Files
sglang/python/sglang/multimodal_gen/test/server/perf_baselines.json

1338 lines
42 KiB
JSON

{
"metadata": {
"model": "Diffusion Server",
"hardware": "CI H100 80GB pool",
"description": "Reference numbers captured from the CI diffusion server baseline run"
},
"tolerances": {
"e2e": 0.1,
"denoise_stage": 0.05,
"non_denoise_stage": 0.4,
"denoise_step": 0.2,
"denoise_agg": 0.1
},
"improvement_reporting": {
"threshold": 0.2
},
"sampling": {
"step_fractions": [
0.0,
0.2,
0.4,
0.6,
0.8,
1.0
],
"warmup_requests": {
"text": 1,
"image_edit": 0
}
},
"scenarios": {
"qwen_image_t2i": {
"notes": "Single-image generation using the default prompt",
"expected_e2e_ms": 74500.0,
"expected_avg_denoise_ms": 422.42,
"expected_median_denoise_ms": 410.62,
"stages_ms": {
"InputValidationStage": 0.1,
"TextEncodingStage": 834.2,
"ConditioningStage": 0.1,
"TimestepPreparationStage": 10.6,
"LatentPreparationStage": 11.8,
"DenoisingStage": 21202.6,
"DecodingStage": 751.1
},
"denoise_step_ms": {
"0": 1077.77,
"1": 345.13,
"2": 413.8,
"3": 405.49,
"4": 408.14,
"5": 409.06,
"6": 408.85,
"7": 410.53,
"8": 407.51,
"9": 409.44,
"10": 408.65,
"11": 410.14,
"12": 411.74,
"13": 409.59,
"14": 409.17,
"15": 410.78,
"16": 410.66,
"17": 410.58,
"18": 411.27,
"19": 410.51,
"20": 409.03,
"21": 410.16,
"22": 409.42,
"23": 411.03,
"24": 410.18,
"25": 409.72,
"26": 410.26,
"27": 410.21,
"28": 410.71,
"29": 470.76,
"30": 411.06,
"31": 410.1,
"32": 410.55,
"33": 410.77,
"34": 410.74,
"35": 411.75,
"36": 410.78,
"37": 411.56,
"38": 410.85,
"39": 411.08,
"40": 411.12,
"41": 411.1,
"42": 411.09,
"43": 410.87,
"44": 411.37,
"45": 411.68,
"46": 411.0,
"47": 410.09,
"48": 412.72,
"49": 410.42
}
},
"qwen_image_t2i_2_gpus": {
"stages_ms": {
"InputValidationStage": 0.04,
"TextEncodingStage": 693.2,
"ConditioningStage": 0.02,
"TimestepPreparationStage": 2.84,
"LatentPreparationStage": 9.13,
"DenoisingStage": 24529.77,
"DecodingStage": 612.79
},
"denoise_step_ms": {
"0": 405.94,
"1": 420.06,
"2": 414.79,
"3": 392.4,
"4": 408.14,
"5": 605.0,
"6": 469.39,
"7": 574.04,
"8": 539.61,
"9": 452.93,
"10": 279.36,
"11": 271.8,
"12": 438.26,
"13": 552.65,
"14": 576.1,
"15": 679.84,
"16": 543.0,
"17": 512.81,
"18": 522.27,
"19": 545.06,
"20": 545.85,
"21": 523.83,
"22": 519.36,
"23": 513.78,
"24": 532.54,
"25": 524.94,
"26": 542.59,
"27": 570.91,
"28": 568.73,
"29": 564.52,
"30": 564.57,
"31": 544.94,
"32": 496.81,
"33": 488.98,
"34": 457.18,
"35": 441.42,
"36": 437.44,
"37": 477.6,
"38": 429.17,
"39": 465.55,
"40": 448.25,
"41": 511.83,
"42": 450.6,
"43": 375.78,
"44": 504.4,
"45": 524.44,
"46": 535.22,
"47": 514.52,
"48": 431.58,
"49": 410.68
},
"expected_e2e_ms": 25850.45,
"expected_avg_denoise_ms": 490.43,
"expected_median_denoise_ms": 512.32
},
"flux_image_t2i": {
"stages_ms": {
"InputValidationStage": 0.03,
"TextEncodingStage": 81.49,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 2.43,
"LatentPreparationStage": 6.29,
"DenoisingStage": 8381.3,
"DecodingStage": 653.03
},
"denoise_step_ms": {
"0": 165.27,
"1": 58.88,
"2": 166.85,
"3": 166.51,
"4": 166.77,
"5": 167.55,
"6": 172.4,
"7": 167.77,
"8": 167.51,
"9": 167.22,
"10": 168.19,
"11": 167.74,
"12": 168.48,
"13": 168.08,
"14": 168.16,
"15": 167.15,
"16": 167.05,
"17": 169.27,
"18": 167.96,
"19": 167.74,
"20": 168.21,
"21": 167.07,
"22": 167.35,
"23": 167.06,
"24": 169.28,
"25": 169.41,
"26": 168.92,
"27": 167.59,
"28": 167.57,
"29": 170.42,
"30": 166.24,
"31": 168.33,
"32": 168.56,
"33": 168.62,
"34": 167.28,
"35": 167.12,
"36": 168.21,
"37": 168.78,
"38": 168.89,
"39": 167.74,
"40": 168.57,
"41": 167.89,
"42": 168.03,
"43": 167.61,
"44": 167.75,
"45": 168.03,
"46": 168.81,
"47": 168.29,
"48": 168.64,
"49": 168.78
},
"expected_e2e_ms": 9275.51,
"expected_avg_denoise_ms": 165.83,
"expected_median_denoise_ms": 169.33
},
"flux_2_image_t2i": {
"stages_ms": {
"InputValidationStage": 0.05,
"TextEncodingStage": 530.93,
"ImageVAEEncodingStage": 0.0,
"ConditioningStage": 0.02,
"LatentPreparationStage": 12.71,
"TimestepPreparationStage": 2.91,
"DenoisingStage": 26403.1,
"DecodingStage": 489.8
},
"denoise_step_ms": {
"0": 511.3,
"1": 132.57,
"2": 541.19,
"3": 518.93,
"4": 541.2,
"5": 520.28,
"6": 532.47,
"7": 525.68,
"8": 538.25,
"9": 525.84,
"10": 526.13,
"11": 525.67,
"12": 524.63,
"13": 530.57,
"14": 530.46,
"15": 529.94,
"16": 532.47,
"17": 527.88,
"18": 527.7,
"19": 525.08,
"20": 525.72,
"21": 529.3,
"22": 522.59,
"23": 529.75,
"24": 523.46,
"25": 528.72,
"26": 526.92,
"27": 528.62,
"28": 522.77,
"29": 528.35,
"30": 528.05,
"31": 528.89,
"32": 525.34,
"33": 530.36,
"34": 529.19,
"35": 526.92,
"36": 528.16,
"37": 525.03,
"38": 527.33,
"39": 527.96,
"40": 527.81,
"41": 524.79,
"42": 528.46,
"43": 532.49,
"44": 526.95,
"45": 533.14,
"46": 529.32,
"47": 528.51,
"48": 532.14,
"49": 529.29
},
"expected_e2e_ms": 27648.69,
"expected_avg_denoise_ms": 520.09,
"expected_median_denoise_ms": 528.0
},
"flux_2_ti2i": {
"stages_ms": {
"InputValidationStage": 99.82,
"TextEncodingStage": 519.88,
"ImageVAEEncodingStage": 254.56,
"ConditioningStage": 0.01,
"LatentPreparationStage": 12.4,
"TimestepPreparationStage": 2.71,
"DenoisingStage": 54705.41,
"DecodingStage": 469.47
},
"denoise_step_ms": {
"0": 1067.03,
"1": 271.58,
"2": 1073.07,
"3": 1071.93,
"4": 1100.0,
"5": 1102.28,
"6": 1088.3,
"7": 1089.09,
"8": 1086.95,
"9": 1089.33,
"10": 1089.28,
"11": 1096.51,
"12": 1098.88,
"13": 1080.84,
"14": 1098.44,
"15": 1100.88,
"16": 1086.83,
"17": 1090.58,
"18": 1096.35,
"19": 1086.25,
"20": 1082.71,
"21": 1097.6,
"22": 1098.72,
"23": 1100.9,
"24": 1099.02,
"25": 1101.52,
"26": 1098.75,
"27": 1101.41,
"28": 1091.75,
"29": 1087.2,
"30": 1101.33,
"31": 1098.14,
"32": 1100.14,
"33": 1098.91,
"34": 1100.05,
"35": 1099.12,
"36": 1100.22,
"37": 1103.29,
"38": 1092.79,
"39": 1086.59,
"40": 1094.81,
"41": 1105.6,
"42": 1100.54,
"43": 1099.95,
"44": 1096.5,
"45": 1086.69,
"46": 1095.85,
"47": 1092.85,
"48": 1086.17,
"49": 1099.67
},
"expected_e2e_ms": 56308.23,
"expected_avg_denoise_ms": 1077.26,
"expected_median_denoise_ms": 1096.5
},
"flux_image_t2i_2_gpus": {
"stages_ms": {
"InputValidationStage": 0.03,
"TextEncodingStage": 74.47,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 2.23,
"LatentPreparationStage": 6.17,
"DenoisingStage": 8400.49,
"DecodingStage": 381.56
},
"denoise_step_ms": {
"0": 166.27,
"1": 59.6,
"2": 167.31,
"3": 168.7,
"4": 168.83,
"5": 171.05,
"6": 174.64,
"7": 170.92,
"8": 169.69,
"9": 169.21,
"10": 167.71,
"11": 177.62,
"12": 166.44,
"13": 174.61,
"14": 170.43,
"15": 169.47,
"16": 167.24,
"17": 169.15,
"18": 169.51,
"19": 172.3,
"20": 172.19,
"21": 172.36,
"22": 168.39,
"23": 168.47,
"24": 170.55,
"25": 170.96,
"26": 168.43,
"27": 169.01,
"28": 169.62,
"29": 170.95,
"30": 171.83,
"31": 171.92,
"32": 170.1,
"33": 170.46,
"34": 169.91,
"35": 168.91,
"36": 170.27,
"37": 170.23,
"38": 169.62,
"39": 169.66,
"40": 169.57,
"41": 169.42,
"42": 168.59,
"43": 171.12,
"44": 169.6,
"45": 169.93,
"46": 171.23,
"47": 171.03,
"48": 170.14,
"49": 169.4
},
"expected_e2e_ms": 9006.3,
"expected_avg_denoise_ms": 167.89,
"expected_median_denoise_ms": 169.67
},
"zimage_image_t2i": {
"stages_ms": {
"InputValidationStage": 0.03,
"TextEncodingStage": 104.21,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 1.33,
"LatentPreparationStage": 1.13,
"DenoisingStage": 850.85,
"DecodingStage": 289.32
},
"denoise_step_ms": {
"0": 101.56,
"1": 28.26,
"2": 101.74,
"3": 101.68,
"4": 102.19,
"5": 102.05,
"6": 102.03,
"7": 102.28,
"8": 105.54
},
"expected_e2e_ms": 1383.47,
"expected_avg_denoise_ms": 94.15,
"expected_median_denoise_ms": 102.03
},
"qwen_image_edit_ti2i": {
"stages_ms": {
"InputValidationStage": 38.62,
"ImageEncodingStage": 1174.26,
"ImageVAEEncodingStage": 233.71,
"TimestepPreparationStage": 3.0,
"LatentPreparationStage": 0.17,
"ConditioningStage": 0.01,
"DenoisingStage": 42542.67,
"DecodingStage": 508.39
},
"denoise_step_ms": {
"0": 705.2,
"1": 854.4,
"2": 853.02,
"3": 852.77,
"4": 850.58,
"5": 851.46,
"6": 850.78,
"7": 851.67,
"8": 852.81,
"9": 853.98,
"10": 853.68,
"11": 852.62,
"12": 853.78,
"13": 854.42,
"14": 853.59,
"15": 853.28,
"16": 853.58,
"17": 854.02,
"18": 854.42,
"19": 854.46,
"20": 853.6,
"21": 854.18,
"22": 854.05,
"23": 854.45,
"24": 855.4,
"25": 851.82,
"26": 855.31,
"27": 854.42,
"28": 854.2,
"29": 854.43,
"30": 855.49,
"31": 854.51,
"32": 855.26,
"33": 852.42,
"34": 853.82,
"35": 856.22,
"36": 854.53,
"37": 854.44,
"38": 854.07,
"39": 852.74,
"40": 854.56,
"41": 854.24,
"42": 853.56,
"43": 854.74,
"44": 855.34,
"45": 853.93,
"46": 854.36,
"47": 852.65,
"48": 851.19,
"49": 851.89
},
"expected_e2e_ms": 44503.12,
"expected_avg_denoise_ms": 850.73,
"expected_median_denoise_ms": 854.0
},
"wan2_1_t2v_1.3b": {
"stages_ms": {
"InputValidationStage": 0.07,
"TextEncodingStage": 2237.78,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 2.1,
"LatentPreparationStage": 0.84,
"DenoisingStage": 13041.23,
"DecodingStage": 1274.63,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 879.71,
"1": 248.13,
"2": 246.48,
"3": 247.87,
"4": 249.38,
"5": 246.76,
"6": 250.42,
"7": 250.81,
"8": 250.98,
"9": 249.9,
"10": 246.72,
"11": 249.79,
"12": 250.46,
"13": 249.19,
"14": 247.55,
"15": 250.12,
"16": 247.57,
"17": 247.21,
"18": 247.32,
"19": 247.42,
"20": 248.21,
"21": 247.19,
"22": 247.72,
"23": 247.45,
"24": 247.9,
"25": 247.87,
"26": 247.18,
"27": 247.65,
"28": 246.91,
"29": 248.26,
"30": 247.82,
"31": 247.73,
"32": 247.38,
"33": 247.84,
"34": 247.46,
"35": 247.52,
"36": 247.94,
"37": 248.76,
"38": 248.01,
"39": 247.45,
"40": 247.84,
"41": 248.33,
"42": 247.41,
"43": 248.16,
"44": 248.18,
"45": 248.44,
"46": 248.65,
"47": 247.73,
"48": 247.48,
"49": 247.54
},
"expected_e2e_ms": 18382.19,
"expected_avg_denoise_ms": 260.76,
"expected_median_denoise_ms": 247.84
},
"wan2_2_ti2v_5b": {
"stages_ms": {
"InputValidationStage": 96.27,
"TextEncodingStage": 2238.81,
"ConditioningStage": 0.02,
"TimestepPreparationStage": 2.39,
"LatentPreparationStage": 27.62,
"DenoisingStage": 134069.79,
"DecodingStage": 13559.79,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 3181.0,
"1": 2561.67,
"2": 2578.49,
"3": 2582.1,
"4": 2572.24,
"5": 2577.72,
"6": 2581.35,
"7": 2578.79,
"8": 2584.98,
"9": 2588.49,
"10": 2594.37,
"11": 2591.19,
"12": 2591.32,
"13": 2595.35,
"14": 2594.35,
"15": 2595.62,
"16": 2596.35,
"17": 2596.11,
"18": 2597.24,
"19": 2603.13,
"20": 2599.9,
"21": 2601.48,
"22": 2603.58,
"23": 2601.13,
"24": 2600.47,
"25": 2604.13,
"26": 2606.04,
"27": 2605.3,
"28": 2602.02,
"29": 2601.83,
"30": 2603.57,
"31": 2606.63,
"32": 2606.1,
"33": 2602.24,
"34": 2603.29,
"35": 2602.34,
"36": 2602.16,
"37": 2608.14,
"38": 2603.48,
"39": 2601.7,
"40": 2603.96,
"41": 2604.58,
"42": 2606.67,
"43": 2603.52,
"44": 2599.88,
"45": 2598.66,
"46": 2600.74,
"47": 2602.31,
"48": 2608.4,
"49": 2606.02
},
"expected_e2e_ms": 150004.2,
"expected_avg_denoise_ms": 2608.84,
"expected_median_denoise_ms": 2601.59
},
"qwen_image_edit_2509_ti2i": {
"stages_ms": {
"InputValidationStage": 213.24,
"ImageEncodingStage": 1089.12,
"ImageVAEEncodingStage": 304.56,
"TimestepPreparationStage": 2.94,
"LatentPreparationStage": 0.2,
"ConditioningStage": 0.01,
"DenoisingStage": 50724.5,
"DecodingStage": 601.02
},
"denoise_step_ms": {
"0": 1057.09,
"1": 1267.06,
"2": 1268.33,
"3": 1268.94,
"4": 1270.36,
"5": 1270.44,
"6": 1268.61,
"7": 1270.21,
"8": 1274.98,
"9": 1271.57,
"10": 1273.15,
"11": 1271.56,
"12": 1272.69,
"13": 1271.62,
"14": 1274.04,
"15": 1276.81,
"16": 1272.2,
"17": 1269.33,
"18": 1275.96,
"19": 1274.43,
"20": 1272.57,
"21": 1275.28,
"22": 1273.63,
"23": 1275.06,
"24": 1277.39,
"25": 1277.27,
"26": 1274.74,
"27": 1273.38,
"28": 1276.77,
"29": 1275.59,
"30": 1275.51,
"31": 1274.9,
"32": 1274.8,
"33": 1279.03,
"34": 1272.9,
"35": 1274.67,
"36": 1272.61,
"37": 1272.82,
"38": 1276.41,
"39": 1273.55
},
"expected_e2e_ms": 52938.04,
"expected_avg_denoise_ms": 1267.96,
"expected_median_denoise_ms": 1273.46
},
"fastwan2_2_ti2v_5b": {
"stages_ms": {
"InputValidationStage": 88.86,
"TextEncodingStage": 2327.87,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 58.66,
"LatentPreparationStage": 28.55,
"DmdDenoisingStage": 4438.3,
"DecodingStage": 14177.77,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 2022.21,
"1": 1263.17,
"2": 1149.59
},
"expected_e2e_ms": 21133.36,
"expected_avg_denoise_ms": 1478.32,
"expected_median_denoise_ms": 1263.17
},
"fast_hunyuan_video": {
"stages_ms": {
"InputValidationStage": 0.34,
"TextEncodingStage": 550.63,
"ConditioningStage": 0.02,
"TimestepPreparationStage": 44.28,
"LatentPreparationStage": 0.29,
"DenoisingStage": 9154.39,
"DecodingStage": 5995.09,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 2518.62,
"1": 578.59,
"2": 1485.76,
"3": 1490.86,
"4": 1489.93,
"5": 1487.02
},
"expected_e2e_ms": 16672.15,
"expected_avg_denoise_ms": 1608.46,
"expected_median_denoise_ms": 1488.48
},
"wan2_2_i2v_a14b_2gpu": {
"stages_ms": {
"InputValidationStage": 18.45,
"TextEncodingStage": 3337.77,
"ConditioningStage": 0.03,
"TimestepPreparationStage": 2.9,
"LatentPreparationStage": 1.25,
"ImageVAEEncodingStage": 1655.89,
"DenoisingStage": 106972.82,
"DecodingStage": 1355.52,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 15659.6,
"1": 1582.6,
"2": 1597.84,
"3": 1601.34,
"4": 1600.86,
"5": 1598.32,
"6": 1600.93,
"7": 1599.88,
"8": 1600.0,
"9": 1600.55,
"10": 1599.27,
"11": 1600.59,
"12": 1600.17,
"13": 1599.72,
"14": 1599.76,
"15": 24098.85,
"16": 1601.29,
"17": 1598.89,
"18": 1600.12,
"19": 1600.52,
"20": 1599.59,
"21": 1600.37,
"22": 1600.35,
"23": 1599.7,
"24": 1599.92,
"25": 1599.75,
"26": 1600.2,
"27": 1600.06,
"28": 1600.41,
"29": 1599.35,
"30": 1600.69,
"31": 1600.15,
"32": 1599.33,
"33": 1599.86,
"34": 1600.52,
"35": 1599.84,
"36": 1600.38,
"37": 1599.23,
"38": 1600.27,
"39": 1599.78
},
"expected_e2e_ms": 123182.9887,
"expected_avg_denoise_ms": 2831.00,
"expected_median_denoise_ms": 1600.09
},
"wan2_1_i2v_14b_480P_2gpu": {
"stages_ms": {
"InputValidationStage": 38.23,
"TextEncodingStage": 3550.36,
"ImageEncodingStage": 3462.55,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 2.6,
"LatentPreparationStage": 9.73,
"ImageVAEEncodingStage": 2290.98,
"DenoisingStage": 415021.17,
"DecodingStage": 3016.1,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 10200.25,
"1": 8222.39,
"2": 8279.38,
"3": 8301.48,
"4": 8338.87,
"5": 8352.39,
"6": 8354.64,
"7": 8353.64,
"8": 8315.58,
"9": 8308.48,
"10": 8299.65,
"11": 8292.7,
"12": 8292.73,
"13": 8285.21,
"14": 8276.06,
"15": 8270.41,
"16": 8273.04,
"17": 8266.04,
"18": 8267.7,
"19": 8264.06,
"20": 8259.32,
"21": 8257.26,
"22": 8253.02,
"23": 8251.77,
"24": 8260.97,
"25": 8251.39,
"26": 8237.43,
"27": 8241.33,
"28": 8235.96,
"29": 8240.6,
"30": 8232.48,
"31": 8237.85,
"32": 8244.3,
"33": 8236.79,
"34": 8239.83,
"35": 8239.89,
"36": 8239.12,
"37": 8246.74,
"38": 8235.67,
"39": 8242.77,
"40": 8241.17,
"41": 8240.24,
"42": 8237.01,
"43": 8231.26,
"44": 8232.85,
"45": 8226.56,
"46": 8236.98,
"47": 8226.73,
"48": 8220.49,
"49": 8217.04
},
"expected_e2e_ms": 426697.37,
"expected_avg_denoise_ms": 8300.19,
"expected_median_denoise_ms": 8267.01
},
"wan2_1_i2v_14b_720P_2gpu": {
"stages_ms": {
"InputValidationStage": 53.67,
"TextEncodingStage": 2838,
"ImageEncodingStage": 3123.99,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 3.39,
"LatentPreparationStage": 8.41,
"ImageVAEEncodingStage": 2261.05,
"DenoisingStage": 417418.12,
"DecodingStage": 2968.35
},
"denoise_step_ms": {
"0": 11848.08,
"1": 8220.3,
"2": 8274.3,
"3": 8298.9,
"4": 8303.34,
"5": 8322.44,
"6": 8314.37,
"7": 8318.54,
"8": 8304.94,
"9": 8303.04,
"10": 8305.22,
"11": 8296.22,
"12": 8289.2,
"13": 8294.19,
"14": 8294.87,
"15": 8285.96,
"16": 8284.98,
"17": 8281.61,
"18": 8277.35,
"19": 8287.46,
"20": 8280.3,
"21": 8279.18,
"22": 8279.37,
"23": 8280.16,
"24": 8282.67,
"25": 8272.14,
"26": 8279.37,
"27": 8271.66,
"28": 8274.6,
"29": 8272.88,
"30": 8273.76,
"31": 8266.17,
"32": 8267.77,
"33": 8266.88,
"34": 8263.14,
"35": 8265.97,
"36": 8267.76,
"37": 8268.03,
"38": 8262.24,
"39": 8261.4,
"40": 8263.65,
"41": 8272.46,
"42": 8254.9,
"43": 8261.03,
"44": 8252.92,
"45": 8262.49,
"46": 8253.67,
"47": 8254.92,
"48": 8257.08,
"49": 8236.56
},
"expected_e2e_ms": 427536.9,
"expected_avg_denoise_ms": 8348.21,
"expected_median_denoise_ms": 8274.45
},
"wan2_2_t2v_a14b_2gpu": {
"stages_ms": {
"InputValidationStage": 0.07,
"TextEncodingStage": 2575.3,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 1.99,
"LatentPreparationStage": 1.26,
"DenoisingStage": 156678.8406,
"DecodingStage": 2702.7,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 17908.3,
"1": 2379.69,
"2": 2393.59,
"3": 2400.91,
"4": 2398.76,
"5": 2403.1,
"6": 2403.26,
"7": 2399.48,
"8": 2401.33,
"9": 2398.4,
"10": 2401.14,
"11": 2409.1,
"12": 2401.16,
"13": 2408.74,
"14": 2404.97,
"15": 2400.51,
"16": 2402.84,
"17": 2401.87,
"18": 2399.67,
"19": 2400.71,
"20": 2399.23,
"21": 2400.13,
"22": 2400.64,
"23": 2399.15,
"24": 2399.58,
"25": 2400.26,
"26": 35247.02,
"27": 2390.25,
"28": 2398.42,
"29": 2399.8,
"30": 2400.08,
"31": 2400.58,
"32": 2403.68,
"33": 2399.37,
"34": 2401.53,
"35": 2399.69,
"36": 2399.9,
"37": 2400.75,
"38": 2398.97,
"39": 2399.12
},
"expected_e2e_ms": 149864.99,
"expected_avg_denoise_ms": 3608.89,
"expected_median_denoise_ms": 2400.38
},
"wan2_1_t2v_14b_2gpu": {
"stages_ms": {
"InputValidationStage": 0.05,
"TextEncodingStage": 2310.34,
"ConditioningStage": 0.02,
"TimestepPreparationStage": 2.42,
"LatentPreparationStage": 27.7,
"DenoisingStage": 803631.52,
"DecodingStage": 8898.74,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 17347.88,
"1": 15956.93,
"2": 16027.54,
"3": 16054.15,
"4": 16081.46,
"5": 16062.7,
"6": 16058.56,
"7": 16057.58,
"8": 16061.04,
"9": 16120.97,
"10": 16036.84,
"11": 16019.6,
"12": 16042.29,
"13": 16039.87,
"14": 16063.0,
"15": 16036.16,
"16": 16079.82,
"17": 16019.7,
"18": 16061.5,
"19": 16039.95,
"20": 16009.42,
"21": 16051.01,
"22": 16039.31,
"23": 16048.22,
"24": 16071.41,
"25": 16078.75,
"26": 16061.78,
"27": 16018.39,
"28": 16041.44,
"29": 16039.64,
"30": 16041.89,
"31": 16039.6,
"32": 16038.97,
"33": 15999.48,
"34": 16019.93,
"35": 16040.27,
"36": 16020.3,
"37": 16039.38,
"38": 15999.4,
"39": 16022.15,
"40": 16042.32,
"41": 16016.62,
"42": 15998.92,
"43": 16041.48,
"44": 15999.63,
"45": 16003.21,
"46": 15995.91,
"47": 16023.52,
"48": 16016.64,
"49": 16019.6
},
"expected_e2e_ms": 814884.71,
"expected_avg_denoise_ms": 16062.92,
"expected_median_denoise_ms": 16039.62
},
"wan2_2_t2v_a14b_lora_2gpu": {
"stages_ms": {
"InputValidationStage": 0.09,
"TextEncodingStage": 2552.97,
"ConditioningStage": 0.03,
"TimestepPreparationStage": 1.99,
"LatentPreparationStage": 1.29,
"DenoisingStage": 154340.69,
"DecodingStage": 2730.86,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 26510.7,
"1": 2381.25,
"2": 2396.9,
"3": 2400.96,
"4": 2402.47,
"5": 2399.6,
"6": 2400.5,
"7": 2401.13,
"8": 2399.32,
"9": 2400.0,
"10": 2401.35,
"11": 2400.04,
"12": 2408.27,
"13": 2407.08,
"14": 2405.92,
"15": 2403.99,
"16": 2402.12,
"17": 2402.52,
"18": 2398.08,
"19": 2399.9,
"20": 2400.14,
"21": 2398.64,
"22": 2401.32,
"23": 2400.75,
"24": 2399.27,
"25": 2400.21,
"26": 36387.55,
"27": 2399.77,
"28": 2398.09,
"29": 2404.64,
"30": 2400.68,
"31": 2404.3,
"32": 2392.44,
"33": 2390.56,
"34": 2396.05,
"35": 2394.86,
"36": 2396.07,
"37": 2398.49,
"38": 2394.77,
"39": 2394.19
},
"expected_e2e_ms": 159643.06,
"expected_avg_denoise_ms": 3851.87,
"expected_median_denoise_ms": 2400.09
},
"wan2_1_t2v_1_3b_lora_1gpu": {
"stages_ms": {
"InputValidationStage": 0.06,
"TextEncodingStage": 2467.44,
"ConditioningStage": 0.02,
"TimestepPreparationStage": 2.96,
"LatentPreparationStage": 1.87,
"DenoisingStage": 14859.47,
"DecodingStage": 1199.31,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 1964.07,
"1": 265.02,
"2": 257.83,
"3": 260.27,
"4": 261.43,
"5": 258.58,
"6": 256.64,
"7": 256.91,
"8": 258.41,
"9": 257.84,
"10": 257.08,
"11": 257.0,
"12": 258.44,
"13": 257.1,
"14": 256.95,
"15": 257.2,
"16": 256.84,
"17": 257.64,
"18": 257.22,
"19": 257.42,
"20": 256.91,
"21": 256.99,
"22": 257.17,
"23": 257.63,
"24": 258.89,
"25": 257.46,
"26": 257.3,
"27": 257.42,
"28": 257.19,
"29": 257.65,
"30": 257.39,
"31": 256.93,
"32": 258.23,
"33": 257.62,
"34": 281.86,
"35": 295.86,
"36": 296.73,
"37": 287.21,
"38": 300.87,
"39": 303.47,
"40": 294.09,
"41": 270.52,
"42": 256.53,
"43": 256.58,
"44": 256.29,
"45": 255.81,
"46": 256.34,
"47": 256.08,
"48": 255.92,
"49": 255.87
},
"expected_e2e_ms": 18547.46,
"expected_avg_denoise_ms": 297.09,
"expected_median_denoise_ms": 257.42
},
"wan2_1_i2v_14b_lora_2gpu": {
"stages_ms": {
"InputValidationStage": 23.97,
"TextEncodingStage": 2485.39,
"ImageEncodingStage": 2372.07,
"ConditioningStage": 0.01,
"TimestepPreparationStage": 2.6,
"LatentPreparationStage": 0.18,
"ImageVAEEncodingStage": 2500.13,
"DenoisingStage": 193514.04,
"DecodingStage": 3341.78,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 7828.3,
"1": 3765.8,
"2": 3774.63,
"3": 3772.93,
"4": 3781.13,
"5": 3778.22,
"6": 3776.41,
"7": 3772.02,
"8": 3776.15,
"9": 3768.82,
"10": 3775.31,
"11": 3771.32,
"12": 3774.33,
"13": 3772.5,
"14": 3778.41,
"15": 3775.31,
"16": 3771.38,
"17": 3774.87,
"18": 3780.01,
"19": 3772.85,
"20": 3773.65,
"21": 3774.47,
"22": 3774.39,
"23": 3773.08,
"24": 3776.71,
"25": 3780.01,
"26": 3774.83,
"27": 3773.27,
"28": 3773.76,
"29": 3772.75,
"30": 3773.01,
"31": 3773.34,
"32": 3773.13,
"33": 3774.12,
"34": 3772.19,
"35": 3774.7,
"36": 3773.98,
"37": 3772.47,
"38": 3771.72,
"39": 3774.07,
"40": 3773.71,
"41": 3773.6,
"42": 3772.12,
"43": 3773.75,
"44": 3782.43,
"45": 3779.66,
"46": 3779.86,
"47": 3774.58,
"48": 3770.54,
"49": 3776.76
},
"expected_e2e_ms": 204257.12,
"expected_avg_denoise_ms": 3855.55,
"expected_median_denoise_ms": 3774.03
},
"flux_2_image_t2i_2_gpus": {
"stages_ms": {
"InputValidationStage": 0.05,
"TextEncodingStage": 518.88,
"ImageVAEEncodingStage": 0.0,
"ConditioningStage": 0.01,
"LatentPreparationStage": 0.45,
"TimestepPreparationStage": 3.41,
"DenoisingStage": 26377.63,
"DecodingStage": 321.94
},
"denoise_step_ms": {
"0": 66.94,
"1": 511.02,
"2": 537.24,
"3": 516.92,
"4": 537.18,
"5": 520.79,
"6": 531.95,
"7": 522.69,
"8": 532.77,
"9": 522.36,
"10": 530.91,
"11": 526.67,
"12": 530.23,
"13": 526.16,
"14": 528.55,
"15": 524.96,
"16": 526.36,
"17": 527.71,
"18": 528.84,
"19": 528.73,
"20": 527.95,
"21": 525.1,
"22": 528.21,
"23": 528.06,
"24": 527.11,
"25": 530.93,
"26": 526.27,
"27": 528.66,
"28": 526.33,
"29": 530.99,
"30": 528.51,
"31": 528.53,
"32": 528.73,
"33": 527.02,
"34": 527.49,
"35": 528.25,
"36": 529.96,
"37": 528.43,
"38": 528.5,
"39": 525.72,
"40": 529.65,
"41": 525.62,
"42": 526.24,
"43": 528.77,
"44": 526.68,
"45": 528.06,
"46": 524.92,
"47": 529.5,
"48": 524.22,
"49": 528.21
},
"expected_e2e_ms": 27624.8,
"expected_avg_denoise_ms": 518.23,
"expected_median_denoise_ms": 528.06
}
}
}