, "aggregateThroughputTokS": 6854.81, "perRequestTokS": 31.8, "ttftMs": 528.86, "energyJoules": 1225.117, "avgPowerWatts": 222.7, "joulesPerToken": 0.0374 }, { "concurrency": 512, "aggregateThroughputTokS": 6373.24, "perRequestTokS": 28.16, "ttftMs": 3120.21, "energyJoules": 2567.526, "avgPowerWatts": 236.79, "joulesPerToken": 0.0392 }, { "concurrency": 1024, "aggregateThroughputTokS": 5808.61, "perRequestTokS": 25.84, "ttftMs": 8657.44, "energyJoules": 5438.167, "avgPowerWatts": 241.69, "joulesPerToken": 0.0415 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:49:39.505751+00:00", "notes": null, "run_key": "Qwen3.5-0.8B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3.5-0.8B-BF16-vllm-xpu-tp1-baseline" }, { "id": "DeepSeek-R1-Distill-Qwen-1.5B-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "DeepSeek-R1-Distill-Qwen-1.5B", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "1.5B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 3.32, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 146.48, "aggregateDecodeTokS": 10875.02, "perRequestDecodeTokS": 146.48, "ttftMs": 47.1, "acceptanceRate": null, "energyJoules": 210.198, "avgPowerWatts": 118.52, "joulesPerToken": 1.6422, "vramUsedGiB": 29.98, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 146.48, "perRequestTokS": 146.48, "ttftMs": 47.1, "energyJoules": 210.198, "avgPowerWatts": 118.52, "joulesPerToken": 1.6422 }, { "concurrency": 2, "aggregateThroughputTokS": 292.86, "perRequestTokS": 147.55, "ttftMs": 48.86, "energyJoules": 210.175, "avgPowerWatts": 120.39, "joulesPerToken": 0.821 }, { "concurrency": 4, "aggregateThroughputTokS": 566.91, "perRequestTokS": 143.43, "ttftMs": 27.61, "energyJoules": 215.469, "avgPowerWatts": 124.78, "joulesPerToken": 0.4208 }, { "concurrency": 8, "aggregateThroughputTokS": 1086.85, "perRequestTokS": 137.04, "ttftMs": 2095.74, "energyJoules": 336.095, "avgPowerWatts": 86.64, "joulesPerToken": 0.3282 }, { "concurrency": 16, "aggregateThroughputTokS": 2091.09, "perRequestTokS": 131.96, "ttftMs": 388.35, "energyJoules": 264.028, "avgPowerWatts": 122.38, "joulesPerToken": 0.1289 }, { "concurrency": 32, "aggregateThroughputTokS": 3484.15, "perRequestTokS": 116.93, "ttftMs": 94.91, "energyJoules": 295.495, "avgPowerWatts": 148.39, "joulesPerToken": 0.0721 }, { "concurrency": 64, "aggregateThroughputTokS": 5613.08, "perRequestTokS": 98.98, "ttftMs": 168.98, "energyJoules": 393.971, "avgPowerWatts": 169.67, "joulesPerToken": 0.0481 }, { "concurrency": 128, "aggregateThroughputTokS": 8290.98, "perRequestTokS": 76.26, "ttftMs": 287.86, "energyJoules": 579.615, "avgPowerWatts": 206.3, "joulesPerToken": 0.0354 }, { "concurrency": 256, "aggregateThroughputTokS": 10763.25, "perRequestTokS": 52.97, "ttftMs": 564.38, "energyJoules": 943.165, "avgPowerWatts": 232.67, "joulesPerToken": 0.0288 }, { "concurrency": 512, "aggregateThroughputTokS": 10875.02, "perRequestTokS": 49.5, "ttftMs": 2079.55, "energyJoules": 1825.423, "avgPowerWatts": 260.73, "joulesPerToken": 0.0279 }, { "concurrency": 1024, "aggregateThroughputTokS": 10823.04, "perRequestTokS": 47.87, "ttftMs": 5168.2, "energyJoules": 3599.979, "avgPowerWatts": 277.32, "joulesPerToken": 0.0275 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:50:10.814726+00:00", "notes": null, "run_key": "DeepSeek-R1-Distill-Qwen-1.5B-vllm-xpu-tp1-baseline", "run_id": "DeepSeek-R1-Distill-Qwen-1.5B-vllm-xpu-tp1-baseline" }, { "id": "Qwen3.5-2B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.5-2B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "2B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 4.26, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 128.35, "aggregateDecodeTokS": 5459.25, "perRequestDecodeTokS": 128.35, "ttftMs": 44.84, "acceptanceRate": null, "energyJoules": 254.167, "avgPowerWatts": 121.92, "joulesPerToken": 1.9857, "vramUsedGiB": 28.12, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 128.35, "perRequestTokS": 128.35, "ttftMs": 44.84, "energyJoules": 254.167, "avgPowerWatts": 121.92, "joulesPerToken": 1.9857 }, { "concurrency": 2, "aggregateThroughputTokS": 247.76, "perRequestTokS": 125.26, "ttftMs": 30.1, "energyJoules": 253.096, "avgPowerWatts": 126.41, "joulesPerToken": 0.9887 }, { "concurrency": 4, "aggregateThroughputTokS": 436.24, "perRequestTokS": 117.47, "ttftMs": 39.47, "energyJoules": 270.055, "avgPowerWatts": 126.09, "joulesPerToken": 0.5275 }, { "concurrency": 8, "aggregateThroughputTokS": 779.66, "perRequestTokS": 116.35, "ttftMs": 66.08, "energyJoules": 279.637, "avgPowerWatts": 135.73, "joulesPerToken": 0.2731 }, { "concurrency": 16, "aggregateThroughputTokS": 1613.9, "perRequestTokS": 104.94, "ttftMs": 75.17, "energyJoules": 328.818, "avgPowerWatts": 146.32, "joulesPerToken": 0.1606 }, { "concurrency": 32, "aggregateThroughputTokS": 2517.93, "perRequestTokS": 87.57, "ttftMs": 116.49, "energyJoules": 414.593, "avgPowerWatts": 166.9, "joulesPerToken": 0.1012 }, { "concurrency": 64, "aggregateThroughputTokS": 3689.84, "perRequestTokS": 64.1, "ttftMs": 193.37, "energyJoules": 596.455, "avgPowerWatts": 190.15, "joulesPerToken": 0.0728 }, { "concurrency": 128, "aggregateThroughputTokS": 4861.68, "perRequestTokS": 44.39, "ttftMs": 358.48, "energyJoules": 918.067, "avgPowerWatts": 216.2, "joulesPerToken": 0.056 }, { "concurrency": 256, "aggregateThroughputTokS": 5459.25, "perRequestTokS": 25.6, "ttftMs": 782.57, "energyJoules": 1715.622, "avgPowerWatts": 249.28, "joulesPerToken": 0.0524 }, { "concurrency": 512, "aggregateThroughputTokS": 5117.88, "perRequestTokS": 23.02, "ttftMs": 3996.05, "energyJoules": 3515.545, "avgPowerWatts": 263.64, "joulesPerToken": 0.0536 }, { "concurrency": 1024, "aggregateThroughputTokS": 4903.53, "perRequestTokS": 22.04, "ttftMs": 10425.55, "energyJoules": 7288.86, "avgPowerWatts": 272.1, "joulesPerToken": 0.0556 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:50:17.454701+00:00", "notes": null, "run_key": "Qwen3.5-2B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3.5-2B-BF16-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-3B-Instruct-AWQ-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-3B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "3B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 2.51, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 163.62, "aggregateDecodeTokS": 6752.3, "perRequestDecodeTokS": 163.62, "ttftMs": 50.51, "acceptanceRate": null, "energyJoules": 206.224, "avgPowerWatts": 113.48, "joulesPerToken": 1.6111, "vramUsedGiB": 29.82, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 163.62, "perRequestTokS": 163.62, "ttftMs": 50.51, "energyJoules": 206.224, "avgPowerWatts": 113.48, "joulesPerToken": 1.6111 }, { "concurrency": 2, "aggregateThroughputTokS": 318.12, "perRequestTokS": 159.76, "ttftMs": 97.45, "energyJoules": 202.086, "avgPowerWatts": 120.33, "joulesPerToken": 0.7894 }, { "concurrency": 4, "aggregateThroughputTokS": 589.37, "perRequestTokS": 161.28, "ttftMs": 119.79, "energyJoules": 210.586, "avgPowerWatts": 124.36, "joulesPerToken": 0.4113 }, { "concurrency": 8, "aggregateThroughputTokS": 1115.62, "perRequestTokS": 141.75, "ttftMs": 919.58, "energyJoules": 279.022, "avgPowerWatts": 105.38, "joulesPerToken": 0.2725 }, { "concurrency": 16, "aggregateThroughputTokS": 2202.89, "perRequestTokS": 157.18, "ttftMs": 137.68, "energyJoules": 261.959, "avgPowerWatts": 143.84, "joulesPerToken": 0.1279 }, { "concurrency": 32, "aggregateThroughputTokS": 3340.94, "perRequestTokS": 119.82, "ttftMs": 165.99, "energyJoules": 383.236, "avgPowerWatts": 188.05, "joulesPerToken": 0.0936 }, { "concurrency": 64, "aggregateThroughputTokS": 4442.81, "perRequestTokS": 86.19, "ttftMs": 280.03, "energyJoules": 522.908, "avgPowerWatts": 203.73, "joulesPerToken": 0.0638 }, { "concurrency": 128, "aggregateThroughputTokS": 6005.33, "perRequestTokS": 60.4, "ttftMs": 514.22, "energyJoules": 832.709, "avgPowerWatts": 237.71, "joulesPerToken": 0.0508 }, { "concurrency": 256, "aggregateThroughputTokS": 6542.33, "perRequestTokS": 34.77, "ttftMs": 1018.79, "energyJoules": 1506.069, "avgPowerWatts": 265.17, "joulesPerToken": 0.046 }, { "concurrency": 512, "aggregateThroughputTokS": 6585.53, "perRequestTokS": 31.56, "ttftMs": 3383.56, "energyJoules": 2969.852, "avgPowerWatts": 284.6, "joulesPerToken": 0.0453 }, { "concurrency": 1024, "aggregateThroughputTokS": 6752.3, "perRequestTokS": 30.11, "ttftMs": 7942.37, "energyJoules": 5907.365, "avgPowerWatts": 299.27, "joulesPerToken": 0.0451 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:51:54.921742+00:00", "notes": null, "run_key": "Qwen2.5-3B-Instruct-AWQ-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-3B-Instruct-AWQ-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-3B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-3B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "3B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 5.76, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 80.09, "aggregateDecodeTokS": 6306.53, "perRequestDecodeTokS": 80.09, "ttftMs": 45.27, "acceptanceRate": null, "energyJoules": 353.591, "avgPowerWatts": 140.28, "joulesPerToken": 2.7624, "vramUsedGiB": 29.83, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 80.09, "perRequestTokS": 80.09, "ttftMs": 45.27, "energyJoules": 353.591, "avgPowerWatts": 140.28, "joulesPerToken": 2.7624 }, { "concurrency": 2, "aggregateThroughputTokS": 159.38, "perRequestTokS": 80.2, "ttftMs": 51.9, "energyJoules": 352.77, "avgPowerWatts": 144.46, "joulesPerToken": 1.378 }, { "concurrency": 4, "aggregateThroughputTokS": 298.7, "perRequestTokS": 79.02, "ttftMs": 37.66, "energyJoules": 364.732, "avgPowerWatts": 146.74, "joulesPerToken": 0.7124 }, { "concurrency": 8, "aggregateThroughputTokS": 586.72, "perRequestTokS": 78.3, "ttftMs": 921.63, "energyJoules": 427.542, "avgPowerWatts": 126.13, "joulesPerToken": 0.4175 }, { "concurrency": 16, "aggregateThroughputTokS": 1138.75, "perRequestTokS": 74.48, "ttftMs": 105.9, "energyJoules": 424.05, "avgPowerWatts": 162.62, "joulesPerToken": 0.2071 }, { "concurrency": 32, "aggregateThroughputTokS": 1983.05, "perRequestTokS": 69.41, "ttftMs": 158.25, "energyJoules": 514.972, "avgPowerWatts": 178.78, "joulesPerToken": 0.1257 }, { "concurrency": 64, "aggregateThroughputTokS": 3270.13, "perRequestTokS": 58.93, "ttftMs": 281.61, "energyJoules": 671.466, "avgPowerWatts": 204.87, "joulesPerToken": 0.082 }, { "concurrency": 128, "aggregateThroughputTokS": 4974.45, "perRequestTokS": 47.54, "ttftMs": 498.91, "energyJoules": 971.145, "avgPowerWatts": 239.79, "joulesPerToken": 0.0593 }, { "concurrency": 256, "aggregateThroughputTokS": 6092.79, "perRequestTokS": 31.45, "ttftMs": 991.33, "energyJoules": 1670.953, "avgPowerWatts": 275.33, "joulesPerToken": 0.051 }, { "concurrency": 512, "aggregateThroughputTokS": 6261.87, "perRequestTokS": 29, "ttftMs": 3451.31, "energyJoules": 3287.017, "avgPowerWatts": 297.76, "joulesPerToken": 0.0502 }, { "concurrency": 1024, "aggregateThroughputTokS": 6306.53, "perRequestTokS": 27.68, "ttftMs": 8289.07, "energyJoules": 6494.215, "avgPowerWatts": 310.72, "joulesPerToken": 0.0495 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:52:06.405673+00:00", "notes": null, "run_key": "Qwen2.5-3B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-3B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen3.5-4B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.5-4B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "4B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 8.7, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 61.53, "aggregateDecodeTokS": 2260.81, "perRequestDecodeTokS": 61.53, "ttftMs": 96.84, "acceptanceRate": null, "energyJoules": 490.509, "avgPowerWatts": 149.57, "joulesPerToken": 3.8321, "vramUsedGiB": 28.23, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 61.53, "perRequestTokS": 61.53, "ttftMs": 96.84, "energyJoules": 490.509, "avgPowerWatts": 149.57, "joulesPerToken": 3.8321 }, { "concurrency": 2, "aggregateThroughputTokS": 118.32, "perRequestTokS": 59.85, "ttftMs": 71.37, "energyJoules": 499.426, "avgPowerWatts": 155, "joulesPerToken": 1.9509 }, { "concurrency": 4, "aggregateThroughputTokS": 227.33, "perRequestTokS": 57.8, "ttftMs": 73.67, "energyJoules": 518.109, "avgPowerWatts": 160.44, "joulesPerToken": 1.0119 }, { "concurrency": 8, "aggregateThroughputTokS": 426.79, "perRequestTokS": 54.79, "ttftMs": 104.69, "energyJoules": 574.301, "avgPowerWatts": 167.62, "joulesPerToken": 0.5608 }, { "concurrency": 16, "aggregateThroughputTokS": 749.43, "perRequestTokS": 49.26, "ttftMs": 170.46, "energyJoules": 686.456, "avgPowerWatts": 178.49, "joulesPerToken": 0.3352 }, { "concurrency": 32, "aggregateThroughputTokS": 1198.07, "perRequestTokS": 40.21, "ttftMs": 256.77, "energyJoules": 899.502, "avgPowerWatts": 204.64, "joulesPerToken": 0.2196 }, { "concurrency": 64, "aggregateThroughputTokS": 1655.89, "perRequestTokS": 28.46, "ttftMs": 432.07, "energyJoules": 1330.045, "avgPowerWatts": 225.03, "joulesPerToken": 0.1624 }, { "concurrency": 128, "aggregateThroughputTokS": 2090.2, "perRequestTokS": 18.42, "ttftMs": 797.81, "energyJoules": 2180.503, "avgPowerWatts": 248.19, "joulesPerToken": 0.1331 }, { "concurrency": 256, "aggregateThroughputTokS": 2077.21, "perRequestTokS": 13.33, "ttftMs": 2363.35, "energyJoules": 4430.877, "avgPowerWatts": 265.74, "joulesPerToken": 0.1352 }, { "concurrency": 512, "aggregateThroughputTokS": 2187.75, "perRequestTokS": 12.11, "ttftMs": 9338.47, "energyJoules": 8647.161, "avgPowerWatts": 280.5, "joulesPerToken": 0.1319 }, { "concurrency": 1024, "aggregateThroughputTokS": 2260.81, "perRequestTokS": 11.24, "ttftMs": 23038.73, "energyJoules": 16974.347, "avgPowerWatts": 291.57, "joulesPerToken": 0.1295 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:53:03.556974+00:00", "notes": null, "run_key": "Qwen3.5-4B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3.5-4B-BF16-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-7B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-7B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 28.39, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 38.59, "aggregateDecodeTokS": 3416.92, "perRequestDecodeTokS": 38.59, "ttftMs": 61.63, "acceptanceRate": null, "energyJoules": 733.915, "avgPowerWatts": 171.52, "joulesPerToken": 5.7337, "vramUsedGiB": 29.49, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 38.59, "perRequestTokS": 38.59, "ttftMs": 61.63, "energyJoules": 733.915, "avgPowerWatts": 171.52, "joulesPerToken": 5.7337 }, { "concurrency": 2, "aggregateThroughputTokS": 76.56, "perRequestTokS": 38.57, "ttftMs": 63.1, "energyJoules": 742.305, "avgPowerWatts": 174.26, "joulesPerToken": 2.8996 }, { "concurrency": 4, "aggregateThroughputTokS": 151.15, "perRequestTokS": 38.3, "ttftMs": 75.33, "energyJoules": 770.054, "avgPowerWatts": 182.63, "joulesPerToken": 1.504 }, { "concurrency": 8, "aggregateThroughputTokS": 303.84, "perRequestTokS": 38.1, "ttftMs": 848.45, "energyJoules": 847.874, "avgPowerWatts": 168.8, "joulesPerToken": 0.828 }, { "concurrency": 16, "aggregateThroughputTokS": 549.4, "perRequestTokS": 36.65, "ttftMs": 233, "energyJoules": 909.28, "avgPowerWatts": 199.04, "joulesPerToken": 0.444 }, { "concurrency": 32, "aggregateThroughputTokS": 990.5, "perRequestTokS": 34.15, "ttftMs": 340.67, "energyJoules": 1078.566, "avgPowerWatts": 220.21, "joulesPerToken": 0.2633 }, { "concurrency": 64, "aggregateThroughputTokS": 1651.05, "perRequestTokS": 29.59, "ttftMs": 602.96, "energyJoules": 1388.57, "avgPowerWatts": 239.44, "joulesPerToken": 0.1695 }, { "concurrency": 128, "aggregateThroughputTokS": 2539.98, "perRequestTokS": 24.55, "ttftMs": 1081.91, "energyJoules": 1950.164, "avgPowerWatts": 269.59, "joulesPerToken": 0.119 }, { "concurrency": 256, "aggregateThroughputTokS": 3366.22, "perRequestTokS": 17.71, "ttftMs": 2072.27, "energyJoules": 3196.027, "avgPowerWatts": 304.72, "joulesPerToken": 0.0975 }, { "concurrency": 512, "aggregateThroughputTokS": 3402.56, "perRequestTokS": 16.08, "ttftMs": 6651.17, "energyJoules": 6250.157, "avgPowerWatts": 316.33, "joulesPerToken": 0.0954 }, { "concurrency": 1024, "aggregateThroughputTokS": 3416.92, "perRequestTokS": 15.14, "ttftMs": 15816.38, "energyJoules": 12391.32, "avgPowerWatts": 322.01, "joulesPerToken": 0.0945 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:54:26.915798+00:00", "notes": null, "run_key": "Qwen2.5-7B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-7B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-VL-7B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-VL-7B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 29.93, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 38.94, "aggregateDecodeTokS": 3292.3, "perRequestDecodeTokS": 38.94, "ttftMs": 62.43, "acceptanceRate": null, "energyJoules": 716.544, "avgPowerWatts": 172.69, "joulesPerToken": 5.598, "vramUsedGiB": 27.95, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 38.94, "perRequestTokS": 38.94, "ttftMs": 62.43, "energyJoules": 716.544, "avgPowerWatts": 172.69, "joulesPerToken": 5.598 }, { "concurrency": 2, "aggregateThroughputTokS": 61.19, "perRequestTokS": 38.84, "ttftMs": 62.21, "energyJoules": 727.622, "avgPowerWatts": 177.09, "joulesPerToken": 2.8423 }, { "concurrency": 4, "aggregateThroughputTokS": 147.3, "perRequestTokS": 38.6, "ttftMs": 74.24, "energyJoules": 765.126, "avgPowerWatts": 183.36, "joulesPerToken": 1.4944 }, { "concurrency": 8, "aggregateThroughputTokS": 265.59, "perRequestTokS": 38.37, "ttftMs": 121.26, "energyJoules": 805.561, "avgPowerWatts": 189.2, "joulesPerToken": 0.7867 }, { "concurrency": 16, "aggregateThroughputTokS": 496.45, "perRequestTokS": 37.49, "ttftMs": 234.07, "energyJoules": 889.211, "avgPowerWatts": 198.34, "joulesPerToken": 0.4342 }, { "concurrency": 32, "aggregateThroughputTokS": 930.69, "perRequestTokS": 34.3, "ttftMs": 343.99, "energyJoules": 1066.703, "avgPowerWatts": 219.2, "joulesPerToken": 0.2604 }, { "concurrency": 64, "aggregateThroughputTokS": 1553.58, "perRequestTokS": 29.99, "ttftMs": 601.09, "energyJoules": 1351.752, "avgPowerWatts": 239.02, "joulesPerToken": 0.165 }, { "concurrency": 128, "aggregateThroughputTokS": 2429.36, "perRequestTokS": 24.95, "ttftMs": 1086.2, "energyJoules": 1889.14, "avgPowerWatts": 267.52, "joulesPerToken": 0.1153 }, { "concurrency": 256, "aggregateThroughputTokS": 3146.07, "perRequestTokS": 17.85, "ttftMs": 2075.47, "energyJoules": 3066.49, "avgPowerWatts": 300.08, "joulesPerToken": 0.0936 }, { "concurrency": 512, "aggregateThroughputTokS": 3202.06, "perRequestTokS": 16.04, "ttftMs": 6352.6, "energyJoules": 5980.282, "avgPowerWatts": 311.87, "joulesPerToken": 0.0913 }, { "concurrency": 1024, "aggregateThroughputTokS": 3292.3, "perRequestTokS": 15.3, "ttftMs": 14724.8, "energyJoules": 11567.679, "avgPowerWatts": 318.9, "joulesPerToken": 0.0883 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:54:53.559424+00:00", "notes": null, "run_key": "Qwen2.5-VL-7B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-VL-7B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-7B-Instruct-AWQ-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-7B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 5.2, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 109.51, "aggregateDecodeTokS": 3657.37, "perRequestDecodeTokS": 109.51, "ttftMs": 58.1, "acceptanceRate": null, "energyJoules": 330.399, "avgPowerWatts": 156.39, "joulesPerToken": 2.5812, "vramUsedGiB": 29.52, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 109.51, "perRequestTokS": 109.51, "ttftMs": 58.1, "energyJoules": 330.399, "avgPowerWatts": 156.39, "joulesPerToken": 2.5812 }, { "concurrency": 2, "aggregateThroughputTokS": 217.13, "perRequestTokS": 109.62, "ttftMs": 143.06, "energyJoules": 339.514, "avgPowerWatts": 160.96, "joulesPerToken": 1.3262 }, { "concurrency": 4, "aggregateThroughputTokS": 413.54, "perRequestTokS": 106.39, "ttftMs": 337.38, "energyJoules": 371.868, "avgPowerWatts": 155.38, "joulesPerToken": 0.7263 }, { "concurrency": 8, "aggregateThroughputTokS": 828.5, "perRequestTokS": 104.33, "ttftMs": 915.4, "energyJoules": 437.026, "avgPowerWatts": 147.9, "joulesPerToken": 0.4268 }, { "concurrency": 16, "aggregateThroughputTokS": 1319.09, "perRequestTokS": 93.9, "ttftMs": 211.9, "energyJoules": 483.611, "avgPowerWatts": 193.43, "joulesPerToken": 0.2361 }, { "concurrency": 32, "aggregateThroughputTokS": 1994.76, "perRequestTokS": 75.62, "ttftMs": 356.77, "energyJoules": 703.814, "avgPowerWatts": 220.51, "joulesPerToken": 0.1718 }, { "concurrency": 64, "aggregateThroughputTokS": 2811.69, "perRequestTokS": 55.87, "ttftMs": 585.8, "energyJoules": 976.204, "avgPowerWatts": 259.06, "joulesPerToken": 0.1192 }, { "concurrency": 128, "aggregateThroughputTokS": 3090.16, "perRequestTokS": 31.38, "ttftMs": 1071.5, "energyJoules": 1721.708, "avgPowerWatts": 279.22, "joulesPerToken": 0.1051 }, { "concurrency": 256, "aggregateThroughputTokS": 3657.37, "perRequestTokS": 19.78, "ttftMs": 2110.33, "energyJoules": 2965.46, "avgPowerWatts": 302.07, "joulesPerToken": 0.0905 }, { "concurrency": 512, "aggregateThroughputTokS": 3656.9, "perRequestTokS": 17.57, "ttftMs": 6487.02, "energyJoules": 5868.85, "avgPowerWatts": 312.9, "joulesPerToken": 0.0896 }, { "concurrency": 1024, "aggregateThroughputTokS": 3651.66, "perRequestTokS": 16.38, "ttftMs": 15030.1, "energyJoules": 11607.469, "avgPowerWatts": 323.01, "joulesPerToken": 0.0886 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:56:27.725677+00:00", "notes": null, "run_key": "Qwen2.5-7B-Instruct-AWQ-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-7B-Instruct-AWQ-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-Coder-7B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-Coder-7B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 14.2, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 38.56, "aggregateDecodeTokS": 2999.35, "perRequestDecodeTokS": 38.56, "ttftMs": 64.35, "acceptanceRate": null, "energyJoules": 636.14, "avgPowerWatts": 167.59, "joulesPerToken": 4.9698, "vramUsedGiB": 29.49, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 38.56, "perRequestTokS": 38.56, "ttftMs": 64.35, "energyJoules": 636.14, "avgPowerWatts": 167.59, "joulesPerToken": 4.9698 }, { "concurrency": 2, "aggregateThroughputTokS": 76.59, "perRequestTokS": 38.59, "ttftMs": 59.95, "energyJoules": 754.474, "avgPowerWatts": 169.43, "joulesPerToken": 2.9472 }, { "concurrency": 4, "aggregateThroughputTokS": 122.48, "perRequestTokS": 38.13, "ttftMs": 75.83, "energyJoules": 768.268, "avgPowerWatts": 178.78, "joulesPerToken": 1.5005 }, { "concurrency": 8, "aggregateThroughputTokS": 240.78, "perRequestTokS": 38.53, "ttftMs": 837.85, "energyJoules": 849.3, "avgPowerWatts": 168.84, "joulesPerToken": 0.8294 }, { "concurrency": 16, "aggregateThroughputTokS": 398.56, "perRequestTokS": 36.55, "ttftMs": 207.22, "energyJoules": 885.28, "avgPowerWatts": 198.62, "joulesPerToken": 0.4323 }, { "concurrency": 32, "aggregateThroughputTokS": 723.52, "perRequestTokS": 33.1, "ttftMs": 369.67, "energyJoules": 1033.172, "avgPowerWatts": 211.05, "joulesPerToken": 0.2522 }, { "concurrency": 64, "aggregateThroughputTokS": 1453.8, "perRequestTokS": 29.05, "ttftMs": 600.66, "energyJoules": 1362.039, "avgPowerWatts": 235.17, "joulesPerToken": 0.1663 }, { "concurrency": 128, "aggregateThroughputTokS": 2005.03, "perRequestTokS": 23.73, "ttftMs": 1079.7, "energyJoules": 1819.961, "avgPowerWatts": 259.64, "joulesPerToken": 0.1111 }, { "concurrency": 256, "aggregateThroughputTokS": 2826.59, "perRequestTokS": 17.07, "ttftMs": 2075.99, "energyJoules": 3002.966, "avgPowerWatts": 295.19, "joulesPerToken": 0.0916 }, { "concurrency": 512, "aggregateThroughputTokS": 2854.88, "perRequestTokS": 15.46, "ttftMs": 5626.82, "energyJoules": 5623.04, "avgPowerWatts": 305.56, "joulesPerToken": 0.0858 }, { "concurrency": 1024, "aggregateThroughputTokS": 2999.35, "perRequestTokS": 14.14, "ttftMs": 13149.65, "energyJoules": 10698.935, "avgPowerWatts": 317.19, "joulesPerToken": 0.0816 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:57:51.478905+00:00", "notes": null, "run_key": "Qwen2.5-Coder-7B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-Coder-7B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-8B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-8B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "8B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 15.27, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 36.38, "aggregateDecodeTokS": 2963.95, "perRequestDecodeTokS": 36.38, "ttftMs": 64.46, "acceptanceRate": null, "energyJoules": 767.284, "avgPowerWatts": 170.84, "joulesPerToken": 5.9944, "vramUsedGiB": 29.28, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 36.38, "perRequestTokS": 36.38, "ttftMs": 64.46, "energyJoules": 767.284, "avgPowerWatts": 170.84, "joulesPerToken": 5.9944 }, { "concurrency": 2, "aggregateThroughputTokS": 72.04, "perRequestTokS": 36.29, "ttftMs": 61.06, "energyJoules": 791.214, "avgPowerWatts": 177.74, "joulesPerToken": 3.0907 }, { "concurrency": 4, "aggregateThroughputTokS": 142.51, "perRequestTokS": 36.1, "ttftMs": 78.83, "energyJoules": 824.866, "avgPowerWatts": 182.85, "joulesPerToken": 1.6111 }, { "concurrency": 8, "aggregateThroughputTokS": 281.76, "perRequestTokS": 35.35, "ttftMs": 910.43, "energyJoules": 915.878, "avgPowerWatts": 165.86, "joulesPerToken": 0.8944 }, { "concurrency": 16, "aggregateThroughputTokS": 521.53, "perRequestTokS": 34.38, "ttftMs": 236.83, "energyJoules": 969.091, "avgPowerWatts": 201.66, "joulesPerToken": 0.4732 }, { "concurrency": 32, "aggregateThroughputTokS": 940.53, "perRequestTokS": 31.89, "ttftMs": 361.22, "energyJoules": 1167.551, "avgPowerWatts": 218.25, "joulesPerToken": 0.285 }, { "concurrency": 64, "aggregateThroughputTokS": 1510.89, "perRequestTokS": 26.74, "ttftMs": 610.95, "energyJoules": 1508.592, "avgPowerWatts": 240.19, "joulesPerToken": 0.1842 }, { "concurrency": 128, "aggregateThroughputTokS": 2330.48, "perRequestTokS": 22.14, "ttftMs": 1099, "energyJoules": 2146.263, "avgPowerWatts": 269.34, "joulesPerToken": 0.131 }, { "concurrency": 256, "aggregateThroughputTokS": 2938.83, "perRequestTokS": 14.78, "ttftMs": 2131.69, "energyJoules": 3695.781, "avgPowerWatts": 305.31, "joulesPerToken": 0.1128 }, { "concurrency": 512, "aggregateThroughputTokS": 2950.01, "perRequestTokS": 13.55, "ttftMs": 7571.32, "energyJoules": 7311.297, "avgPowerWatts": 317.35, "joulesPerToken": 0.1116 }, { "concurrency": 1024, "aggregateThroughputTokS": 2963.95, "perRequestTokS": 12.92, "ttftMs": 18278.98, "energyJoules": 14429.818, "avgPowerWatts": 322.7, "joulesPerToken": 0.1101 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T14:58:18.641350+00:00", "notes": null, "run_key": "Qwen3-8B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3-8B-BF16-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-VL-8B-Instruct-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-VL-8B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "8B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 32.68, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 36.31, "aggregateDecodeTokS": 2942.84, "perRequestDecodeTokS": 36.31, "ttftMs": 68.97, "acceptanceRate": null, "energyJoules": 789.445, "avgPowerWatts": 173.12, "joulesPerToken": 6.1675, "vramUsedGiB": 27.96, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 36.31, "perRequestTokS": 36.31, "ttftMs": 68.97, "energyJoules": 789.445, "avgPowerWatts": 173.12, "joulesPerToken": 6.1675 }, { "concurrency": 2, "aggregateThroughputTokS": 72.25, "perRequestTokS": 36.38, "ttftMs": 96.88, "energyJoules": 796.408, "avgPowerWatts": 179.11, "joulesPerToken": 3.111 }, { "concurrency": 4, "aggregateThroughputTokS": 143, "perRequestTokS": 36.08, "ttftMs": 81.39, "energyJoules": 820.769, "avgPowerWatts": 185.07, "joulesPerToken": 1.6031 }, { "concurrency": 8, "aggregateThroughputTokS": 281.18, "perRequestTokS": 35.28, "ttftMs": 892.17, "energyJoules": 911.249, "avgPowerWatts": 169.23, "joulesPerToken": 0.8899 }, { "concurrency": 16, "aggregateThroughputTokS": 520.66, "perRequestTokS": 34.31, "ttftMs": 236.7, "energyJoules": 971.199, "avgPowerWatts": 201.8, "joulesPerToken": 0.4742 }, { "concurrency": 32, "aggregateThroughputTokS": 938.4, "perRequestTokS": 31.8, "ttftMs": 358.25, "energyJoules": 1164.492, "avgPowerWatts": 216.94, "joulesPerToken": 0.2843 }, { "concurrency": 64, "aggregateThroughputTokS": 1507.08, "perRequestTokS": 26.68, "ttftMs": 614.41, "energyJoules": 1512.437, "avgPowerWatts": 239.07, "joulesPerToken": 0.1846 }, { "concurrency": 128, "aggregateThroughputTokS": 2312.22, "perRequestTokS": 22.12, "ttftMs": 1153.79, "energyJoules": 2157.264, "avgPowerWatts": 269.35, "joulesPerToken": 0.1317 }, { "concurrency": 256, "aggregateThroughputTokS": 2632.86, "perRequestTokS": 14.65, "ttftMs": 2133.96, "energyJoules": 3982.749, "avgPowerWatts": 298.04, "joulesPerToken": 0.1215 }, { "concurrency": 512, "aggregateThroughputTokS": 2779.15, "perRequestTokS": 13.35, "ttftMs": 7589.26, "energyJoules": 7624.523, "avgPowerWatts": 311.47, "joulesPerToken": 0.1163 }, { "concurrency": 1024, "aggregateThroughputTokS": 2942.84, "perRequestTokS": 12.83, "ttftMs": 18344.97, "energyJoules": 14520.129, "avgPowerWatts": 323.58, "joulesPerToken": 0.1108 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:00:09.865288+00:00", "notes": null, "run_key": "Qwen3-VL-8B-Instruct-vllm-xpu-tp1-baseline", "run_id": "Qwen3-VL-8B-Instruct-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-8B-AWQ-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-8B-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "8B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 5.69, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 100.04, "aggregateDecodeTokS": 3242.01, "perRequestDecodeTokS": 100.04, "ttftMs": 70.21, "acceptanceRate": null, "energyJoules": 362.502, "avgPowerWatts": 146.58, "joulesPerToken": 2.832, "vramUsedGiB": 29.28, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 100.04, "perRequestTokS": 100.04, "ttftMs": 70.21, "energyJoules": 362.502, "avgPowerWatts": 146.58, "joulesPerToken": 2.832 }, { "concurrency": 2, "aggregateThroughputTokS": 198.02, "perRequestTokS": 100.21, "ttftMs": 249.76, "energyJoules": 367.754, "avgPowerWatts": 151.04, "joulesPerToken": 1.4365 }, { "concurrency": 4, "aggregateThroughputTokS": 381.13, "perRequestTokS": 97.35, "ttftMs": 125.45, "energyJoules": 385.497, "avgPowerWatts": 167.23, "joulesPerToken": 0.7529 }, { "concurrency": 8, "aggregateThroughputTokS": 731.91, "perRequestTokS": 92.18, "ttftMs": 935.34, "energyJoules": 468.575, "avgPowerWatts": 147.21, "joulesPerToken": 0.4576 }, { "concurrency": 16, "aggregateThroughputTokS": 1201.58, "perRequestTokS": 85.53, "ttftMs": 229.15, "energyJoules": 514.883, "avgPowerWatts": 198.32, "joulesPerToken": 0.2514 }, { "concurrency": 32, "aggregateThroughputTokS": 1826.59, "perRequestTokS": 67.3, "ttftMs": 352.09, "energyJoules": 749.75, "avgPowerWatts": 240.64, "joulesPerToken": 0.183 }, { "concurrency": 64, "aggregateThroughputTokS": 2485.19, "perRequestTokS": 48.12, "ttftMs": 613.38, "energyJoules": 1058.75, "avgPowerWatts": 256.33, "joulesPerToken": 0.1292 }, { "concurrency": 128, "aggregateThroughputTokS": 2922.95, "perRequestTokS": 29.26, "ttftMs": 1125.27, "energyJoules": 1806.979, "avgPowerWatts": 276.08, "joulesPerToken": 0.1103 }, { "concurrency": 256, "aggregateThroughputTokS": 3242.01, "perRequestTokS": 16.93, "ttftMs": 2209.05, "energyJoules": 3342.947, "avgPowerWatts": 303.16, "joulesPerToken": 0.102 }, { "concurrency": 512, "aggregateThroughputTokS": 3226.24, "perRequestTokS": 15.16, "ttftMs": 7202.97, "energyJoules": 6718.107, "avgPowerWatts": 315.99, "joulesPerToken": 0.1025 }, { "concurrency": 1024, "aggregateThroughputTokS": 3203.07, "perRequestTokS": 14.15, "ttftMs": 17202.16, "energyJoules": 13382.92, "avgPowerWatts": 323.46, "joulesPerToken": 0.1021 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:00:43.270497+00:00", "notes": null, "run_key": "Qwen3-8B-AWQ-vllm-xpu-tp1-baseline", "run_id": "Qwen3-8B-AWQ-vllm-xpu-tp1-baseline" }, { "id": "Qwen3.5-9B-BF16-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.5-9B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "9B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 36, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 34.85, "aggregateDecodeTokS": 1475.38, "perRequestDecodeTokS": 34.85, "ttftMs": 102.55, "acceptanceRate": null, "energyJoules": 815.596, "avgPowerWatts": 169.38, "joulesPerToken": 6.3718, "vramUsedGiB": 27.9, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 34.85, "perRequestTokS": 34.85, "ttftMs": 102.55, "energyJoules": 815.596, "avgPowerWatts": 169.38, "joulesPerToken": 6.3718 }, { "concurrency": 2, "aggregateThroughputTokS": 67.88, "perRequestTokS": 34.29, "ttftMs": 79.12, "energyJoules": 834.924, "avgPowerWatts": 175.26, "joulesPerToken": 3.2614 }, { "concurrency": 4, "aggregateThroughputTokS": 132.45, "perRequestTokS": 33.6, "ttftMs": 103.23, "energyJoules": 879.086, "avgPowerWatts": 180.42, "joulesPerToken": 1.717 }, { "concurrency": 8, "aggregateThroughputTokS": 244.54, "perRequestTokS": 31.35, "ttftMs": 155.92, "energyJoules": 967.898, "avgPowerWatts": 185, "joulesPerToken": 0.9452 }, { "concurrency": 16, "aggregateThroughputTokS": 460.75, "perRequestTokS": 30.32, "ttftMs": 272.85, "energyJoules": 1105.515, "avgPowerWatts": 204.47, "joulesPerToken": 0.5398 }, { "concurrency": 32, "aggregateThroughputTokS": 775.35, "perRequestTokS": 26.21, "ttftMs": 405.69, "energyJoules": 1413.85, "avgPowerWatts": 226.22, "joulesPerToken": 0.3452 }, { "concurrency": 64, "aggregateThroughputTokS": 1124.38, "perRequestTokS": 19.58, "ttftMs": 704.29, "energyJoules": 1986.771, "avgPowerWatts": 240.21, "joulesPerToken": 0.2425 }, { "concurrency": 128, "aggregateThroughputTokS": 1178.02, "perRequestTokS": 17.71, "ttftMs": 2549.77, "energyJoules": 3776.719, "avgPowerWatts": 252.96, "joulesPerToken": 0.2305 }, { "concurrency": 256, "aggregateThroughputTokS": 1355.45, "perRequestTokS": 16.54, "ttftMs": 7877.48, "energyJoules": 6762.497, "avgPowerWatts": 269.94, "joulesPerToken": 0.2064 }, { "concurrency": 512, "aggregateThroughputTokS": 1475.38, "perRequestTokS": 15.52, "ttftMs": 18525.4, "energyJoules": 12628.926, "avgPowerWatts": 279.29, "joulesPerToken": 0.1927 }, { "concurrency": 1024, "aggregateThroughputTokS": 1450.99, "perRequestTokS": 15.2, "ttftMs": 40440.79, "energyJoules": 25347.151, "avgPowerWatts": 281.24, "joulesPerToken": 0.1934 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:02:36.271816+00:00", "notes": null, "run_key": "Qwen3.5-9B-BF16-vllm-xpu-tp1-baseline", "run_id": "Qwen3.5-9B-BF16-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-14B-Instruct-AWQ-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-14B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 9.31, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 59.44, "aggregateDecodeTokS": 1760.09, "perRequestDecodeTokS": 59.44, "ttftMs": 87.72, "acceptanceRate": null, "energyJoules": 575.705, "avgPowerWatts": 182.21, "joulesPerToken": 4.4977, "vramUsedGiB": 28.99, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 59.44, "perRequestTokS": 59.44, "ttftMs": 87.72, "energyJoules": 575.705, "avgPowerWatts": 182.21, "joulesPerToken": 4.4977 }, { "concurrency": 2, "aggregateThroughputTokS": 105.18, "perRequestTokS": 59.56, "ttftMs": 207.71, "energyJoules": 596.932, "avgPowerWatts": 184.73, "joulesPerToken": 2.3318 }, { "concurrency": 4, "aggregateThroughputTokS": 219.31, "perRequestTokS": 58.18, "ttftMs": 238.57, "energyJoules": 638.193, "avgPowerWatts": 194.51, "joulesPerToken": 1.2465 }, { "concurrency": 8, "aggregateThroughputTokS": 436.62, "perRequestTokS": 55.69, "ttftMs": 838.37, "energyJoules": 745.171, "avgPowerWatts": 188.06, "joulesPerToken": 0.7277 }, { "concurrency": 16, "aggregateThroughputTokS": 648.94, "perRequestTokS": 50.22, "ttftMs": 424.81, "energyJoules": 930.359, "avgPowerWatts": 238.49, "joulesPerToken": 0.4543 }, { "concurrency": 32, "aggregateThroughputTokS": 1051.33, "perRequestTokS": 40.71, "ttftMs": 671.14, "energyJoules": 1312.776, "avgPowerWatts": 279.85, "joulesPerToken": 0.3205 }, { "concurrency": 64, "aggregateThroughputTokS": 1423.38, "perRequestTokS": 29.76, "ttftMs": 1161.16, "energyJoules": 1865.736, "avgPowerWatts": 293.02, "joulesPerToken": 0.2278 }, { "concurrency": 128, "aggregateThroughputTokS": 1533.77, "perRequestTokS": 15.91, "ttftMs": 2100.82, "energyJoules": 3334.805, "avgPowerWatts": 298.23, "joulesPerToken": 0.2035 }, { "concurrency": 256, "aggregateThroughputTokS": 1760.09, "perRequestTokS": 9.75, "ttftMs": 4113.05, "energyJoules": 5834.157, "avgPowerWatts": 317.12, "joulesPerToken": 0.178 }, { "concurrency": 512, "aggregateThroughputTokS": 1751.87, "perRequestTokS": 8.58, "ttftMs": 12667.75, "energyJoules": 11653.04, "avgPowerWatts": 323.05, "joulesPerToken": 0.1778 }, { "concurrency": 1024, "aggregateThroughputTokS": 1747.19, "perRequestTokS": 7.94, "ttftMs": 30090.76, "energyJoules": 23336.195, "avgPowerWatts": 325.99, "joulesPerToken": 0.178 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:03:32.529831+00:00", "notes": null, "run_key": "Qwen2.5-14B-Instruct-AWQ-vllm-xpu-tp1-baseline", "run_id": "Qwen2.5-14B-Instruct-AWQ-vllm-xpu-tp1-baseline" }, { "id": "Qwen3-30B-A3B-GPTQ-Int4-vllm-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-30B-A3B-GPTQ-Int4", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "30B", "quantFormat": "GPTQ", "quantBits": 4, "checkpointDiskGiB": 15.79, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 79.44, "aggregateDecodeTokS": 3295.94, "perRequestDecodeTokS": 79.44, "ttftMs": 92.43, "acceptanceRate": null, "energyJoules": 307.422, "avgPowerWatts": 111.55, "joulesPerToken": 2.4017, "vramUsedGiB": 29.47, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 79.44, "perRequestTokS": 79.44, "ttftMs": 92.43, "energyJoules": 307.422, "avgPowerWatts": 111.55, "joulesPerToken": 2.4017 }, { "concurrency": 2, "aggregateThroughputTokS": 155.39, "perRequestTokS": 78.08, "ttftMs": 130.67, "energyJoules": 333.314, "avgPowerWatts": 128.78, "joulesPerToken": 1.302 }, { "concurrency": 4, "aggregateThroughputTokS": 283.59, "perRequestTokS": 72.11, "ttftMs": 81.86, "energyJoules": 392.837, "avgPowerWatts": 145.25, "joulesPerToken": 0.7673 }, { "concurrency": 8, "aggregateThroughputTokS": 562.93, "perRequestTokS": 73.13, "ttftMs": 126.25, "energyJoules": 482.584, "avgPowerWatts": 163.02, "joulesPerToken": 0.4713 }, { "concurrency": 16, "aggregateThroughputTokS": 899.49, "perRequestTokS": 60.52, "ttftMs": 198.26, "energyJoules": 611.909, "avgPowerWatts": 194.93, "joulesPerToken": 0.2988 }, { "concurrency": 32, "aggregateThroughputTokS": 1362.7, "perRequestTokS": 46.76, "ttftMs": 293.21, "energyJoules": 861.594, "avgPowerWatts": 219.87, "joulesPerToken": 0.2104 }, { "concurrency": 64, "aggregateThroughputTokS": 1913.46, "perRequestTokS": 33.82, "ttftMs": 493.77, "energyJoules": 1288.429, "avgPowerWatts": 248.28, "joulesPerToken": 0.1573 }, { "concurrency": 128, "aggregateThroughputTokS": 2685.09, "perRequestTokS": 24.98, "ttftMs": 888.92, "energyJoules": 1911.965, "avgPowerWatts": 271.33, "joulesPerToken": 0.1167 }, { "concurrency": 256, "aggregateThroughputTokS": 3260.43, "perRequestTokS": 15.9, "ttftMs": 1726.8, "energyJoules": 3008.927, "avgPowerWatts": 276.68, "joulesPerToken": 0.0918 }, { "concurrency": 512, "aggregateThroughputTokS": 3277.61, "perRequestTokS": 14.75, "ttftMs": 6634.86, "energyJoules": 6100.177, "avgPowerWatts": 292.37, "joulesPerToken": 0.0931 }, { "concurrency": 1024, "aggregateThroughputTokS": 3295.94, "perRequestTokS": 14.19, "ttftMs": 16325.72, "energyJoules": 12027.376, "avgPowerWatts": 298.21, "joulesPerToken": 0.0918 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:07:47.496567+00:00", "notes": null, "run_key": "Qwen3-30B-A3B-GPTQ-Int4-vllm-xpu-tp1-baseline", "run_id": "Qwen3-30B-A3B-GPTQ-Int4-vllm-xpu-tp1-baseline" }, { "id": "Qwen2.5-0.5B-Instruct-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-0.5B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.5B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 0.93, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 103.47, "aggregateDecodeTokS": 2164.48, "perRequestDecodeTokS": 103.47, "ttftMs": 46.56, "acceptanceRate": null, "energyJoules": 327.299, "avgPowerWatts": 74.31, "joulesPerToken": 2.557, "vramUsedGiB": 30.69, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 103.47, "perRequestTokS": 103.47, "ttftMs": 46.56, "energyJoules": 327.299, "avgPowerWatts": 74.31, "joulesPerToken": 2.557 }, { "concurrency": 2, "aggregateThroughputTokS": 131.73, "perRequestTokS": 121.09, "ttftMs": 41.32, "energyJoules": 183.638, "avgPowerWatts": 88.2, "joulesPerToken": 0.7173 }, { "concurrency": 4, "aggregateThroughputTokS": 302.5, "perRequestTokS": 101.64, "ttftMs": 81.06, "energyJoules": 200.684, "avgPowerWatts": 93.63, "joulesPerToken": 0.392 }, { "concurrency": 8, "aggregateThroughputTokS": 584.31, "perRequestTokS": 101.99, "ttftMs": 101.54, "energyJoules": 217.878, "avgPowerWatts": 99.12, "joulesPerToken": 0.2128 }, { "concurrency": 16, "aggregateThroughputTokS": 853.62, "perRequestTokS": 95.17, "ttftMs": 177.47, "energyJoules": 261.079, "avgPowerWatts": 105.19, "joulesPerToken": 0.1275 }, { "concurrency": 32, "aggregateThroughputTokS": 1582.27, "perRequestTokS": 82.34, "ttftMs": 246.92, "energyJoules": 326.966, "avgPowerWatts": 122.73, "joulesPerToken": 0.0798 }, { "concurrency": 64, "aggregateThroughputTokS": 2164.48, "perRequestTokS": 54.39, "ttftMs": 385.67, "energyJoules": 463.585, "avgPowerWatts": 143.17, "joulesPerToken": 0.0566 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 513.02, "avgPowerWatts": 70.97, "joulesPerToken": 0.0313 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:09:57.786231+00:00", "notes": null, "run_key": "Qwen2.5-0.5B-Instruct-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-0.5B-Instruct-sglang-xpu-tp1-baseline" }, { "id": "Qwen3-0.6B-BF16-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-0.6B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "0.6B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 1.41, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 86.3, "aggregateDecodeTokS": 1011.92, "perRequestDecodeTokS": 86.3, "ttftMs": 70.11, "acceptanceRate": null, "energyJoules": 313.312, "avgPowerWatts": 85.86, "joulesPerToken": 2.4478, "vramUsedGiB": 30.82, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 86.3, "perRequestTokS": 86.3, "ttftMs": 70.11, "energyJoules": 313.312, "avgPowerWatts": 85.86, "joulesPerToken": 2.4478 }, { "concurrency": 2, "aggregateThroughputTokS": 169.67, "perRequestTokS": 85.16, "ttftMs": 74.6, "energyJoules": 257.836, "avgPowerWatts": 98.92, "joulesPerToken": 1.0072 }, { "concurrency": 4, "aggregateThroughputTokS": 341.12, "perRequestTokS": 85.82, "ttftMs": 178.36, "energyJoules": 285.557, "avgPowerWatts": 110.52, "joulesPerToken": 0.5577 }, { "concurrency": 8, "aggregateThroughputTokS": 550.33, "perRequestTokS": 69.29, "ttftMs": 206.61, "energyJoules": 371.913, "avgPowerWatts": 126.75, "joulesPerToken": 0.3632 }, { "concurrency": 16, "aggregateThroughputTokS": 731.58, "perRequestTokS": 46.59, "ttftMs": 435.25, "energyJoules": 574.746, "avgPowerWatts": 135.41, "joulesPerToken": 0.2806 }, { "concurrency": 32, "aggregateThroughputTokS": 890.7, "perRequestTokS": 29.73, "ttftMs": 634.73, "energyJoules": 915.606, "avgPowerWatts": 155.43, "joulesPerToken": 0.2235 }, { "concurrency": 64, "aggregateThroughputTokS": 979.69, "perRequestTokS": 16.73, "ttftMs": 1040.74, "energyJoules": 1622.632, "avgPowerWatts": 169.51, "joulesPerToken": 0.1981 }, { "concurrency": 128, "aggregateThroughputTokS": 1011.92, "perRequestTokS": 8.66, "ttftMs": 1711.67, "energyJoules": 3107.579, "avgPowerWatts": 179.47, "joulesPerToken": 0.1897 }, { "concurrency": 256, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 499.539, "avgPowerWatts": 69.84, "joulesPerToken": 0.0152 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:09:58.031339+00:00", "notes": null, "run_key": "Qwen3-0.6B-BF16-sglang-xpu-tp1-baseline", "run_id": "Qwen3-0.6B-BF16-sglang-xpu-tp1-baseline" }, { "id": "DeepSeek-R1-Distill-Qwen-1.5B-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "DeepSeek-R1-Distill-Qwen-1.5B", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "1.5B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 3.32, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 94.43, "aggregateDecodeTokS": 1976.67, "perRequestDecodeTokS": 94.43, "ttftMs": 74.13, "acceptanceRate": null, "energyJoules": 316.624, "avgPowerWatts": 106.35, "joulesPerToken": 2.4736, "vramUsedGiB": 30.77, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 94.43, "perRequestTokS": 94.43, "ttftMs": 74.13, "energyJoules": 316.624, "avgPowerWatts": 106.35, "joulesPerToken": 2.4736 }, { "concurrency": 2, "aggregateThroughputTokS": 184.27, "perRequestTokS": 92.53, "ttftMs": 75.96, "energyJoules": 272.2, "avgPowerWatts": 115.47, "joulesPerToken": 1.0633 }, { "concurrency": 4, "aggregateThroughputTokS": 300.01, "perRequestTokS": 75.46, "ttftMs": 216.2, "energyJoules": 325.765, "avgPowerWatts": 113.58, "joulesPerToken": 0.6363 }, { "concurrency": 8, "aggregateThroughputTokS": 567.67, "perRequestTokS": 71.49, "ttftMs": 197.9, "energyJoules": 373.279, "avgPowerWatts": 120.37, "joulesPerToken": 0.3645 }, { "concurrency": 16, "aggregateThroughputTokS": 970.84, "perRequestTokS": 61.23, "ttftMs": 334.08, "energyJoules": 460.731, "avgPowerWatts": 137.31, "joulesPerToken": 0.225 }, { "concurrency": 32, "aggregateThroughputTokS": 1394.78, "perRequestTokS": 47.71, "ttftMs": 583.65, "energyJoules": 657.461, "avgPowerWatts": 156.68, "joulesPerToken": 0.1605 }, { "concurrency": 64, "aggregateThroughputTokS": 1779.79, "perRequestTokS": 32.09, "ttftMs": 901.87, "energyJoules": 1031.724, "avgPowerWatts": 176.32, "joulesPerToken": 0.1259 }, { "concurrency": 128, "aggregateThroughputTokS": 1976.67, "perRequestTokS": 18.16, "ttftMs": 1488.9, "energyJoules": 1814.546, "avgPowerWatts": 192.06, "joulesPerToken": 0.1108 }, { "concurrency": 256, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1501.207, "avgPowerWatts": 108.17, "joulesPerToken": 0.0458 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:10:45.346232+00:00", "notes": null, "run_key": "DeepSeek-R1-Distill-Qwen-1.5B-sglang-xpu-tp1-baseline", "run_id": "DeepSeek-R1-Distill-Qwen-1.5B-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-3B-Instruct-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-3B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "3B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 5.76, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 60.89, "aggregateDecodeTokS": 1260.65, "perRequestDecodeTokS": 60.89, "ttftMs": 96.03, "acceptanceRate": null, "energyJoules": 514.997, "avgPowerWatts": 113.44, "joulesPerToken": 4.0234, "vramUsedGiB": 30.58, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 60.89, "perRequestTokS": 60.89, "ttftMs": 96.03, "energyJoules": 514.997, "avgPowerWatts": 113.44, "joulesPerToken": 4.0234 }, { "concurrency": 2, "aggregateThroughputTokS": 100.39, "perRequestTokS": 62.64, "ttftMs": 110.74, "energyJoules": 423.378, "avgPowerWatts": 137.77, "joulesPerToken": 1.6538 }, { "concurrency": 4, "aggregateThroughputTokS": 194.58, "perRequestTokS": 48.91, "ttftMs": 274.49, "energyJoules": 508.864, "avgPowerWatts": 134.61, "joulesPerToken": 0.9939 }, { "concurrency": 8, "aggregateThroughputTokS": 384.07, "perRequestTokS": 49.46, "ttftMs": 320.16, "energyJoules": 560.493, "avgPowerWatts": 143.44, "joulesPerToken": 0.5474 }, { "concurrency": 16, "aggregateThroughputTokS": 681.41, "perRequestTokS": 43.55, "ttftMs": 528.59, "energyJoules": 696.625, "avgPowerWatts": 156.38, "joulesPerToken": 0.3401 }, { "concurrency": 32, "aggregateThroughputTokS": 993.05, "perRequestTokS": 35.1, "ttftMs": 911.39, "energyJoules": 964.072, "avgPowerWatts": 175.62, "joulesPerToken": 0.2354 }, { "concurrency": 64, "aggregateThroughputTokS": 1260.65, "perRequestTokS": 23.67, "ttftMs": 1512.33, "energyJoules": 1493.671, "avgPowerWatts": 191.53, "joulesPerToken": 0.1823 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1069.843, "avgPowerWatts": 109.16, "joulesPerToken": 0.0653 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:11:24.296513+00:00", "notes": null, "run_key": "Qwen2.5-3B-Instruct-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-3B-Instruct-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-3B-Instruct-AWQ-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-3B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "3B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 2.51, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 74.25, "aggregateDecodeTokS": 1336.76, "perRequestDecodeTokS": 74.25, "ttftMs": 90.34, "acceptanceRate": null, "energyJoules": 372.332, "avgPowerWatts": 91.09, "joulesPerToken": 2.9088, "vramUsedGiB": 30.82, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 74.25, "perRequestTokS": 74.25, "ttftMs": 90.34, "energyJoules": 372.332, "avgPowerWatts": 91.09, "joulesPerToken": 2.9088 }, { "concurrency": 2, "aggregateThroughputTokS": 160.64, "perRequestTokS": 80.62, "ttftMs": 98.83, "energyJoules": 294.802, "avgPowerWatts": 115.97, "joulesPerToken": 1.1516 }, { "concurrency": 4, "aggregateThroughputTokS": 275.74, "perRequestTokS": 69.4, "ttftMs": 320.09, "energyJoules": 353.114, "avgPowerWatts": 115.62, "joulesPerToken": 0.6897 }, { "concurrency": 8, "aggregateThroughputTokS": 591.36, "perRequestTokS": 74.6, "ttftMs": 322.49, "energyJoules": 384.063, "avgPowerWatts": 128.9, "joulesPerToken": 0.3751 }, { "concurrency": 16, "aggregateThroughputTokS": 1006.4, "perRequestTokS": 69.65, "ttftMs": 672.08, "energyJoules": 501.701, "avgPowerWatts": 148.88, "joulesPerToken": 0.245 }, { "concurrency": 32, "aggregateThroughputTokS": 1134.94, "perRequestTokS": 43.1, "ttftMs": 1127.35, "energyJoules": 835.191, "avgPowerWatts": 166.43, "joulesPerToken": 0.2039 }, { "concurrency": 64, "aggregateThroughputTokS": 1336.76, "perRequestTokS": 26.11, "ttftMs": 1588.59, "energyJoules": 1371.926, "avgPowerWatts": 184.34, "joulesPerToken": 0.1675 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1094.841, "avgPowerWatts": 105.21, "joulesPerToken": 0.0668 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:11:58.721346+00:00", "notes": null, "run_key": "Qwen2.5-3B-Instruct-AWQ-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-3B-Instruct-AWQ-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-7B-Instruct-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-7B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 28.39, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 33.52, "aggregateDecodeTokS": 826.67, "perRequestDecodeTokS": 33.52, "ttftMs": 147.31, "acceptanceRate": null, "energyJoules": 919.344, "avgPowerWatts": 141.66, "joulesPerToken": 7.1824, "vramUsedGiB": 30.59, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 33.52, "perRequestTokS": 33.52, "ttftMs": 147.31, "energyJoules": 919.344, "avgPowerWatts": 141.66, "joulesPerToken": 7.1824 }, { "concurrency": 2, "aggregateThroughputTokS": 63.14, "perRequestTokS": 31.69, "ttftMs": 160.91, "energyJoules": 856.693, "avgPowerWatts": 164.07, "joulesPerToken": 3.3465 }, { "concurrency": 4, "aggregateThroughputTokS": 109.01, "perRequestTokS": 32.41, "ttftMs": 300.03, "energyJoules": 878.675, "avgPowerWatts": 169.72, "joulesPerToken": 1.7162 }, { "concurrency": 8, "aggregateThroughputTokS": 228.03, "perRequestTokS": 29.77, "ttftMs": 441.72, "energyJoules": 1006.153, "avgPowerWatts": 175.31, "joulesPerToken": 0.9826 }, { "concurrency": 16, "aggregateThroughputTokS": 409.43, "perRequestTokS": 26.7, "ttftMs": 732.02, "energyJoules": 1221.149, "avgPowerWatts": 187.23, "joulesPerToken": 0.5963 }, { "concurrency": 32, "aggregateThroughputTokS": 613.88, "perRequestTokS": 20.8, "ttftMs": 1330.58, "energyJoules": 1684.927, "avgPowerWatts": 199.09, "joulesPerToken": 0.4114 }, { "concurrency": 64, "aggregateThroughputTokS": 826.67, "perRequestTokS": 15.14, "ttftMs": 2223.08, "energyJoules": 2487.532, "avgPowerWatts": 213.23, "joulesPerToken": 0.3037 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1602.737, "avgPowerWatts": 139.3, "joulesPerToken": 0.0978 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:12:34.204302+00:00", "notes": null, "run_key": "Qwen2.5-7B-Instruct-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-7B-Instruct-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-7B-Instruct-AWQ-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-7B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 5.2, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 71.56, "aggregateDecodeTokS": 1006.78, "perRequestDecodeTokS": 71.56, "ttftMs": 92.52, "acceptanceRate": null, "energyJoules": 512.896, "avgPowerWatts": 108.24, "joulesPerToken": 4.007, "vramUsedGiB": 31.15, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 71.56, "perRequestTokS": 71.56, "ttftMs": 92.52, "energyJoules": 512.896, "avgPowerWatts": 108.24, "joulesPerToken": 4.007 }, { "concurrency": 2, "aggregateThroughputTokS": 132.58, "perRequestTokS": 66.56, "ttftMs": 143.78, "energyJoules": 433.792, "avgPowerWatts": 144.71, "joulesPerToken": 1.6945 }, { "concurrency": 4, "aggregateThroughputTokS": 277.16, "perRequestTokS": 69.72, "ttftMs": 381.17, "energyJoules": 473.105, "avgPowerWatts": 153.28, "joulesPerToken": 0.924 }, { "concurrency": 8, "aggregateThroughputTokS": 471.44, "perRequestTokS": 61.88, "ttftMs": 407.42, "energyJoules": 557.452, "avgPowerWatts": 168.44, "joulesPerToken": 0.5444 }, { "concurrency": 16, "aggregateThroughputTokS": 728.41, "perRequestTokS": 46.84, "ttftMs": 824.95, "energyJoules": 798.249, "avgPowerWatts": 179.9, "joulesPerToken": 0.3898 }, { "concurrency": 32, "aggregateThroughputTokS": 925.53, "perRequestTokS": 33.17, "ttftMs": 1469.01, "energyJoules": 1237.293, "avgPowerWatts": 199.93, "joulesPerToken": 0.3021 }, { "concurrency": 64, "aggregateThroughputTokS": 1006.78, "perRequestTokS": 19.56, "ttftMs": 2341.12, "energyJoules": 2000.628, "avgPowerWatts": 201.51, "joulesPerToken": 0.2442 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1522.728, "avgPowerWatts": 132.68, "joulesPerToken": 0.0929 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:13:04.079222+00:00", "notes": null, "run_key": "Qwen2.5-7B-Instruct-AWQ-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-7B-Instruct-AWQ-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-Coder-7B-Instruct-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-Coder-7B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "7B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 14.2, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 32.99, "aggregateDecodeTokS": 653.26, "perRequestDecodeTokS": 32.99, "ttftMs": 145.86, "acceptanceRate": null, "energyJoules": 909.868, "avgPowerWatts": 144.3, "joulesPerToken": 7.1083, "vramUsedGiB": 30.6, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 32.99, "perRequestTokS": 32.99, "ttftMs": 145.86, "energyJoules": 909.868, "avgPowerWatts": 144.3, "joulesPerToken": 7.1083 }, { "concurrency": 2, "aggregateThroughputTokS": 63.83, "perRequestTokS": 32.04, "ttftMs": 162.34, "energyJoules": 837.41, "avgPowerWatts": 165.14, "joulesPerToken": 3.2711 }, { "concurrency": 4, "aggregateThroughputTokS": 118.99, "perRequestTokS": 31.73, "ttftMs": 327.53, "energyJoules": 889.131, "avgPowerWatts": 167.48, "joulesPerToken": 1.7366 }, { "concurrency": 8, "aggregateThroughputTokS": 168.93, "perRequestTokS": 31.14, "ttftMs": 440, "energyJoules": 948.677, "avgPowerWatts": 173.18, "joulesPerToken": 0.9264 }, { "concurrency": 16, "aggregateThroughputTokS": 301.68, "perRequestTokS": 28.59, "ttftMs": 717.82, "energyJoules": 1120.081, "avgPowerWatts": 182.56, "joulesPerToken": 0.5469 }, { "concurrency": 32, "aggregateThroughputTokS": 505.36, "perRequestTokS": 24.02, "ttftMs": 1304.64, "energyJoules": 1482.972, "avgPowerWatts": 196.99, "joulesPerToken": 0.3621 }, { "concurrency": 64, "aggregateThroughputTokS": 653.26, "perRequestTokS": 16.33, "ttftMs": 2217.27, "energyJoules": 2166.756, "avgPowerWatts": 209.61, "joulesPerToken": 0.2645 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 1596.451, "avgPowerWatts": 140.52, "joulesPerToken": 0.0974 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:13:36.305076+00:00", "notes": null, "run_key": "Qwen2.5-Coder-7B-Instruct-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-Coder-7B-Instruct-sglang-xpu-tp1-baseline" }, { "id": "Qwen3-8B-BF16-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-8B-BF16", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "8B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 15.27, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 29.27, "aggregateDecodeTokS": 608.37, "perRequestDecodeTokS": 29.27, "ttftMs": 262.34, "acceptanceRate": null, "energyJoules": 985.415, "avgPowerWatts": 147.9, "joulesPerToken": 7.6986, "vramUsedGiB": 30.51, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 29.27, "perRequestTokS": 29.27, "ttftMs": 262.34, "energyJoules": 985.415, "avgPowerWatts": 147.9, "joulesPerToken": 7.6986 }, { "concurrency": 2, "aggregateThroughputTokS": 58.13, "perRequestTokS": 29.18, "ttftMs": 244.3, "energyJoules": 937.467, "avgPowerWatts": 167.4, "joulesPerToken": 3.662 }, { "concurrency": 4, "aggregateThroughputTokS": 110.07, "perRequestTokS": 27.68, "ttftMs": 375.74, "energyJoules": 1040.888, "avgPowerWatts": 168.26, "joulesPerToken": 2.033 }, { "concurrency": 8, "aggregateThroughputTokS": 194.47, "perRequestTokS": 24.47, "ttftMs": 567.02, "energyJoules": 1212.124, "avgPowerWatts": 179.33, "joulesPerToken": 1.1837 }, { "concurrency": 16, "aggregateThroughputTokS": 318.9, "perRequestTokS": 20.2, "ttftMs": 1009.87, "energyJoules": 1552.888, "avgPowerWatts": 186.73, "joulesPerToken": 0.7582 }, { "concurrency": 32, "aggregateThroughputTokS": 448.66, "perRequestTokS": 15.42, "ttftMs": 1824.09, "energyJoules": 2222.445, "avgPowerWatts": 199.45, "joulesPerToken": 0.5426 }, { "concurrency": 64, "aggregateThroughputTokS": 542.47, "perRequestTokS": 9.8, "ttftMs": 2989.67, "energyJoules": 3529.733, "avgPowerWatts": 207.52, "joulesPerToken": 0.4309 }, { "concurrency": 128, "aggregateThroughputTokS": 608.37, "perRequestTokS": 5.6, "ttftMs": 4887.39, "energyJoules": 6150.978, "avgPowerWatts": 212.47, "joulesPerToken": 0.3754 }, { "concurrency": 256, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 3755.499, "avgPowerWatts": 179.06, "joulesPerToken": 0.1146 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:14:02.163664+00:00", "notes": null, "run_key": "Qwen3-8B-BF16-sglang-xpu-tp1-baseline", "run_id": "Qwen3-8B-BF16-sglang-xpu-tp1-baseline" }, { "id": "Qwen3-8B-AWQ-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-8B-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "8B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 5.69, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 61.71, "aggregateDecodeTokS": 619.51, "perRequestDecodeTokS": 61.71, "ttftMs": 166.68, "acceptanceRate": null, "energyJoules": 522.538, "avgPowerWatts": 129.16, "joulesPerToken": 4.0823, "vramUsedGiB": 30.87, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 61.71, "perRequestTokS": 61.71, "ttftMs": 166.68, "energyJoules": 522.538, "avgPowerWatts": 129.16, "joulesPerToken": 4.0823 }, { "concurrency": 2, "aggregateThroughputTokS": 122.94, "perRequestTokS": 61.74, "ttftMs": 232.52, "energyJoules": 497.768, "avgPowerWatts": 143.23, "joulesPerToken": 1.9444 }, { "concurrency": 4, "aggregateThroughputTokS": 216.27, "perRequestTokS": 54.39, "ttftMs": 454.75, "energyJoules": 605.577, "avgPowerWatts": 147.13, "joulesPerToken": 1.1828 }, { "concurrency": 8, "aggregateThroughputTokS": 347.82, "perRequestTokS": 43.79, "ttftMs": 586.13, "energyJoules": 760.521, "avgPowerWatts": 165.23, "joulesPerToken": 0.7427 }, { "concurrency": 16, "aggregateThroughputTokS": 503.33, "perRequestTokS": 32.08, "ttftMs": 1096.2, "energyJoules": 1093.939, "avgPowerWatts": 175.2, "joulesPerToken": 0.5341 }, { "concurrency": 32, "aggregateThroughputTokS": 587.06, "perRequestTokS": 20.97, "ttftMs": 1924.57, "energyJoules": 1802.773, "avgPowerWatts": 197.26, "joulesPerToken": 0.4401 }, { "concurrency": 64, "aggregateThroughputTokS": 619.51, "perRequestTokS": 11.59, "ttftMs": 3165.65, "energyJoules": 2990.972, "avgPowerWatts": 195.81, "joulesPerToken": 0.3651 }, { "concurrency": 128, "aggregateThroughputTokS": 611.88, "perRequestTokS": 5.7, "ttftMs": 5175.3, "energyJoules": 5556.234, "avgPowerWatts": 186.18, "joulesPerToken": 0.3391 }, { "concurrency": 256, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 3972.36, "avgPowerWatts": 157.34, "joulesPerToken": 0.1212 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:14:57.811802+00:00", "notes": null, "run_key": "Qwen3-8B-AWQ-sglang-xpu-tp1-baseline", "run_id": "Qwen3-8B-AWQ-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-14B-Instruct-AWQ-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-14B-Instruct-AWQ", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "AWQ", "quantBits": 4, "checkpointDiskGiB": 9.31, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 40.7, "aggregateDecodeTokS": 450.83, "perRequestDecodeTokS": 40.7, "ttftMs": 203.15, "acceptanceRate": null, "energyJoules": 841.543, "avgPowerWatts": 133.56, "joulesPerToken": 6.5746, "vramUsedGiB": 30.92, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 40.7, "perRequestTokS": 40.7, "ttftMs": 203.15, "energyJoules": 841.543, "avgPowerWatts": 133.56, "joulesPerToken": 6.5746 }, { "concurrency": 2, "aggregateThroughputTokS": 81.76, "perRequestTokS": 41.04, "ttftMs": 338.51, "energyJoules": 772.925, "avgPowerWatts": 166.08, "joulesPerToken": 3.0192 }, { "concurrency": 4, "aggregateThroughputTokS": 144.96, "perRequestTokS": 36.96, "ttftMs": 638.8, "energyJoules": 901.221, "avgPowerWatts": 172.43, "joulesPerToken": 1.7602 }, { "concurrency": 8, "aggregateThroughputTokS": 230.25, "perRequestTokS": 31.54, "ttftMs": 904.83, "energyJoules": 1282.198, "avgPowerWatts": 137.93, "joulesPerToken": 1.2521 }, { "concurrency": 16, "aggregateThroughputTokS": 356.13, "perRequestTokS": 22.86, "ttftMs": 1743.83, "energyJoules": 1671.307, "avgPowerWatts": 193.09, "joulesPerToken": 0.8161 }, { "concurrency": 32, "aggregateThroughputTokS": 432.81, "perRequestTokS": 15.85, "ttftMs": 3180.87, "energyJoules": 2741.628, "avgPowerWatts": 202.49, "joulesPerToken": 0.6693 }, { "concurrency": 64, "aggregateThroughputTokS": 450.83, "perRequestTokS": 9.03, "ttftMs": 5190.35, "energyJoules": 4322.1, "avgPowerWatts": 204.59, "joulesPerToken": 0.5276 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 3031.511, "avgPowerWatts": 164.06, "joulesPerToken": 0.185 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:17:26.856446+00:00", "notes": null, "run_key": "Qwen2.5-14B-Instruct-AWQ-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-14B-Instruct-AWQ-sglang-xpu-tp1-baseline" }, { "id": "phi-4-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "phi-4", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "unknown", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.66, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 18.31, "aggregateDecodeTokS": 252.57, "perRequestDecodeTokS": 18.31, "ttftMs": 227.05, "acceptanceRate": null, "energyJoules": 1649.04, "avgPowerWatts": 151.6, "joulesPerToken": 12.8831, "vramUsedGiB": 30.39, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 18.31, "perRequestTokS": 18.31, "ttftMs": 227.05, "energyJoules": 1649.04, "avgPowerWatts": 151.6, "joulesPerToken": 12.8831 }, { "concurrency": 2, "aggregateThroughputTokS": 35.95, "perRequestTokS": 17.97, "ttftMs": 371.33, "energyJoules": 1570.609, "avgPowerWatts": 178.2, "joulesPerToken": 6.1352 }, { "concurrency": 4, "aggregateThroughputTokS": 67.99, "perRequestTokS": 17.17, "ttftMs": 418.59, "energyJoules": 1693.517, "avgPowerWatts": 180.24, "joulesPerToken": 3.3077 }, { "concurrency": 8, "aggregateThroughputTokS": 123.92, "perRequestTokS": 15.76, "ttftMs": 731.11, "energyJoules": 1919.191, "avgPowerWatts": 189.48, "joulesPerToken": 1.8742 }, { "concurrency": 16, "aggregateThroughputTokS": 218.2, "perRequestTokS": 13.7, "ttftMs": 1929.57, "energyJoules": 2418.31, "avgPowerWatts": 194.67, "joulesPerToken": 1.1808 }, { "concurrency": 32, "aggregateThroughputTokS": 199.84, "perRequestTokS": 11.27, "ttftMs": 3054.81, "energyJoules": 4788.954, "avgPowerWatts": 204.42, "joulesPerToken": 1.1692 }, { "concurrency": 64, "aggregateThroughputTokS": 229.16, "perRequestTokS": 11.06, "ttftMs": 11052.37, "energyJoules": 8146.45, "avgPowerWatts": 214.44, "joulesPerToken": 0.9944 }, { "concurrency": 128, "aggregateThroughputTokS": 249.55, "perRequestTokS": 10.81, "ttftMs": 26579.17, "energyJoules": 14823.068, "avgPowerWatts": 219.81, "joulesPerToken": 0.9047 }, { "concurrency": 256, "aggregateThroughputTokS": 243.89, "perRequestTokS": 10.5, "ttftMs": 57500.05, "energyJoules": 29667.65, "avgPowerWatts": 221.26, "joulesPerToken": 0.9054 }, { "concurrency": 512, "aggregateThroughputTokS": 249.96, "perRequestTokS": 10.41, "ttftMs": 120216.39, "energyJoules": 58096.605, "avgPowerWatts": 222.95, "joulesPerToken": 0.8865 }, { "concurrency": 1024, "aggregateThroughputTokS": 252.57, "perRequestTokS": 10.38, "ttftMs": 245118.14, "energyJoules": 114111.492, "avgPowerWatts": 223.5, "joulesPerToken": 0.8706 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:18:04.386478+00:00", "notes": null, "run_key": "phi-4-sglang-xpu-tp1-baseline", "run_id": "phi-4-sglang-xpu-tp1-baseline" }, { "id": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "DeepSeek-R1-Distill-Qwen-14B", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 36.69, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 17.34, "aggregateDecodeTokS": 224.03, "perRequestDecodeTokS": 17.34, "ttftMs": 296.28, "acceptanceRate": null, "energyJoules": 1667.593, "avgPowerWatts": 161.51, "joulesPerToken": 13.0281, "vramUsedGiB": 30.41, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 17.34, "perRequestTokS": 17.34, "ttftMs": 296.28, "energyJoules": 1667.593, "avgPowerWatts": 161.51, "joulesPerToken": 13.0281 }, { "concurrency": 2, "aggregateThroughputTokS": 34.49, "perRequestTokS": 17.31, "ttftMs": 383.6, "energyJoules": 1628.778, "avgPowerWatts": 177.66, "joulesPerToken": 6.3624 }, { "concurrency": 4, "aggregateThroughputTokS": 65.41, "perRequestTokS": 16.45, "ttftMs": 617.17, "energyJoules": 1759.458, "avgPowerWatts": 184.93, "joulesPerToken": 3.4364 }, { "concurrency": 8, "aggregateThroughputTokS": 120.4, "perRequestTokS": 15.15, "ttftMs": 991.11, "energyJoules": 2036.001, "avgPowerWatts": 189.61, "joulesPerToken": 1.9883 }, { "concurrency": 16, "aggregateThroughputTokS": 208.98, "perRequestTokS": 13.16, "ttftMs": 1669.06, "energyJoules": 2525.928, "avgPowerWatts": 201.67, "joulesPerToken": 1.2334 }, { "concurrency": 32, "aggregateThroughputTokS": 189.08, "perRequestTokS": 12.07, "ttftMs": 5162.79, "energyJoules": 5091.569, "avgPowerWatts": 209.57, "joulesPerToken": 1.2431 }, { "concurrency": 64, "aggregateThroughputTokS": 219.69, "perRequestTokS": 11.62, "ttftMs": 13909.37, "energyJoules": 8637.761, "avgPowerWatts": 217.2, "joulesPerToken": 1.0544 }, { "concurrency": 128, "aggregateThroughputTokS": 215.4, "perRequestTokS": 11.51, "ttftMs": 31651.94, "energyJoules": 17305.134, "avgPowerWatts": 220.87, "joulesPerToken": 1.0562 }, { "concurrency": 256, "aggregateThroughputTokS": 224.03, "perRequestTokS": 11.3, "ttftMs": 67641.81, "energyJoules": 33157.901, "avgPowerWatts": 224.08, "joulesPerToken": 1.0119 }, { "concurrency": 512, "aggregateThroughputTokS": 222.76, "perRequestTokS": 11.28, "ttftMs": 139814.48, "energyJoules": 66377.938, "avgPowerWatts": 225.14, "joulesPerToken": 1.0128 }, { "concurrency": 1024, "aggregateThroughputTokS": 221.48, "perRequestTokS": 11.32, "ttftMs": 281766.24, "energyJoules": 131994.666, "avgPowerWatts": 225.43, "joulesPerToken": 1.007 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:19:34.824992+00:00", "notes": null, "run_key": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp1-baseline", "run_id": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp1-baseline" }, { "id": "Qwen2.5-14B-Instruct-sglang-xpu-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-14B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.57, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 17.23, "aggregateDecodeTokS": 236.66, "perRequestDecodeTokS": 17.23, "ttftMs": 301.1, "acceptanceRate": null, "energyJoules": 1645.913, "avgPowerWatts": 161.48, "joulesPerToken": 12.8587, "vramUsedGiB": 30.49, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 17.23, "perRequestTokS": 17.23, "ttftMs": 301.1, "energyJoules": 1645.913, "avgPowerWatts": 161.48, "joulesPerToken": 12.8587 }, { "concurrency": 2, "aggregateThroughputTokS": 24.51, "perRequestTokS": 17.6, "ttftMs": 384.54, "energyJoules": 1435.633, "avgPowerWatts": 176.41, "joulesPerToken": 5.6079 }, { "concurrency": 4, "aggregateThroughputTokS": 65.72, "perRequestTokS": 16.53, "ttftMs": 649.91, "energyJoules": 1723.009, "avgPowerWatts": 181.57, "joulesPerToken": 3.3653 }, { "concurrency": 8, "aggregateThroughputTokS": 115.5, "perRequestTokS": 15.24, "ttftMs": 944.46, "energyJoules": 1966.617, "avgPowerWatts": 189.53, "joulesPerToken": 1.9205 }, { "concurrency": 16, "aggregateThroughputTokS": 195.91, "perRequestTokS": 13.43, "ttftMs": 1622.89, "energyJoules": 2411.033, "avgPowerWatts": 197.2, "joulesPerToken": 1.1773 }, { "concurrency": 32, "aggregateThroughputTokS": 198.16, "perRequestTokS": 10.95, "ttftMs": 3658.39, "energyJoules": 4652.556, "avgPowerWatts": 205.46, "joulesPerToken": 1.1359 }, { "concurrency": 64, "aggregateThroughputTokS": 221.12, "perRequestTokS": 10.27, "ttftMs": 11174.46, "energyJoules": 7980.65, "avgPowerWatts": 211.77, "joulesPerToken": 0.9742 }, { "concurrency": 128, "aggregateThroughputTokS": 229.56, "perRequestTokS": 10.15, "ttftMs": 27231.26, "energyJoules": 15259.357, "avgPowerWatts": 216.61, "joulesPerToken": 0.9314 }, { "concurrency": 256, "aggregateThroughputTokS": 236.09, "perRequestTokS": 8.99, "ttftMs": 59229.02, "energyJoules": 29517.566, "avgPowerWatts": 219, "joulesPerToken": 0.9008 }, { "concurrency": 512, "aggregateThroughputTokS": 236.66, "perRequestTokS": 8.52, "ttftMs": 123725.92, "energyJoules": 58179.734, "avgPowerWatts": 219.86, "joulesPerToken": 0.8878 }, { "concurrency": 1024, "aggregateThroughputTokS": 234.21, "perRequestTokS": 8.33, "ttftMs": 252440.13, "energyJoules": 116110.177, "avgPowerWatts": 219.86, "joulesPerToken": 0.8859 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:20:20.783200+00:00", "notes": null, "run_key": "Qwen2.5-14B-Instruct-sglang-xpu-tp1-baseline", "run_id": "Qwen2.5-14B-Instruct-sglang-xpu-tp1-baseline" }, { "id": "Qwen3-0.6B-GGUF-llama.cpp-sycl-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-0.6B-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.6B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 0.45, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 337.9, "aggregateDecodeTokS": 899.69, "perRequestDecodeTokS": 337.9, "ttftMs": 36.76, "acceptanceRate": null, "energyJoules": 240.427, "avgPowerWatts": 66.31, "joulesPerToken": 1.8783, "vramUsedGiB": 2.49, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 337.9, "perRequestTokS": 337.9, "ttftMs": 36.76, "energyJoules": 240.427, "avgPowerWatts": 66.31, "joulesPerToken": 1.8783 }, { "concurrency": 2, "aggregateThroughputTokS": 613.72, "perRequestTokS": 307.04, "ttftMs": 43.77, "energyJoules": 226.099, "avgPowerWatts": 70.46, "joulesPerToken": 0.8832 }, { "concurrency": 4, "aggregateThroughputTokS": 899.69, "perRequestTokS": 225.04, "ttftMs": 50.61, "energyJoules": 278.72, "avgPowerWatts": 77.21, "joulesPerToken": 0.5444 }, { "concurrency": 8, "aggregateThroughputTokS": 740.7, "perRequestTokS": 218.89, "ttftMs": 467.24, "energyJoules": 418.577, "avgPowerWatts": 93.15, "joulesPerToken": 0.4088 }, { "concurrency": 16, "aggregateThroughputTokS": 681.98, "perRequestTokS": 223.58, "ttftMs": 1252.54, "energyJoules": 670.288, "avgPowerWatts": 112.49, "joulesPerToken": 0.3273 }, { "concurrency": 32, "aggregateThroughputTokS": 644.85, "perRequestTokS": 208.31, "ttftMs": 2926.8, "energyJoules": 1210.028, "avgPowerWatts": 130.59, "joulesPerToken": 0.2954 }, { "concurrency": 64, "aggregateThroughputTokS": 628.99, "perRequestTokS": 205.27, "ttftMs": 6264.19, "energyJoules": 2302.582, "avgPowerWatts": 143.37, "joulesPerToken": 0.2811 }, { "concurrency": 128, "aggregateThroughputTokS": 610.71, "perRequestTokS": 200.21, "ttftMs": 13181.65, "energyJoules": 4518.966, "avgPowerWatts": 152.29, "joulesPerToken": 0.2758 }, { "concurrency": 256, "aggregateThroughputTokS": 595.9, "perRequestTokS": 183.16, "ttftMs": 27431.59, "energyJoules": 9095.546, "avgPowerWatts": 157.51, "joulesPerToken": 0.2776 }, { "concurrency": 512, "aggregateThroughputTokS": 590.03, "perRequestTokS": 178.85, "ttftMs": 55982.71, "energyJoules": 18302.85, "avgPowerWatts": 161.16, "joulesPerToken": 0.2793 }, { "concurrency": 1024, "aggregateThroughputTokS": 570.31, "perRequestTokS": 172.57, "ttftMs": 115761.22, "energyJoules": 36914.206, "avgPowerWatts": 160.83, "joulesPerToken": 0.2816 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:40:01.063576+00:00", "notes": null, "run_key": "Qwen3-0.6B-GGUF-llama.cpp-sycl-tp1-baseline", "run_id": "Qwen3-0.6B-GGUF-llama.cpp-sycl-tp1-baseline" }, { "id": "Qwen2.5-0.5B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-0.5B-Instruct-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.5B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 0.92, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 205.39, "aggregateDecodeTokS": 571.92, "perRequestDecodeTokS": 205.39, "ttftMs": 30.31, "acceptanceRate": null, "energyJoules": 240.38, "avgPowerWatts": 63.86, "joulesPerToken": 1.878, "vramUsedGiB": 0.95, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 205.39, "perRequestTokS": 205.39, "ttftMs": 30.31, "energyJoules": 240.38, "avgPowerWatts": 63.86, "joulesPerToken": 1.878 }, { "concurrency": 2, "aggregateThroughputTokS": 429.51, "perRequestTokS": 215.1, "ttftMs": 63.5, "energyJoules": 202.427, "avgPowerWatts": 59.47, "joulesPerToken": 0.7907 }, { "concurrency": 4, "aggregateThroughputTokS": 365.67, "perRequestTokS": 190.05, "ttftMs": 40.78, "energyJoules": 288.213, "avgPowerWatts": 77.78, "joulesPerToken": 0.5629 }, { "concurrency": 8, "aggregateThroughputTokS": 534.85, "perRequestTokS": 180.7, "ttftMs": 451.48, "energyJoules": 431.769, "avgPowerWatts": 92.99, "joulesPerToken": 0.4216 }, { "concurrency": 16, "aggregateThroughputTokS": 571.92, "perRequestTokS": 176.73, "ttftMs": 677.19, "energyJoules": 477.991, "avgPowerWatts": 99.28, "joulesPerToken": 0.2334 }, { "concurrency": 32, "aggregateThroughputTokS": 566.93, "perRequestTokS": 159.77, "ttftMs": 2490.61, "energyJoules": 1098.46, "avgPowerWatts": 125.49, "joulesPerToken": 0.2682 }, { "concurrency": 64, "aggregateThroughputTokS": 560.91, "perRequestTokS": 158.58, "ttftMs": 4695.07, "energyJoules": 1781.478, "avgPowerWatts": 136.2, "joulesPerToken": 0.2175 }, { "concurrency": 128, "aggregateThroughputTokS": 552.33, "perRequestTokS": 154.07, "ttftMs": 10086.89, "energyJoules": 3430.579, "avgPowerWatts": 144.8, "joulesPerToken": 0.2094 }, { "concurrency": 256, "aggregateThroughputTokS": 539.78, "perRequestTokS": 151.11, "ttftMs": 21803.75, "energyJoules": 7052.486, "avgPowerWatts": 150.14, "joulesPerToken": 0.2152 }, { "concurrency": 512, "aggregateThroughputTokS": 514.44, "perRequestTokS": 142.47, "ttftMs": 43860.78, "energyJoules": 13873.488, "avgPowerWatts": 151.27, "joulesPerToken": 0.2117 }, { "concurrency": 1024, "aggregateThroughputTokS": 492.18, "perRequestTokS": 138.37, "ttftMs": 88159.53, "energyJoules": 26868.678, "avgPowerWatts": 150.66, "joulesPerToken": 0.205 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:40:01.192275+00:00", "notes": null, "run_key": "Qwen2.5-0.5B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline", "run_id": "Qwen2.5-0.5B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline" }, { "id": "Qwen_Qwen3.5-0.8B-GGUF-llama.cpp-sycl-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen_Qwen3.5-0.8B-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "0.8B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 1.27, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 293.68, "aggregateDecodeTokS": 624.84, "perRequestDecodeTokS": 293.68, "ttftMs": 40.88, "acceptanceRate": null, "energyJoules": 255.774, "avgPowerWatts": 65.1, "joulesPerToken": 1.9982, "vramUsedGiB": 1.43, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 293.68, "perRequestTokS": 293.68, "ttftMs": 40.88, "energyJoules": 255.774, "avgPowerWatts": 65.1, "joulesPerToken": 1.9982 }, { "concurrency": 2, "aggregateThroughputTokS": 446.76, "perRequestTokS": 223.47, "ttftMs": 91.37, "energyJoules": 257.912, "avgPowerWatts": 74.83, "joulesPerToken": 1.0075 }, { "concurrency": 4, "aggregateThroughputTokS": 624.84, "perRequestTokS": 156.4, "ttftMs": 83.6, "energyJoules": 327.056, "avgPowerWatts": 85.17, "joulesPerToken": 0.6388 }, { "concurrency": 8, "aggregateThroughputTokS": 583.04, "perRequestTokS": 156.92, "ttftMs": 540.48, "energyJoules": 508.549, "avgPowerWatts": 106.03, "joulesPerToken": 0.4966 }, { "concurrency": 16, "aggregateThroughputTokS": 562.63, "perRequestTokS": 157.13, "ttftMs": 1475.89, "energyJoules": 865.239, "avgPowerWatts": 129.96, "joulesPerToken": 0.4225 }, { "concurrency": 32, "aggregateThroughputTokS": 554.96, "perRequestTokS": 156.59, "ttftMs": 3335.82, "energyJoules": 1592.657, "avgPowerWatts": 150.38, "joulesPerToken": 0.3888 }, { "concurrency": 64, "aggregateThroughputTokS": 548.67, "perRequestTokS": 155.63, "ttftMs": 7087.22, "energyJoules": 3024.576, "avgPowerWatts": 171.17, "joulesPerToken": 0.3692 }, { "concurrency": 128, "aggregateThroughputTokS": 528.72, "perRequestTokS": 145.66, "ttftMs": 15102.26, "energyJoules": 6089.922, "avgPowerWatts": 181.23, "joulesPerToken": 0.3717 }, { "concurrency": 256, "aggregateThroughputTokS": 519.58, "perRequestTokS": 142.75, "ttftMs": 31266.58, "energyJoules": 12212.387, "avgPowerWatts": 186.56, "joulesPerToken": 0.3727 }, { "concurrency": 512, "aggregateThroughputTokS": 505.59, "perRequestTokS": 138.69, "ttftMs": 65134.22, "energyJoules": 24698.306, "avgPowerWatts": 187.77, "joulesPerToken": 0.3769 }, { "concurrency": 1024, "aggregateThroughputTokS": 488.79, "perRequestTokS": 133.8, "ttftMs": 134680.61, "energyJoules": 49750.145, "avgPowerWatts": 186.27, "joulesPerToken": 0.3796 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:40:01.356751+00:00", "notes": null, "run_key": "Qwen_Qwen3.5-0.8B-GGUF-llama.cpp-sycl-tp1-baseline", "run_id": "Qwen_Qwen3.5-0.8B-GGUF-llama.cpp-sycl-tp1-baseline" }, { "id": "Qwen2.5-3B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-3B-Instruct-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "3B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 1.96, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 153.12, "aggregateDecodeTokS": 330.6, "perRequestDecodeTokS": 153.12, "ttftMs": 62.5, "acceptanceRate": null, "energyJoules": 391.954, "avgPowerWatts": 97.92, "joulesPerToken": 3.0621, "vramUsedGiB": 2.93, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 153.12, "perRequestTokS": 153.12, "ttftMs": 62.5, "energyJoules": 391.954, "avgPowerWatts": 97.92, "joulesPerToken": 3.0621 }, { "concurrency": 2, "aggregateThroughputTokS": 253.64, "perRequestTokS": 126.86, "ttftMs": 117.15, "energyJoules": 450.169, "avgPowerWatts": 108.69, "joulesPerToken": 1.7585 }, { "concurrency": 4, "aggregateThroughputTokS": 330.6, "perRequestTokS": 91.4, "ttftMs": 91.35, "energyJoules": 562.338, "avgPowerWatts": 123.9, "joulesPerToken": 1.0983 }, { "concurrency": 8, "aggregateThroughputTokS": 305.56, "perRequestTokS": 84.23, "ttftMs": 919.25, "energyJoules": 1023.128, "avgPowerWatts": 156.91, "joulesPerToken": 0.9991 }, { "concurrency": 16, "aggregateThroughputTokS": 286.18, "perRequestTokS": 84.01, "ttftMs": 2639.5, "energyJoules": 1823.697, "avgPowerWatts": 186.03, "joulesPerToken": 0.8905 }, { "concurrency": 32, "aggregateThroughputTokS": 277.9, "perRequestTokS": 75.09, "ttftMs": 6455.08, "energyJoules": 3728.822, "avgPowerWatts": 210.95, "joulesPerToken": 0.9104 }, { "concurrency": 64, "aggregateThroughputTokS": 277.31, "perRequestTokS": 76.41, "ttftMs": 13157.5, "energyJoules": 7122.389, "avgPowerWatts": 227.49, "joulesPerToken": 0.8694 }, { "concurrency": 128, "aggregateThroughputTokS": 270.52, "perRequestTokS": 73.08, "ttftMs": 26971.67, "energyJoules": 14183.262, "avgPowerWatts": 238.15, "joulesPerToken": 0.8657 }, { "concurrency": 256, "aggregateThroughputTokS": 267.11, "perRequestTokS": 71.57, "ttftMs": 57349.48, "energyJoules": 28892.276, "avgPowerWatts": 242.1, "joulesPerToken": 0.8817 }, { "concurrency": 512, "aggregateThroughputTokS": 264.28, "perRequestTokS": 70.46, "ttftMs": 119218.69, "energyJoules": 58554.67, "avgPowerWatts": 242.45, "joulesPerToken": 0.8935 }, { "concurrency": 1024, "aggregateThroughputTokS": 258.35, "perRequestTokS": 68.89, "ttftMs": 242514.44, "energyJoules": 116697.108, "avgPowerWatts": 240.82, "joulesPerToken": 0.8903 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:46:29.139387+00:00", "notes": null, "run_key": "Qwen2.5-3B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline", "run_id": "Qwen2.5-3B-Instruct-GGUF-llama.cpp-sycl-tp1-baseline" }, { "id": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp1-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-30B-A3B-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "30B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 17.28, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 1, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-1", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 116.33, "aggregateDecodeTokS": 126.49, "perRequestDecodeTokS": 116.33, "ttftMs": 685.51, "acceptanceRate": null, "energyJoules": 544.094, "avgPowerWatts": 114.1, "joulesPerToken": 4.2507, "vramUsedGiB": 19.43, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 116.33, "perRequestTokS": 116.33, "ttftMs": 685.51, "energyJoules": 544.094, "avgPowerWatts": 114.1, "joulesPerToken": 4.2507 }, { "concurrency": 2, "aggregateThroughputTokS": 98.5, "perRequestTokS": 49.25, "ttftMs": 295.34, "energyJoules": 682.454, "avgPowerWatts": 118.68, "joulesPerToken": 2.6658 }, { "concurrency": 4, "aggregateThroughputTokS": 126.49, "perRequestTokS": 31.63, "ttftMs": 413.46, "energyJoules": 997.499, "avgPowerWatts": 133.44, "joulesPerToken": 1.9482 }, { "concurrency": 8, "aggregateThroughputTokS": 116.44, "perRequestTokS": 31.58, "ttftMs": 2712.95, "energyJoules": 1886.589, "avgPowerWatts": 152.32, "joulesPerToken": 1.8424 }, { "concurrency": 16, "aggregateThroughputTokS": 110.37, "perRequestTokS": 31.31, "ttftMs": 7520.33, "energyJoules": 3685.596, "avgPowerWatts": 167.66, "joulesPerToken": 1.7996 }, { "concurrency": 32, "aggregateThroughputTokS": 106.53, "perRequestTokS": 30.29, "ttftMs": 17626.27, "energyJoules": 7478.519, "avgPowerWatts": 176.74, "joulesPerToken": 1.8258 }, { "concurrency": 64, "aggregateThroughputTokS": 104.94, "perRequestTokS": 30.08, "ttftMs": 37185.62, "energyJoules": 14882.065, "avgPowerWatts": 183.13, "joulesPerToken": 1.8167 }, { "concurrency": 128, "aggregateThroughputTokS": 104.39, "perRequestTokS": 30.06, "ttftMs": 76384.82, "energyJoules": 29663.436, "avgPowerWatts": 185.84, "joulesPerToken": 1.8105 }, { "concurrency": 256, "aggregateThroughputTokS": 103.35, "perRequestTokS": 29.77, "ttftMs": 156258.85, "energyJoules": 59729.126, "avgPowerWatts": 187.57, "joulesPerToken": 1.8228 }, { "concurrency": 512, "aggregateThroughputTokS": 120.78, "perRequestTokS": 33.66, "ttftMs": 58239.11, "energyJoules": 212926.863, "avgPowerWatts": 188.44, "joulesPerToken": 3.327 }, { "concurrency": 1024, "aggregateThroughputTokS": 57.93, "perRequestTokS": 18.46, "ttftMs": 32324.91, "energyJoules": 79049.785, "avgPowerWatts": 43.27, "joulesPerToken": 1.1188 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-08T15:49:22.649801+00:00", "notes": null, "run_key": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp1-baseline", "run_id": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp1-baseline" }, { "id": "Qwen3.6-35B-A3B-FP8-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.6-35B-A3B-FP8", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "35B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 34.92, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 17.58, "aggregateDecodeTokS": 1291.78, "perRequestDecodeTokS": 17.58, "ttftMs": 92.91, "acceptanceRate": null, "energyJoules": 1837.707, "avgPowerWatts": 171.37, "joulesPerToken": 14.3571, "vramUsedGiB": 58.88, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 17.58, "perRequestTokS": 17.58, "ttftMs": 92.91, "energyJoules": 1837.707, "avgPowerWatts": 171.37, "joulesPerToken": 14.3571 }, { "concurrency": 2, "aggregateThroughputTokS": 39.86, "perRequestTokS": 20.12, "ttftMs": 108.02, "energyJoules": 1564.762, "avgPowerWatts": 209.11, "joulesPerToken": 6.1124 }, { "concurrency": 4, "aggregateThroughputTokS": 74.81, "perRequestTokS": 19.66, "ttftMs": 371.8, "energyJoules": 1716.663, "avgPowerWatts": 218.1, "joulesPerToken": 3.3529 }, { "concurrency": 8, "aggregateThroughputTokS": 151.75, "perRequestTokS": 19.55, "ttftMs": 259.02, "energyJoules": 1832.548, "avgPowerWatts": 236.26, "joulesPerToken": 1.7896 }, { "concurrency": 16, "aggregateThroughputTokS": 287.32, "perRequestTokS": 18.62, "ttftMs": 318.4, "energyJoules": 2110.224, "avgPowerWatts": 259.08, "joulesPerToken": 1.0304 }, { "concurrency": 32, "aggregateThroughputTokS": 520.11, "perRequestTokS": 17.23, "ttftMs": 439.27, "energyJoules": 2619.737, "avgPowerWatts": 295.55, "joulesPerToken": 0.6396 }, { "concurrency": 64, "aggregateThroughputTokS": 869.91, "perRequestTokS": 14.76, "ttftMs": 677.75, "energyJoules": 3531.223, "avgPowerWatts": 338.55, "joulesPerToken": 0.4311 }, { "concurrency": 128, "aggregateThroughputTokS": 1291.78, "perRequestTokS": 11.35, "ttftMs": 1194.15, "energyJoules": 5233.587, "avgPowerWatts": 381.51, "joulesPerToken": 0.3194 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:20:52.784063+00:00", "notes": null, "run_key": "Qwen3.6-35B-A3B-FP8-vllm-xpu-tp2-baseline", "run_id": "Qwen3.6-35B-A3B-FP8-vllm-xpu-tp2-baseline" }, { "id": "Qwen3.8-27B-FP8-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.8-27B-FP8", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "27B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 29.4, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 13.84, "aggregateDecodeTokS": 493.96, "perRequestDecodeTokS": 13.84, "ttftMs": 278.59, "acceptanceRate": null, "energyJoules": 4032.959, "avgPowerWatts": 350.05, "joulesPerToken": 31.5075, "vramUsedGiB": 58.96, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 13.84, "perRequestTokS": 13.84, "ttftMs": 278.59, "energyJoules": 4032.959, "avgPowerWatts": 350.05, "joulesPerToken": 31.5075 }, { "concurrency": 2, "aggregateThroughputTokS": 28.08, "perRequestTokS": 14.2, "ttftMs": 206.31, "energyJoules": 4023.939, "avgPowerWatts": 390.94, "joulesPerToken": 15.7185 }, { "concurrency": 4, "aggregateThroughputTokS": 54.19, "perRequestTokS": 14.07, "ttftMs": 447.32, "energyJoules": 4225.272, "avgPowerWatts": 398.5, "joulesPerToken": 8.2525 }, { "concurrency": 8, "aggregateThroughputTokS": 102.81, "perRequestTokS": 13.81, "ttftMs": 777.79, "energyJoules": 4541.779, "avgPowerWatts": 408.16, "joulesPerToken": 4.4353 }, { "concurrency": 16, "aggregateThroughputTokS": 194.23, "perRequestTokS": 13.5, "ttftMs": 1176.54, "energyJoules": 5185.173, "avgPowerWatts": 442.93, "joulesPerToken": 2.5318 }, { "concurrency": 32, "aggregateThroughputTokS": 327.98, "perRequestTokS": 11.86, "ttftMs": 1688.08, "energyJoules": 6510.534, "avgPowerWatts": 477.45, "joulesPerToken": 1.5895 }, { "concurrency": 64, "aggregateThroughputTokS": 493.96, "perRequestTokS": 9.39, "ttftMs": 2797.56, "energyJoules": 9221.354, "avgPowerWatts": 519.04, "joulesPerToken": 1.1257 }, { "concurrency": 128, "aggregateThroughputTokS": 452.31, "perRequestTokS": 7.02, "ttftMs": 7908.58, "energyJoules": 19926.235, "avgPowerWatts": 535.78, "joulesPerToken": 1.2162 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:25:06.323032+00:00", "notes": null, "run_key": "Qwen3.8-27B-FP8-vllm-xpu-tp2-baseline", "run_id": "Qwen3.8-27B-FP8-vllm-xpu-tp2-baseline" }, { "id": "gemma-4-26B-A4B-it-FP8-Dynamic-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "gemma-4-26B-A4B-it-FP8-Dynamic", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "26B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 26.7, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 8.7, "aggregateDecodeTokS": 314.45, "perRequestDecodeTokS": 8.7, "ttftMs": 1400.93, "acceptanceRate": null, "energyJoules": 3983.094, "avgPowerWatts": 124.94, "joulesPerToken": 31.1179, "vramUsedGiB": 60.24, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 8.7, "perRequestTokS": 8.7, "ttftMs": 1400.93, "energyJoules": 3983.094, "avgPowerWatts": 124.94, "joulesPerToken": 31.1179 }, { "concurrency": 2, "aggregateThroughputTokS": 28.25, "perRequestTokS": 14.97, "ttftMs": 648.91, "energyJoules": 2564.676, "avgPowerWatts": 164.07, "joulesPerToken": 10.0183 }, { "concurrency": 4, "aggregateThroughputTokS": 57.24, "perRequestTokS": 16.19, "ttftMs": 619.53, "energyJoules": 2486.93, "avgPowerWatts": 240.92, "joulesPerToken": 4.8573 }, { "concurrency": 8, "aggregateThroughputTokS": 79.75, "perRequestTokS": 12.86, "ttftMs": 2019.24, "energyJoules": 3195.813, "avgPowerWatts": 238.33, "joulesPerToken": 3.1209 }, { "concurrency": 16, "aggregateThroughputTokS": 143.33, "perRequestTokS": 10.46, "ttftMs": 1857.84, "energyJoules": 4146.034, "avgPowerWatts": 270.27, "joulesPerToken": 2.0244 }, { "concurrency": 32, "aggregateThroughputTokS": 244.56, "perRequestTokS": 8.47, "ttftMs": 1746.79, "energyJoules": 5582.024, "avgPowerWatts": 306.76, "joulesPerToken": 1.3628 }, { "concurrency": 64, "aggregateThroughputTokS": 314.45, "perRequestTokS": 5.62, "ttftMs": 2762.12, "energyJoules": 9259.016, "avgPowerWatts": 342.38, "joulesPerToken": 1.1303 }, { "concurrency": 128, "aggregateThroughputTokS": 147.53, "perRequestTokS": 1.29, "ttftMs": 4859.36, "energyJoules": 28159.107, "avgPowerWatts": 260.74, "joulesPerToken": 1.7187 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:30:35.445121+00:00", "notes": null, "run_key": "gemma-4-26B-A4B-it-FP8-Dynamic-vllm-xpu-tp2-baseline", "run_id": "gemma-4-26B-A4B-it-FP8-Dynamic-vllm-xpu-tp2-baseline" }, { "id": "gemma-4-31B-it-FP8-Dynamic-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "gemma-4-31B-it-FP8-Dynamic", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "31B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 31.01, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 16.01, "aggregateDecodeTokS": 376.52, "perRequestDecodeTokS": 16.01, "ttftMs": 433.37, "acceptanceRate": null, "energyJoules": 584.324, "avgPowerWatts": 169.38, "joulesPerToken": 4.565, "vramUsedGiB": 62.12, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 16.01, "perRequestTokS": 16.01, "ttftMs": 433.37, "energyJoules": 584.324, "avgPowerWatts": 169.38, "joulesPerToken": 4.565 }, { "concurrency": 2, "aggregateThroughputTokS": 30.95, "perRequestTokS": 16.3, "ttftMs": 469.98, "energyJoules": 2792.834, "avgPowerWatts": 289.26, "joulesPerToken": 10.9095 }, { "concurrency": 4, "aggregateThroughputTokS": 56.82, "perRequestTokS": 16.3, "ttftMs": 271.2, "energyJoules": 3002.547, "avgPowerWatts": 318.82, "joulesPerToken": 5.8643 }, { "concurrency": 8, "aggregateThroughputTokS": 87.32, "perRequestTokS": 14.13, "ttftMs": 1188.92, "energyJoules": 3567.613, "avgPowerWatts": 307.51, "joulesPerToken": 3.484 }, { "concurrency": 16, "aggregateThroughputTokS": 141.04, "perRequestTokS": 10.39, "ttftMs": 844.13, "energyJoules": 4671.957, "avgPowerWatts": 329.42, "joulesPerToken": 2.2812 }, { "concurrency": 32, "aggregateThroughputTokS": 279.22, "perRequestTokS": 11.29, "ttftMs": 1217.39, "energyJoules": 5443.701, "avgPowerWatts": 408.08, "joulesPerToken": 1.329 }, { "concurrency": 64, "aggregateThroughputTokS": 355.14, "perRequestTokS": 7.62, "ttftMs": 2031.77, "energyJoules": 8772.072, "avgPowerWatts": 457.1, "joulesPerToken": 1.0708 }, { "concurrency": 128, "aggregateThroughputTokS": 376.52, "perRequestTokS": 4.17, "ttftMs": 3633.58, "energyJoules": 17183.005, "avgPowerWatts": 452.38, "joulesPerToken": 1.0488 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:37:38.405993+00:00", "notes": null, "run_key": "gemma-4-31B-it-FP8-Dynamic-vllm-xpu-tp2-baseline", "run_id": "gemma-4-31B-it-FP8-Dynamic-vllm-xpu-tp2-baseline" }, { "id": "phi-4-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "phi-4", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.66, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 24.35, "aggregateDecodeTokS": 1704.65, "perRequestDecodeTokS": 24.35, "ttftMs": 248.22, "acceptanceRate": null, "energyJoules": 1872.507, "avgPowerWatts": 280.62, "joulesPerToken": 14.629, "vramUsedGiB": 59.46, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 24.35, "perRequestTokS": 24.35, "ttftMs": 248.22, "energyJoules": 1872.507, "avgPowerWatts": 280.62, "joulesPerToken": 14.629 }, { "concurrency": 2, "aggregateThroughputTokS": 48.45, "perRequestTokS": 24.64, "ttftMs": 152.04, "energyJoules": 1843.56, "avgPowerWatts": 299.55, "joulesPerToken": 7.2014 }, { "concurrency": 4, "aggregateThroughputTokS": 96.2, "perRequestTokS": 24.56, "ttftMs": 147.97, "energyJoules": 1886.105, "avgPowerWatts": 307.11, "joulesPerToken": 3.6838 }, { "concurrency": 8, "aggregateThroughputTokS": 187.18, "perRequestTokS": 24.34, "ttftMs": 261.15, "energyJoules": 1998.786, "avgPowerWatts": 316.92, "joulesPerToken": 1.9519 }, { "concurrency": 16, "aggregateThroughputTokS": 360.01, "perRequestTokS": 24.06, "ttftMs": 377.05, "energyJoules": 2188.127, "avgPowerWatts": 338.37, "joulesPerToken": 1.0684 }, { "concurrency": 32, "aggregateThroughputTokS": 669.54, "perRequestTokS": 22.82, "ttftMs": 541.16, "energyJoules": 2636.565, "avgPowerWatts": 379.1, "joulesPerToken": 0.6437 }, { "concurrency": 64, "aggregateThroughputTokS": 1092.15, "perRequestTokS": 19.45, "ttftMs": 854.73, "energyJoules": 3414.75, "avgPowerWatts": 411.34, "joulesPerToken": 0.4168 }, { "concurrency": 128, "aggregateThroughputTokS": 1704.65, "perRequestTokS": 16.31, "ttftMs": 1472.26, "energyJoules": 4812.936, "avgPowerWatts": 467.34, "joulesPerToken": 0.2938 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:41:46.334065+00:00", "notes": null, "run_key": "phi-4-vllm-xpu-tp2-baseline", "run_id": "phi-4-vllm-xpu-tp2-baseline" }, { "id": "Qwen2.5-14B-Instruct-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-14B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.57, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 20.98, "aggregateDecodeTokS": 1522.92, "perRequestDecodeTokS": 20.98, "ttftMs": 253.43, "acceptanceRate": null, "energyJoules": 2003.134, "avgPowerWatts": 270.38, "joulesPerToken": 15.6495, "vramUsedGiB": 59.34, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 20.98, "perRequestTokS": 20.98, "ttftMs": 253.43, "energyJoules": 2003.134, "avgPowerWatts": 270.38, "joulesPerToken": 15.6495 }, { "concurrency": 2, "aggregateThroughputTokS": 41.05, "perRequestTokS": 20.99, "ttftMs": 153.7, "energyJoules": 2008.255, "avgPowerWatts": 280.47, "joulesPerToken": 7.8447 }, { "concurrency": 4, "aggregateThroughputTokS": 78.72, "perRequestTokS": 20.99, "ttftMs": 204.25, "energyJoules": 2060.51, "avgPowerWatts": 288.82, "joulesPerToken": 4.0244 }, { "concurrency": 8, "aggregateThroughputTokS": 133.07, "perRequestTokS": 20.25, "ttftMs": 1055.77, "energyJoules": 2251.369, "avgPowerWatts": 275.55, "joulesPerToken": 2.1986 }, { "concurrency": 16, "aggregateThroughputTokS": 290.88, "perRequestTokS": 20.61, "ttftMs": 386.23, "energyJoules": 2347.214, "avgPowerWatts": 314.86, "joulesPerToken": 1.1461 }, { "concurrency": 32, "aggregateThroughputTokS": 552.91, "perRequestTokS": 19.69, "ttftMs": 551.74, "energyJoules": 2767.847, "avgPowerWatts": 349.51, "joulesPerToken": 0.6757 }, { "concurrency": 64, "aggregateThroughputTokS": 935.08, "perRequestTokS": 17.01, "ttftMs": 864.55, "energyJoules": 3561.321, "avgPowerWatts": 382.73, "joulesPerToken": 0.4347 }, { "concurrency": 128, "aggregateThroughputTokS": 1522.92, "perRequestTokS": 14.75, "ttftMs": 1499.44, "energyJoules": 4972.615, "avgPowerWatts": 442.22, "joulesPerToken": 0.3035 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:44:50.058348+00:00", "notes": null, "run_key": "Qwen2.5-14B-Instruct-vllm-xpu-tp2-baseline", "run_id": "Qwen2.5-14B-Instruct-vllm-xpu-tp2-baseline" }, { "id": "DeepSeek-R1-Distill-Qwen-14B-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "DeepSeek-R1-Distill-Qwen-14B", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 36.69, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 21.54, "aggregateDecodeTokS": 1609.25, "perRequestDecodeTokS": 21.54, "ttftMs": 253.43, "acceptanceRate": null, "energyJoules": 1949.749, "avgPowerWatts": 273.21, "joulesPerToken": 15.2324, "vramUsedGiB": 59.39, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 21.54, "perRequestTokS": 21.54, "ttftMs": 253.43, "energyJoules": 1949.749, "avgPowerWatts": 273.21, "joulesPerToken": 15.2324 }, { "concurrency": 2, "aggregateThroughputTokS": 40.74, "perRequestTokS": 21.11, "ttftMs": 233.2, "energyJoules": 2011.682, "avgPowerWatts": 279.53, "joulesPerToken": 7.8581 }, { "concurrency": 4, "aggregateThroughputTokS": 83.76, "perRequestTokS": 21.47, "ttftMs": 192.69, "energyJoules": 2026.755, "avgPowerWatts": 290.18, "joulesPerToken": 3.9585 }, { "concurrency": 8, "aggregateThroughputTokS": 121.85, "perRequestTokS": 20.79, "ttftMs": 2206.53, "energyJoules": 2333.89, "avgPowerWatts": 251.24, "joulesPerToken": 2.2792 }, { "concurrency": 16, "aggregateThroughputTokS": 305.99, "perRequestTokS": 21.15, "ttftMs": 691.33, "energyJoules": 2337.793, "avgPowerWatts": 307.93, "joulesPerToken": 1.1415 }, { "concurrency": 32, "aggregateThroughputTokS": 597.97, "perRequestTokS": 20.21, "ttftMs": 549.31, "energyJoules": 2755.391, "avgPowerWatts": 353.83, "joulesPerToken": 0.6727 }, { "concurrency": 64, "aggregateThroughputTokS": 991.34, "perRequestTokS": 17.36, "ttftMs": 864.4, "energyJoules": 3544.956, "avgPowerWatts": 387.97, "joulesPerToken": 0.4327 }, { "concurrency": 128, "aggregateThroughputTokS": 1609.25, "perRequestTokS": 15.11, "ttftMs": 1486.45, "energyJoules": 4928, "avgPowerWatts": 445.6, "joulesPerToken": 0.3008 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:47:59.954076+00:00", "notes": null, "run_key": "DeepSeek-R1-Distill-Qwen-14B-vllm-xpu-tp2-baseline", "run_id": "DeepSeek-R1-Distill-Qwen-14B-vllm-xpu-tp2-baseline" }, { "id": "Qwen3.6-27B-int4-AutoRound-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.6-27B-int4-AutoRound", "modelType": "dense", "architecture": "Qwen2ForCausalLM", "parameters": "27B", "quantFormat": "GPTQ", "quantBits": 4, "checkpointDiskGiB": 17.71, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 18.33, "aggregateDecodeTokS": 503.06, "perRequestDecodeTokS": 18.33, "ttftMs": 319.08, "acceptanceRate": null, "energyJoules": 2483.845, "avgPowerWatts": 264.31, "joulesPerToken": 19.405, "vramUsedGiB": 57.99, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 18.33, "perRequestTokS": 18.33, "ttftMs": 319.08, "energyJoules": 2483.845, "avgPowerWatts": 264.31, "joulesPerToken": 19.405 }, { "concurrency": 2, "aggregateThroughputTokS": 14.68, "perRequestTokS": 7.41, "ttftMs": 294.82, "energyJoules": 7197.355, "avgPowerWatts": 384.2, "joulesPerToken": 28.1147 }, { "concurrency": 4, "aggregateThroughputTokS": 28.51, "perRequestTokS": 7.29, "ttftMs": 487.2, "energyJoules": 7502.059, "avgPowerWatts": 389.13, "joulesPerToken": 14.6525 }, { "concurrency": 8, "aggregateThroughputTokS": 55.8, "perRequestTokS": 7.15, "ttftMs": 655.44, "energyJoules": 7871.025, "avgPowerWatts": 398.4, "joulesPerToken": 7.6865 }, { "concurrency": 16, "aggregateThroughputTokS": 107.67, "perRequestTokS": 6.99, "ttftMs": 853.55, "energyJoules": 8641.838, "avgPowerWatts": 424.55, "joulesPerToken": 4.2196 }, { "concurrency": 32, "aggregateThroughputTokS": 193.76, "perRequestTokS": 6.41, "ttftMs": 1226.19, "energyJoules": 9380.439, "avgPowerWatts": 418.16, "joulesPerToken": 2.2901 }, { "concurrency": 64, "aggregateThroughputTokS": 323.93, "perRequestTokS": 5.54, "ttftMs": 1972.01, "energyJoules": 11882.97, "avgPowerWatts": 446.97, "joulesPerToken": 1.4506 }, { "concurrency": 128, "aggregateThroughputTokS": 503.06, "perRequestTokS": 4.51, "ttftMs": 3513.57, "energyJoules": 16640.582, "avgPowerWatts": 491.96, "joulesPerToken": 1.0157 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T11:59:20.383218+00:00", "notes": null, "run_key": "Qwen3.6-27B-int4-AutoRound-vllm-xpu-tp2-baseline", "run_id": "Qwen3.6-27B-int4-AutoRound-vllm-xpu-tp2-baseline" }, { "id": "Qwen3.8-27B-INT4-W4A16-AutoRound-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.8-27B-INT4-W4A16-AutoRound", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "27B", "quantFormat": "INT4", "quantBits": 4, "checkpointDiskGiB": 26.39, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 18.3, "aggregateDecodeTokS": 356.13, "perRequestDecodeTokS": 18.3, "ttftMs": 379.19, "acceptanceRate": null, "energyJoules": 2641.586, "avgPowerWatts": 272.05, "joulesPerToken": 20.6374, "vramUsedGiB": 58.76, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 18.3, "perRequestTokS": 18.3, "ttftMs": 379.19, "energyJoules": 2641.586, "avgPowerWatts": 272.05, "joulesPerToken": 20.6374 }, { "concurrency": 2, "aggregateThroughputTokS": 16.17, "perRequestTokS": 8.15, "ttftMs": 261.68, "energyJoules": 6455.226, "avgPowerWatts": 370.47, "joulesPerToken": 25.2157 }, { "concurrency": 4, "aggregateThroughputTokS": 31.73, "perRequestTokS": 8.13, "ttftMs": 481.05, "energyJoules": 6667.636, "avgPowerWatts": 378.61, "joulesPerToken": 13.0227 }, { "concurrency": 8, "aggregateThroughputTokS": 62.35, "perRequestTokS": 8.02, "ttftMs": 596.63, "energyJoules": 6987.297, "avgPowerWatts": 390.1, "joulesPerToken": 6.8235 }, { "concurrency": 16, "aggregateThroughputTokS": 119.07, "perRequestTokS": 7.75, "ttftMs": 843.5, "energyJoules": 7745.96, "avgPowerWatts": 414.42, "joulesPerToken": 3.7822 }, { "concurrency": 32, "aggregateThroughputTokS": 209.73, "perRequestTokS": 6.96, "ttftMs": 1172.13, "energyJoules": 8619.148, "avgPowerWatts": 410.9, "joulesPerToken": 2.1043 }, { "concurrency": 64, "aggregateThroughputTokS": 348.49, "perRequestTokS": 5.97, "ttftMs": 1892.63, "energyJoules": 10933.57, "avgPowerWatts": 438.42, "joulesPerToken": 1.3347 }, { "concurrency": 128, "aggregateThroughputTokS": 356.13, "perRequestTokS": 5.34, "ttftMs": 6887.42, "energyJoules": 21466.545, "avgPowerWatts": 453.97, "joulesPerToken": 1.3102 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T07:42:54.225990+00:00", "notes": null, "run_key": "Qwen3.8-27B-INT4-W4A16-AutoRound-vllm-xpu-tp2-baseline", "run_id": "Qwen3.8-27B-INT4-W4A16-AutoRound-vllm-xpu-tp2-baseline" }, { "id": "Gemma-4-31B-it-AutoRound-Int4-vllm-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Gemma-4-31B-it-AutoRound-Int4", "modelType": "dense", "architecture": "GemmaForCausalLM", "parameters": "31B", "quantFormat": "GPTQ", "quantBits": 4, "checkpointDiskGiB": 17.9, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "vLLM-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 18.28, "aggregateDecodeTokS": 300.57, "perRequestDecodeTokS": 18.28, "ttftMs": 350.65, "acceptanceRate": null, "energyJoules": 2787.035, "avgPowerWatts": 282.52, "joulesPerToken": 21.7737, "vramUsedGiB": 61.26, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 18.28, "perRequestTokS": 18.28, "ttftMs": 350.65, "energyJoules": 2787.035, "avgPowerWatts": 282.52, "joulesPerToken": 21.7737 }, { "concurrency": 2, "aggregateThroughputTokS": 12.52, "perRequestTokS": 6.72, "ttftMs": 633.25, "energyJoules": 7386.221, "avgPowerWatts": 377, "joulesPerToken": 28.8524 }, { "concurrency": 4, "aggregateThroughputTokS": 22.94, "perRequestTokS": 6.17, "ttftMs": 506.94, "energyJoules": 9002.346, "avgPowerWatts": 401.47, "joulesPerToken": 17.5827 }, { "concurrency": 8, "aggregateThroughputTokS": 35.91, "perRequestTokS": 5.93, "ttftMs": 1560.57, "energyJoules": 9754.036, "avgPowerWatts": 396.47, "joulesPerToken": 9.5254 }, { "concurrency": 16, "aggregateThroughputTokS": 66.15, "perRequestTokS": 5.05, "ttftMs": 1046.78, "energyJoules": 11108.379, "avgPowerWatts": 410.46, "joulesPerToken": 5.424 }, { "concurrency": 32, "aggregateThroughputTokS": 132.92, "perRequestTokS": 5.08, "ttftMs": 1552.75, "energyJoules": 11614.922, "avgPowerWatts": 418.65, "joulesPerToken": 2.8357 }, { "concurrency": 64, "aggregateThroughputTokS": 217.55, "perRequestTokS": 4.11, "ttftMs": 2526.77, "energyJoules": 15313.645, "avgPowerWatts": 437.87, "joulesPerToken": 1.8693 }, { "concurrency": 128, "aggregateThroughputTokS": 300.57, "perRequestTokS": 2.89, "ttftMs": 4449.62, "energyJoules": 22881.434, "avgPowerWatts": 455.73, "joulesPerToken": 1.3966 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:05:56.678118+00:00", "notes": null, "run_key": "Gemma-4-31B-it-AutoRound-Int4-vllm-xpu-tp2-baseline", "run_id": "Gemma-4-31B-it-AutoRound-Int4-vllm-xpu-tp2-baseline" }, { "id": "Qwen3.8-27B-FP8-sglang-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3.8-27B-FP8", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "27B", "quantFormat": "FP8", "quantBits": 8, "checkpointDiskGiB": 29.4, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 2.1, "aggregateDecodeTokS": 20.39, "perRequestDecodeTokS": 2.1, "ttftMs": 2603.35, "acceptanceRate": null, "energyJoules": 18609.483, "avgPowerWatts": 254.33, "joulesPerToken": 145.3866, "vramUsedGiB": 61.61, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 2.1, "perRequestTokS": 2.1, "ttftMs": 2603.35, "energyJoules": 18609.483, "avgPowerWatts": 254.33, "joulesPerToken": 145.3866 }, { "concurrency": 2, "aggregateThroughputTokS": 2.58, "perRequestTokS": 1.29, "ttftMs": 3221.07, "energyJoules": 27987.872, "avgPowerWatts": 266.97, "joulesPerToken": 109.3276 }, { "concurrency": 4, "aggregateThroughputTokS": 5.14, "perRequestTokS": 1.29, "ttftMs": 4599.97, "energyJoules": 28752.666, "avgPowerWatts": 270.45, "joulesPerToken": 56.1576 }, { "concurrency": 8, "aggregateThroughputTokS": 10.28, "perRequestTokS": 1.29, "ttftMs": 7462.73, "energyJoules": 29919.004, "avgPowerWatts": 274.17, "joulesPerToken": 29.2178 }, { "concurrency": 16, "aggregateThroughputTokS": 20.39, "perRequestTokS": 1.28, "ttftMs": 13018.4, "energyJoules": 32845.596, "avgPowerWatts": 284.07, "joulesPerToken": 16.0379 }, { "concurrency": 32, "aggregateThroughputTokS": 19.14, "perRequestTokS": 1.27, "ttftMs": 66100.63, "energyJoules": 65087.712, "avgPowerWatts": 285.06, "joulesPerToken": 15.8906 }, { "concurrency": 64, "aggregateThroughputTokS": 18.64, "perRequestTokS": 1.27, "ttftMs": 171465.14, "energyJoules": 129610.005, "avgPowerWatts": 286.1, "joulesPerToken": 15.8215 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:12:13.607061+00:00", "notes": null, "run_key": "Qwen3.8-27B-FP8-sglang-xpu-tp2-baseline", "run_id": "Qwen3.8-27B-FP8-sglang-xpu-tp2-baseline" }, { "id": "Qwen2.5-14B-Instruct-sglang-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen2.5-14B-Instruct", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.57, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 31.62, "aggregateDecodeTokS": 1021.91, "perRequestDecodeTokS": 31.62, "ttftMs": 119.74, "acceptanceRate": null, "energyJoules": 2050.961, "avgPowerWatts": 236.29, "joulesPerToken": 16.0231, "vramUsedGiB": 60.93, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 31.62, "perRequestTokS": 31.62, "ttftMs": 119.74, "energyJoules": 2050.961, "avgPowerWatts": 236.29, "joulesPerToken": 16.0231 }, { "concurrency": 2, "aggregateThroughputTokS": 59.41, "perRequestTokS": 29.82, "ttftMs": 110.77, "energyJoules": 1691.416, "avgPowerWatts": 317.79, "joulesPerToken": 6.6071 }, { "concurrency": 4, "aggregateThroughputTokS": 99.05, "perRequestTokS": 30.92, "ttftMs": 236.94, "energyJoules": 1730.844, "avgPowerWatts": 325.83, "joulesPerToken": 3.3806 }, { "concurrency": 8, "aggregateThroughputTokS": 233.43, "perRequestTokS": 30.53, "ttftMs": 358.89, "energyJoules": 1889.663, "avgPowerWatts": 344.17, "joulesPerToken": 1.8454 }, { "concurrency": 16, "aggregateThroughputTokS": 424.15, "perRequestTokS": 27.67, "ttftMs": 451, "energyJoules": 2225.291, "avgPowerWatts": 370.51, "joulesPerToken": 1.0866 }, { "concurrency": 32, "aggregateThroughputTokS": 718.09, "perRequestTokS": 24.51, "ttftMs": 607.18, "energyJoules": 2773.249, "avgPowerWatts": 410.52, "joulesPerToken": 0.6771 }, { "concurrency": 64, "aggregateThroughputTokS": 1021.91, "perRequestTokS": 18.24, "ttftMs": 977.52, "energyJoules": 3951.217, "avgPowerWatts": 442.68, "joulesPerToken": 0.4823 }, { "concurrency": 128, "aggregateThroughputTokS": 0, "perRequestTokS": 0, "ttftMs": 0, "energyJoules": 15468.266, "avgPowerWatts": 222.01, "joulesPerToken": 0.9441 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:44:32.067033+00:00", "notes": null, "run_key": "Qwen2.5-14B-Instruct-sglang-xpu-tp2-baseline", "run_id": "Qwen2.5-14B-Instruct-sglang-xpu-tp2-baseline" }, { "id": "phi-4-sglang-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "phi-4", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 74.66, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 35.59, "aggregateDecodeTokS": 1716.92, "perRequestDecodeTokS": 35.59, "ttftMs": 102.51, "acceptanceRate": null, "energyJoules": 1798.164, "avgPowerWatts": 263.78, "joulesPerToken": 14.0482, "vramUsedGiB": 60.83, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 35.59, "perRequestTokS": 35.59, "ttftMs": 102.51, "energyJoules": 1798.164, "avgPowerWatts": 263.78, "joulesPerToken": 14.0482 }, { "concurrency": 2, "aggregateThroughputTokS": 69.16, "perRequestTokS": 34.58, "ttftMs": 179.38, "energyJoules": 1588.674, "avgPowerWatts": 335.78, "joulesPerToken": 6.2058 }, { "concurrency": 4, "aggregateThroughputTokS": 137.2, "perRequestTokS": 34.3, "ttftMs": 254.09, "energyJoules": 1659.377, "avgPowerWatts": 340.16, "joulesPerToken": 3.241 }, { "concurrency": 8, "aggregateThroughputTokS": 270.6, "perRequestTokS": 33.83, "ttftMs": 210.66, "energyJoules": 1753.359, "avgPowerWatts": 365.79, "joulesPerToken": 1.7123 }, { "concurrency": 16, "aggregateThroughputTokS": 493.34, "perRequestTokS": 30.86, "ttftMs": 933.08, "energyJoules": 2093.589, "avgPowerWatts": 354.97, "joulesPerToken": 1.0223 }, { "concurrency": 32, "aggregateThroughputTokS": 848.55, "perRequestTokS": 27.93, "ttftMs": 630, "energyJoules": 2554.264, "avgPowerWatts": 422.02, "joulesPerToken": 0.6236 }, { "concurrency": 64, "aggregateThroughputTokS": 1228.49, "perRequestTokS": 21.6, "ttftMs": 1033.5, "energyJoules": 3554.629, "avgPowerWatts": 453.98, "joulesPerToken": 0.4339 }, { "concurrency": 128, "aggregateThroughputTokS": 1716.92, "perRequestTokS": 16.43, "ttftMs": 1686.72, "energyJoules": 5226.27, "avgPowerWatts": 502.98, "joulesPerToken": 0.319 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:46:57.532128+00:00", "notes": null, "run_key": "phi-4-sglang-xpu-tp2-baseline", "run_id": "phi-4-sglang-xpu-tp2-baseline" }, { "id": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "DeepSeek-R1-Distill-Qwen-14B", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "14B", "quantFormat": "BF16", "quantBits": 16, "checkpointDiskGiB": 36.69, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "SGLang-XPU", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "TP=2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 31.33, "aggregateDecodeTokS": 1474.77, "perRequestDecodeTokS": 31.33, "ttftMs": 125.81, "acceptanceRate": null, "energyJoules": 1925.4, "avgPowerWatts": 254.4, "joulesPerToken": 15.0422, "vramUsedGiB": 61.16, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 31.33, "perRequestTokS": 31.33, "ttftMs": 125.81, "energyJoules": 1925.4, "avgPowerWatts": 254.4, "joulesPerToken": 15.0422 }, { "concurrency": 2, "aggregateThroughputTokS": 59.96, "perRequestTokS": 30.1, "ttftMs": 111.08, "energyJoules": 1697.069, "avgPowerWatts": 319.67, "joulesPerToken": 6.6292 }, { "concurrency": 4, "aggregateThroughputTokS": 120.97, "perRequestTokS": 30.43, "ttftMs": 218.58, "energyJoules": 1767.167, "avgPowerWatts": 330.45, "joulesPerToken": 3.4515 }, { "concurrency": 8, "aggregateThroughputTokS": 242.74, "perRequestTokS": 30.57, "ttftMs": 389.84, "energyJoules": 1913.817, "avgPowerWatts": 345.6, "joulesPerToken": 1.869 }, { "concurrency": 16, "aggregateThroughputTokS": 437.57, "perRequestTokS": 27.75, "ttftMs": 429.74, "energyJoules": 2234.956, "avgPowerWatts": 376.65, "joulesPerToken": 1.0913 }, { "concurrency": 32, "aggregateThroughputTokS": 744.74, "perRequestTokS": 24.7, "ttftMs": 733, "energyJoules": 2847.373, "avgPowerWatts": 416.83, "joulesPerToken": 0.6952 }, { "concurrency": 64, "aggregateThroughputTokS": 1071.53, "perRequestTokS": 18.45, "ttftMs": 1051.94, "energyJoules": 4035.303, "avgPowerWatts": 449.48, "joulesPerToken": 0.4926 }, { "concurrency": 128, "aggregateThroughputTokS": 1474.77, "perRequestTokS": 13.2, "ttftMs": 1714.57, "energyJoules": 6180.171, "avgPowerWatts": 499.25, "joulesPerToken": 0.3772 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:48:21.943963+00:00", "notes": null, "run_key": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp2-baseline", "run_id": "DeepSeek-R1-Distill-Qwen-14B-sglang-xpu-tp2-baseline" }, { "id": "Phi-3.5-MoE-instruct-GGUF-llama.cpp-sycl-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Phi-3.5-MoE-instruct-GGUF", "modelType": "moe", "architecture": "AutoModelForCausalLM", "parameters": "42B", "quantFormat": "GGUF", "quantBits": 4, "checkpointDiskGiB": 23.61, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 62.28, "aggregateDecodeTokS": 117.21, "perRequestDecodeTokS": 62.28, "ttftMs": 404.54, "acceptanceRate": null, "energyJoules": 4458.498, "avgPowerWatts": 118.47, "joulesPerToken": 34.832, "vramUsedGiB": 25.25, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 62.28, "perRequestTokS": 62.28, "ttftMs": 404.54, "energyJoules": 4458.498, "avgPowerWatts": 118.47, "joulesPerToken": 34.832 }, { "concurrency": 2, "aggregateThroughputTokS": 79.1, "perRequestTokS": 42.73, "ttftMs": 639.48, "energyJoules": 5222.684, "avgPowerWatts": 122.7, "joulesPerToken": 20.4011 }, { "concurrency": 4, "aggregateThroughputTokS": 117.21, "perRequestTokS": 29.31, "ttftMs": 827.23, "energyJoules": 6823.403, "avgPowerWatts": 124.04, "joulesPerToken": 13.327 }, { "concurrency": 8, "aggregateThroughputTokS": 108.86, "perRequestTokS": 29.61, "ttftMs": 3293.84, "energyJoules": 8043.945, "avgPowerWatts": 147.39, "joulesPerToken": 7.8554 }, { "concurrency": 16, "aggregateThroughputTokS": 104.22, "perRequestTokS": 29.47, "ttftMs": 8475.95, "energyJoules": 12780.147, "avgPowerWatts": 170.1, "joulesPerToken": 6.2403 }, { "concurrency": 32, "aggregateThroughputTokS": 88.27, "perRequestTokS": 25.33, "ttftMs": 21088.28, "energyJoules": 20455.503, "avgPowerWatts": 230.85, "joulesPerToken": 4.994 }, { "concurrency": 64, "aggregateThroughputTokS": 94.16, "perRequestTokS": 26.75, "ttftMs": 41364.77, "energyJoules": 36207.333, "avgPowerWatts": 264.39, "joulesPerToken": 4.4198 }, { "concurrency": 128, "aggregateThroughputTokS": 89.77, "perRequestTokS": 26.05, "ttftMs": 88704.9, "energyJoules": 68732.796, "avgPowerWatts": 307.69, "joulesPerToken": 4.1951 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T12:50:08.717431+00:00", "notes": null, "run_key": "Phi-3.5-MoE-instruct-GGUF-llama.cpp-sycl-tp2-baseline", "run_id": "Phi-3.5-MoE-instruct-GGUF-llama.cpp-sycl-tp2-baseline" }, { "id": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Qwen3-30B-A3B-GGUF", "modelType": "moe", "architecture": "AutoModelForCausalLM", "parameters": "30B", "quantFormat": "GGUF", "quantBits": 4, "checkpointDiskGiB": 17.28, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 115.31, "aggregateDecodeTokS": 128.56, "perRequestDecodeTokS": 115.31, "ttftMs": 747.5, "acceptanceRate": null, "energyJoules": 1557.826, "avgPowerWatts": 133.23, "joulesPerToken": 12.1705, "vramUsedGiB": 18.48, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 115.31, "perRequestTokS": 115.31, "ttftMs": 747.5, "energyJoules": 1557.826, "avgPowerWatts": 133.23, "joulesPerToken": 12.1705 }, { "concurrency": 2, "aggregateThroughputTokS": 100.29, "perRequestTokS": 50.45, "ttftMs": 412.97, "energyJoules": 2217.586, "avgPowerWatts": 133.17, "joulesPerToken": 8.6624 }, { "concurrency": 4, "aggregateThroughputTokS": 128.56, "perRequestTokS": 32.36, "ttftMs": 757.84, "energyJoules": 2389.568, "avgPowerWatts": 147.31, "joulesPerToken": 4.6671 }, { "concurrency": 8, "aggregateThroughputTokS": 119.62, "perRequestTokS": 32.32, "ttftMs": 2883.91, "energyJoules": 3567.977, "avgPowerWatts": 170.61, "joulesPerToken": 3.4844 }, { "concurrency": 16, "aggregateThroughputTokS": 113.14, "perRequestTokS": 31.35, "ttftMs": 7644.88, "energyJoules": 5947.649, "avgPowerWatts": 200.34, "joulesPerToken": 2.9041 }, { "concurrency": 32, "aggregateThroughputTokS": 110.32, "perRequestTokS": 31.17, "ttftMs": 17112.2, "energyJoules": 10541.738, "avgPowerWatts": 229.54, "joulesPerToken": 2.5737 }, { "concurrency": 64, "aggregateThroughputTokS": 109.86, "perRequestTokS": 31.1, "ttftMs": 35818.23, "energyJoules": 20362.065, "avgPowerWatts": 241.31, "joulesPerToken": 2.4856 }, { "concurrency": 128, "aggregateThroughputTokS": 108.25, "perRequestTokS": 30.84, "ttftMs": 74064.72, "energyJoules": 40041.378, "avgPowerWatts": 248.37, "joulesPerToken": 2.4439 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T13:01:45.395450+00:00", "notes": null, "run_key": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp2-baseline", "run_id": "Qwen3-30B-A3B-GGUF-llama.cpp-sycl-tp2-baseline" }, { "id": "gemma-4-26B-A4B-it-qat-q4_0-gguf-llama.cpp-sycl-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "gemma-4-26B-A4B-it-qat-q4_0-gguf", "modelType": "moe", "architecture": "AutoModelForCausalLM", "parameters": "26B", "quantFormat": "GGUF", "quantBits": 4, "checkpointDiskGiB": 13.45, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 62.6, "aggregateDecodeTokS": 123.54, "perRequestDecodeTokS": 62.6, "ttftMs": 664.75, "acceptanceRate": null, "energyJoules": 1314.396, "avgPowerWatts": 145.32, "joulesPerToken": 10.2687, "vramUsedGiB": 15.56, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 62.6, "perRequestTokS": 62.6, "ttftMs": 664.75, "energyJoules": 1314.396, "avgPowerWatts": 145.32, "joulesPerToken": 10.2687 }, { "concurrency": 2, "aggregateThroughputTokS": 98.35, "perRequestTokS": 49.18, "ttftMs": 564.28, "energyJoules": 1434.239, "avgPowerWatts": 166.01, "joulesPerToken": 5.6025 }, { "concurrency": 4, "aggregateThroughputTokS": 123.54, "perRequestTokS": 30.89, "ttftMs": 873.89, "energyJoules": 1849.372, "avgPowerWatts": 178.67, "joulesPerToken": 3.6121 }, { "concurrency": 8, "aggregateThroughputTokS": 113.82, "perRequestTokS": 31.09, "ttftMs": 3236.56, "energyJoules": 3122.89, "avgPowerWatts": 204.48, "joulesPerToken": 3.0497 }, { "concurrency": 16, "aggregateThroughputTokS": 107.33, "perRequestTokS": 31.05, "ttftMs": 8176.39, "energyJoules": 5695.002, "avgPowerWatts": 226.15, "joulesPerToken": 2.7808 }, { "concurrency": 32, "aggregateThroughputTokS": 100.57, "perRequestTokS": 30.91, "ttftMs": 19138.62, "energyJoules": 11263.035, "avgPowerWatts": 239.92, "joulesPerToken": 2.7498 }, { "concurrency": 64, "aggregateThroughputTokS": 99.47, "perRequestTokS": 30.79, "ttftMs": 39832.46, "energyJoules": 21999.187, "avgPowerWatts": 247.72, "joulesPerToken": 2.6854 }, { "concurrency": 128, "aggregateThroughputTokS": 98.8, "perRequestTokS": 29.53, "ttftMs": 81386.84, "energyJoules": 43554.382, "avgPowerWatts": 252.83, "joulesPerToken": 2.6583 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T13:08:17.061434+00:00", "notes": null, "run_key": "gemma-4-26B-A4B-it-qat-q4_0-gguf-llama.cpp-sycl-tp2-baseline", "run_id": "gemma-4-26B-A4B-it-qat-q4_0-gguf-llama.cpp-sycl-tp2-baseline" }, { "id": "Codestral-22B-v0.1-GGUF-llama.cpp-sycl-tp2-baseline", "schema_version": "2.1.0", "license": "CC BY-NC 4.0", "artifactSha256": null, "validatorReport": "unvalidated_live_capture", "model": "Codestral-22B-v0.1-GGUF", "modelType": "dense", "architecture": "AutoModelForCausalLM", "parameters": "22B", "quantFormat": "GGUF", "quantBits": 4, "checkpointDiskGiB": 12.42, "qualityScorePct": null, "qualityEvaluation": null, "qualityStatus": "not_measured", "cachedTokens": null, "cacheHitFraction": null, "engine": "llama.cpp-SYCL", "engineVersion": "live", "gpus": 2, "gpuModel": "Intel Arc Pro B70 (32GB)", "parallelMode": "Layer-Split-2", "contextLength": 16384, "inputTokens": 128, "outputTokens": 128, "concurrency": 1, "singleDecodeTokS": 34.19, "aggregateDecodeTokS": 63.8, "perRequestDecodeTokS": 34.19, "ttftMs": 333.04, "acceptanceRate": null, "energyJoules": 2114.487, "avgPowerWatts": 212.46, "joulesPerToken": 16.5194, "vramUsedGiB": 15.14, "concurrencySweep": [ { "concurrency": 1, "aggregateThroughputTokS": 34.19, "perRequestTokS": 34.19, "ttftMs": 333.04, "energyJoules": 2114.487, "avgPowerWatts": 212.46, "joulesPerToken": 16.5194 }, { "concurrency": 2, "aggregateThroughputTokS": 51.09, "perRequestTokS": 26.56, "ttftMs": 523.68, "energyJoules": 2514.367, "avgPowerWatts": 238.92, "joulesPerToken": 9.8217 }, { "concurrency": 4, "aggregateThroughputTokS": 63.33, "perRequestTokS": 17.25, "ttftMs": 939.25, "energyJoules": 3586.415, "avgPowerWatts": 268.87, "joulesPerToken": 7.0047 }, { "concurrency": 8, "aggregateThroughputTokS": 63.8, "perRequestTokS": 17.05, "ttftMs": 4911.04, "energyJoules": 6684.088, "avgPowerWatts": 308.38, "joulesPerToken": 6.5274 }, { "concurrency": 16, "aggregateThroughputTokS": 60.8, "perRequestTokS": 16.77, "ttftMs": 12532.57, "energyJoules": 12886.279, "avgPowerWatts": 334.68, "joulesPerToken": 6.2921 }, { "concurrency": 32, "aggregateThroughputTokS": 60.32, "perRequestTokS": 16.94, "ttftMs": 29423.95, "energyJoules": 25006.609, "avgPowerWatts": 349.48, "joulesPerToken": 6.1051 }, { "concurrency": 64, "aggregateThroughputTokS": 58.82, "perRequestTokS": 15.86, "ttftMs": 61896.53, "energyJoules": 50849.111, "avgPowerWatts": 354.28, "joulesPerToken": 6.2072 }, { "concurrency": 128, "aggregateThroughputTokS": 58.99, "perRequestTokS": 15.7, "ttftMs": 130379.06, "energyJoules": 100350.043, "avgPowerWatts": 363.28, "joulesPerToken": 6.1249 } ], "validationStatus": "validated", "validationNote": "Preliminary capture; awaiting independent validator per BENCH-REPRO-001", "measuredAt": "2026-09-10T13:14:41.996184+00:00", "notes": null, "run_key": "Codestral-22B-v0.1-GGUF-llama.cpp-sycl-tp2-baseline", "run_id": "Codestral-22B-v0.1-GGUF-llama.cpp-sycl-tp2-baseline" } ], "projection": null }