From e13b727e928354e53b1e5ba3f613d8bb447a5594 Mon Sep 17 00:00:00 2001 From: Prozac614 Date: Sat, 7 Feb 2026 00:28:44 +0800 Subject: [PATCH] [diffusion] CI: update perf baseline (#17512) --- .../test/server/perf_baselines.json | 717 ++++++++++-------- 1 file changed, 392 insertions(+), 325 deletions(-) diff --git a/python/sglang/multimodal_gen/test/server/perf_baselines.json b/python/sglang/multimodal_gen/test/server/perf_baselines.json index 4e0f64f83..f7317283d 100644 --- a/python/sglang/multimodal_gen/test/server/perf_baselines.json +++ b/python/sglang/multimodal_gen/test/server/perf_baselines.json @@ -36,70 +36,70 @@ "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 + "ConditioningStage": 0.02, + "DecodingStage": 51.86, + "TextEncodingStage": 611.83, + "InputValidationStage": 0.05, + "DenoisingStage": 14289.46, + "LatentPreparationStage": 0.2, + "TimestepPreparationStage": 3.34 }, "denoise_step_ms": { - "0": 400.06, - "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 - } + "0": 240.5, + "1": 279.1, + "2": 283.29, + "3": 296.63, + "4": 287.72, + "5": 283.39, + "6": 283.98, + "7": 291.82, + "8": 283.1, + "9": 284.43, + "10": 288.95, + "11": 285.6, + "12": 285.99, + "13": 285.47, + "14": 289.66, + "15": 285.74, + "16": 284.15, + "17": 290.27, + "18": 288.04, + "19": 284.57, + "20": 286.69, + "21": 288.95, + "22": 287.09, + "23": 285.6, + "24": 289.31, + "25": 285.48, + "26": 285.53, + "27": 288.13, + "28": 287.65, + "29": 285.97, + "30": 288.9, + "31": 287.97, + "32": 286.48, + "33": 285.38, + "34": 286.62, + "35": 288.22, + "36": 285.6, + "37": 286.61, + "38": 287.06, + "39": 286.2, + "40": 284.6, + "41": 285.69, + "42": 288.46, + "43": 285.53, + "44": 285.34, + "45": 285.74, + "46": 287.25, + "47": 285.0, + "48": 286.82, + "49": 287.19 + }, + "expected_e2e_ms": 14959.11, + "expected_avg_denoise_ms": 285.67, + "expected_median_denoise_ms": 286.1 }, "qwen_image_t2i_2_gpus": { "stages_ms": { @@ -169,157 +169,157 @@ }, "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 + "DecodingStage": 32.72, + "TextEncodingStage": 51.96, + "InputValidationStage": 0.03, + "DenoisingStage": 7545.16, + "LatentPreparationStage": 0.2, + "TimestepPreparationStage": 2.43 }, "denoise_step_ms": { - "0": 58.88, - "1": 165.27, - "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 + "0": 50.06, + "1": 58.88, + "2": 151.24, + "3": 150.97, + "4": 151.23, + "5": 151.63, + "6": 159.11, + "7": 158.31, + "8": 153.42, + "9": 151.42, + "10": 151.91, + "11": 151.05, + "12": 151.52, + "13": 157.2, + "14": 152.76, + "15": 153.85, + "16": 153.02, + "17": 151.09, + "18": 151.49, + "19": 155.13, + "20": 155.2, + "21": 152.82, + "22": 152.2, + "23": 150.99, + "24": 152.74, + "25": 153.45, + "26": 153.63, + "27": 154.92, + "28": 152.72, + "29": 151.84, + "30": 151.84, + "31": 152.44, + "32": 153.03, + "33": 154.07, + "34": 152.36, + "35": 153.48, + "36": 152.05, + "37": 152.45, + "38": 152.42, + "39": 154.91, + "40": 152.68, + "41": 153.43, + "42": 151.62, + "43": 153.52, + "44": 153.13, + "45": 152.85, + "46": 152.33, + "47": 151.61, + "48": 152.4, + "49": 152.33 }, - "expected_e2e_ms": 9275.51, - "expected_avg_denoise_ms": 165.83, - "expected_median_denoise_ms": 169.33 + "expected_e2e_ms": 7798.99, + "expected_avg_denoise_ms": 150.77, + "expected_median_denoise_ms": 152.45 }, "flux_2_image_t2i": { "stages_ms": { - "InputValidationStage": 0.05, - "TextEncodingStage": 530.93, + "LatentPreparationStage": 0.52, + "TimestepPreparationStage": 2.91, + "TextEncodingStage": 518.54, "ImageVAEEncodingStage": 0.0, "ConditioningStage": 0.02, - "LatentPreparationStage": 12.71, - "TimestepPreparationStage": 2.91, - "DenoisingStage": 26403.1, - "DecodingStage": 489.8 + "InputValidationStage": 0.05, + "DenoisingStage": 24901.97, + "DecodingStage": 8.98 }, "denoise_step_ms": { - "0": 68.7, - "1": 511.3, - "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 + "0": 69.14, + "1": 132.57, + "2": 508.67, + "3": 493.52, + "4": 504.31, + "5": 492.99, + "6": 501.91, + "7": 495.18, + "8": 500.87, + "9": 497.36, + "10": 498.74, + "11": 497.46, + "12": 499.08, + "13": 494.65, + "14": 500.35, + "15": 496.89, + "16": 500.23, + "17": 497.01, + "18": 501.68, + "19": 493.8, + "20": 501.1, + "21": 494.81, + "22": 501.04, + "23": 499.27, + "24": 500.04, + "25": 497.14, + "26": 499.05, + "27": 494.91, + "28": 496.89, + "29": 498.53, + "30": 497.94, + "31": 497.09, + "32": 497.7, + "33": 497.58, + "34": 496.43, + "35": 497.7, + "36": 497.37, + "37": 497.17, + "38": 499.27, + "39": 495.52, + "40": 501.67, + "41": 495.11, + "42": 500.69, + "43": 501.61, + "44": 501.91, + "45": 495.58, + "46": 499.37, + "47": 496.8, + "48": 497.49, + "49": 495.69 }, - "expected_e2e_ms": 27648.69, - "expected_avg_denoise_ms": 520.09, - "expected_median_denoise_ms": 528.0 + "expected_e2e_ms": 25832.82, + "expected_avg_denoise_ms": 489.43, + "expected_median_denoise_ms": 497.53 }, "flux_2_klein_image_t2i": { "stages_ms": { + "ConditioningStage": 0.01, + "DecodingStage": 9.27, + "TextEncodingStage": 92.17, "InputValidationStage": 0.05, - "TextEncodingStage": 530.93, "ImageVAEEncodingStage": 0.0, - "ConditioningStage": 0.02, - "LatentPreparationStage": 12.71, - "TimestepPreparationStage": 2.91, - "DenoisingStage": 2112.24, - "DecodingStage": 489.8 + "DenoisingStage": 252.01, + "LatentPreparationStage": 0.42, + "TimestepPreparationStage": 1.5 }, "denoise_step_ms": { - "0": 511.3, - "1": 541.19, - "2": 518.93, - "3": 541.2 + "0": 19.91, + "1": 19.32, + "2": 51.99, + "3": 61.78 }, - "expected_e2e_ms": 3148.75, - "expected_avg_denoise_ms": 528.06, - "expected_median_denoise_ms": 526.56 + "expected_e2e_ms": 430.73, + "expected_avg_denoise_ms": 38.25, + "expected_median_denoise_ms": 35.95 }, "flux_2_image_t2i_layerwise_offload": { "stages_ms": { @@ -523,28 +523,28 @@ }, "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 + "DecodingStage": 33.93, + "TextEncodingStage": 104.21, + "InputValidationStage": 0.03, + "DenoisingStage": 808.69, + "LatentPreparationStage": 0.14, + "TimestepPreparationStage": 1.33 }, "denoise_step_ms": { - "0": 101.56, + "0": 23.4, "1": 28.26, - "2": 101.74, - "3": 101.68, - "4": 102.19, - "5": 102.05, - "6": 102.03, - "7": 102.28, - "8": 105.54 + "2": 100.72, + "3": 100.48, + "4": 99.62, + "5": 99.83, + "6": 99.67, + "7": 99.3, + "8": 100.83 }, - "expected_e2e_ms": 1383.47, - "expected_avg_denoise_ms": 94.15, - "expected_median_denoise_ms": 102.03 + "expected_e2e_ms": 952.09, + "expected_avg_denoise_ms": 89.36, + "expected_median_denoise_ms": 99.67 }, "zimage_image_t2i_multi_lora": { "stages_ms": { @@ -623,70 +623,70 @@ }, "qwen_image_edit_ti2i": { "stages_ms": { - "InputValidationStage": 38.62, + "LatentPreparationStage": 0.16, + "TimestepPreparationStage": 2.62, "ImageEncodingStage": 1174.26, - "ImageVAEEncodingStage": 233.71, - "TimestepPreparationStage": 3.0, - "LatentPreparationStage": 0.17, + "ImageVAEEncodingStage": 132.67, "ConditioningStage": 0.01, - "DenoisingStage": 42542.67, - "DecodingStage": 508.39 + "InputValidationStage": 38.1, + "DenoisingStage": 38135.64, + "DecodingStage": 139.72 }, "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 + "0": 618.31, + "1": 769.07, + "2": 766.91, + "3": 762.77, + "4": 764.26, + "5": 765.27, + "6": 767.35, + "7": 764.18, + "8": 766.16, + "9": 766.89, + "10": 766.1, + "11": 764.96, + "12": 763.52, + "13": 765.22, + "14": 765.44, + "15": 763.9, + "16": 763.19, + "17": 764.83, + "18": 765.36, + "19": 765.19, + "20": 765.96, + "21": 765.74, + "22": 765.87, + "23": 764.85, + "24": 765.44, + "25": 765.95, + "26": 766.21, + "27": 767.91, + "28": 765.45, + "29": 764.81, + "30": 766.26, + "31": 765.37, + "32": 766.71, + "33": 765.67, + "34": 766.64, + "35": 765.98, + "36": 766.04, + "37": 764.19, + "38": 765.15, + "39": 766.33, + "40": 767.68, + "41": 765.36, + "42": 766.61, + "43": 766.06, + "44": 765.26, + "45": 765.29, + "46": 764.64, + "47": 766.07, + "48": 762.89, + "49": 763.01 }, - "expected_e2e_ms": 44503.12, - "expected_avg_denoise_ms": 850.73, - "expected_median_denoise_ms": 854.0 + "expected_e2e_ms": 39706.9, + "expected_avg_denoise_ms": 762.57, + "expected_median_denoise_ms": 765.44 }, "qwen_image_t2i_cache_dit_enabled": { "stages_ms": { @@ -821,6 +821,73 @@ "expected_avg_denoise_ms": 260.76, "expected_median_denoise_ms": 247.84 }, + "wan2_1_t2v_1.3b_teacache_enabled": { + "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": 240.71, + "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_1_t2v_1.3b": { "stages_ms": { "InputValidationStage": 0.07, @@ -1169,66 +1236,66 @@ }, "qwen_image_layered_i2i": { "stages_ms": { - "QwenImageLayeredBeforeDenoisingStage": 3240.24, - "TimestepPreparationStage": 3.14, - "DenoisingStage": 41551.18, - "DecodingStage": 312.93 + "QwenImageLayeredBeforeDenoisingStage": 2897.28, + "DecodingStage": 312.93, + "DenoisingStage": 39417.66, + "TimestepPreparationStage": 2.29 }, "denoise_step_ms": { - "0": 809.93, - "1": 836.69, - "2": 834.98, - "3": 826.84, - "4": 827.15, - "5": 827.28, - "6": 830.97, - "7": 827.7, - "8": 829.4, - "9": 832.14, - "10": 825.99, - "11": 831.65, - "12": 829.31, - "13": 829.46, - "14": 828.33, - "15": 831.14, - "16": 830.44, - "17": 831.6, - "18": 829.18, - "19": 831.64, - "20": 828.21, - "21": 831.02, - "22": 831.39, - "23": 830.16, - "24": 832.21, - "25": 831.04, - "26": 830.48, - "27": 831.88, - "28": 833.5, - "29": 837.31, - "30": 828.16, - "31": 832.24, - "32": 833.56, - "33": 829.08, - "34": 833.11, - "35": 831.07, - "36": 832.71, - "37": 833.12, - "38": 830.65, - "39": 831.59, - "40": 833.24, - "41": 831.92, - "42": 832.77, - "43": 830.88, - "44": 833.75, - "45": 831.29, - "46": 834.48, - "47": 832.6, - "48": 835.24, - "49": 832.49 + "0": 657.28, + "1": 799.2, + "2": 790.35, + "3": 785.79, + "4": 792.9, + "5": 795.78, + "6": 791.28, + "7": 790.87, + "8": 786.47, + "9": 791.03, + "10": 788.77, + "11": 790.57, + "12": 788.7, + "13": 786.01, + "14": 791.43, + "15": 789.88, + "16": 791.18, + "17": 792.78, + "18": 792.06, + "19": 790.47, + "20": 792.48, + "21": 789.13, + "22": 792.12, + "23": 789.36, + "24": 790.2, + "25": 790.87, + "26": 792.37, + "27": 794.92, + "28": 792.9, + "29": 791.43, + "30": 793.01, + "31": 793.71, + "32": 794.15, + "33": 787.93, + "34": 792.12, + "35": 794.01, + "36": 789.05, + "37": 790.51, + "38": 793.29, + "39": 791.94, + "40": 788.94, + "41": 788.85, + "42": 789.76, + "43": 788.89, + "44": 791.62, + "45": 788.04, + "46": 790.03, + "47": 786.82, + "48": 789.75, + "49": 789.0 }, - "expected_e2e_ms": 45109.63, - "expected_avg_denoise_ms": 830.86, - "expected_median_denoise_ms": 831.34 + "expected_e2e_ms": 42660.88, + "expected_avg_denoise_ms": 788.2, + "expected_median_denoise_ms": 790.72 }, "fastwan2_2_ti2v_5b": { "stages_ms": {