From f3ccd1c0e42926e795a4304d50935b8cbd210169 Mon Sep 17 00:00:00 2001 From: Mick Date: Mon, 7 Sep 2026 23:13:14 +0800 Subject: [PATCH] [diffusion] CI: baseline the e2e of ten unguarded perf cases (#38335) Co-authored-by: Mick Qian Co-authored-by: Claude Fable 5.1 --- .../test/server/perf_baselines/b200.json | 224 ++++++++++++++++-- .../test/server/perf_baselines/h100.json | 122 ++++++++-- 2 files changed, 302 insertions(+), 44 deletions(-) diff --git a/python/sglang/multimodal_gen/test/server/perf_baselines/b200.json b/python/sglang/multimodal_gen/test/server/perf_baselines/b200.json index 6af1b6dfe..aa81eee87 100644 --- a/python/sglang/multimodal_gen/test/server/perf_baselines/b200.json +++ b/python/sglang/multimodal_gen/test/server/perf_baselines/b200.json @@ -40,43 +40,217 @@ }, "scenarios": { "flux1_modelopt_nvfp4_t2i": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "TextEncodingStage": 16.55, + "LatentPreparationStage": 0.14, + "TimestepPreparationStage": 18.55, + "DenoisingStage": 623.33, + "DecodingStage": 18.67 + }, + "denoise_step_ms": { + "0": 61.5, + "1": 51.43, + "2": 51.19, + "3": 50.75, + "4": 51.09, + "5": 51.0, + "6": 51.14, + "7": 50.65, + "8": 50.41, + "9": 50.45, + "10": 50.42, + "11": 50.9 + }, + "expected_e2e_ms": 836.71, + "expected_avg_denoise_ms": 51.74, + "expected_median_denoise_ms": 50.95, "estimated_full_test_time_s": 71.2 }, "flux2_modelopt_nvfp4_t2i": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "TextEncodingStage": 330.43, + "ImageVAEEncodingStage": 0.01, + "LatentPreparationStage": 0.32, + "TimestepPreparationStage": 12.36, + "DenoisingStage": 1184.2, + "DecodingStage": 17.18 + }, + "denoise_step_ms": { + "0": 96.2, + "1": 30.93, + "2": 66.88, + "3": 98.59, + "4": 97.53, + "5": 100.31, + "6": 100.42, + "7": 98.55, + "8": 97.93, + "9": 98.11, + "10": 99.03, + "11": 99.53 + }, + "expected_e2e_ms": 1990.84, + "expected_avg_denoise_ms": 90.33, + "expected_median_denoise_ms": 98.33, "estimated_full_test_time_s": 592.3 }, "qwen_image_2512_modelopt_nvfp4_t2i": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "TextEncodingStage": 221.4, + "LatentPreparationStage": 0.13, + "TimestepPreparationStage": 10.48, + "DenoisingStage": 9380.87, + "DecodingStage": 27.2 + }, + "denoise_step_ms": { + "0": 203.71, + "1": 190.75, + "2": 186.34, + "3": 186.88, + "4": 187.28, + "5": 185.4, + "6": 185.44, + "7": 186.61, + "8": 185.09, + "9": 190.7, + "10": 186.22, + "11": 185.68, + "12": 190.85, + "13": 186.54, + "14": 187.12, + "15": 185.83, + "16": 185.53, + "17": 186.04, + "18": 185.14, + "19": 185.39, + "20": 184.56, + "21": 188.41, + "22": 184.77, + "23": 185.5, + "24": 190.32, + "25": 187.97, + "26": 185.62, + "27": 191.78, + "28": 186.4, + "29": 187.19, + "30": 185.61, + "31": 188.86, + "32": 185.2, + "33": 186.15, + "34": 189.01, + "35": 187.05, + "36": 186.95, + "37": 189.05, + "38": 187.4, + "39": 188.8, + "40": 193.13, + "41": 187.5, + "42": 184.56, + "43": 190.36, + "44": 185.99, + "45": 185.96, + "46": 188.3, + "47": 191.32, + "48": 186.01, + "49": 185.95 + }, + "expected_e2e_ms": 9650.06, + "expected_avg_denoise_ms": 187.56, + "expected_median_denoise_ms": 186.57, "estimated_full_test_time_s": 120.0 }, "wan22_modelopt_nvfp4_t2v": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "TextEncodingStage": 316.23, + "LatentPreparationStage": 0.11, + "TimestepPreparationStage": 1.44, + "DenoisingStage": 5413.33, + "DecodingStage": 82.95 + }, + "denoise_step_ms": { + "0": 306.5, + "1": 410.26, + "2": 409.53, + "3": 407.43, + "4": 406.74, + "5": 406.78, + "6": 406.75, + "7": 406.78, + "8": 899.72, + "9": 406.75, + "10": 406.77, + "11": 404.41 + }, + "expected_e2e_ms": 5964.13, + "expected_avg_denoise_ms": 439.87, + "expected_median_denoise_ms": 406.77, "estimated_full_test_time_s": 181.8 }, "ideogram4_nvfp4_t2i": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "Ideogram4TextEncodingStage": 129.35, + "LatentPreparationStage": 0.1, + "DenoisingStage": 5233.76, + "Ideogram4DecodingStage": 53.08 + }, + "denoise_step_ms": { + "0": 88.71, + "1": 83.99, + "2": 109.67, + "3": 110.03, + "4": 109.89, + "5": 109.98, + "6": 109.73, + "7": 109.89, + "8": 111.44, + "9": 108.37, + "10": 110.23, + "11": 110.11, + "12": 109.81, + "13": 109.9, + "14": 109.81, + "15": 110.09, + "16": 109.8, + "17": 109.88, + "18": 109.82, + "19": 110.23, + "20": 109.81, + "21": 109.99, + "22": 110.21, + "23": 110.04, + "24": 109.87, + "25": 110.39, + "26": 109.8, + "27": 109.92, + "28": 109.83, + "29": 109.88, + "30": 109.83, + "31": 109.89, + "32": 109.82, + "33": 110.16, + "34": 110.24, + "35": 110.02, + "36": 109.67, + "37": 110.06, + "38": 109.8, + "39": 110.35, + "40": 110.19, + "41": 110.43, + "42": 110.15, + "43": 110.02, + "44": 109.8, + "45": 109.94, + "46": 109.81, + "47": 110.02 + }, + "expected_e2e_ms": 5421.76, + "expected_avg_denoise_ms": 108.99, + "expected_median_denoise_ms": 109.9, "estimated_full_test_time_s": 300.0 } } diff --git a/python/sglang/multimodal_gen/test/server/perf_baselines/h100.json b/python/sglang/multimodal_gen/test/server/perf_baselines/h100.json index 528e11b99..d7b2b0cd0 100644 --- a/python/sglang/multimodal_gen/test/server/perf_baselines/h100.json +++ b/python/sglang/multimodal_gen/test/server/perf_baselines/h100.json @@ -2865,17 +2865,50 @@ "estimated_full_test_time_s": 149.4 }, "lingbot_video_moe_t2v": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.06, + "LingBotVideoTextEncodingStage": 89.03, + "LatentPreparationStage": 0.24, + "TimestepPreparationStage": 2.73, + "DenoisingStage": 5228.81, + "DecodingStage": 201.36 + }, + "denoise_step_ms": { + "0": 318.64, + "1": 340.93, + "2": 338.25, + "3": 337.36, + "4": 337.34, + "5": 337.5, + "6": 337.66, + "7": 337.43, + "8": 337.52, + "9": 337.32, + "10": 337.61, + "11": 337.96 + }, + "expected_e2e_ms": 6639.9, + "expected_avg_denoise_ms": 336.29, + "expected_median_denoise_ms": 337.51, + "load_peak_vram_mb": 92.0, + "runtime_peak_vram_mb": 66634.0, + "load_peak_allocated_mb": 80.0, + "runtime_peak_allocated_mb": 58436.0, "estimated_full_test_time_s": 600.0 }, "lingbot_world_realtime_plastic_beach": { - "stages_ms": {}, + "stages_ms": { + "RealtimeInputValidationStage": 0.09, + "RealtimeTextEncodingStage": 0.04, + "AuxiliaryConditionEncodingStage": 1.89, + "RealtimeImageVAEEncodingStage": 0.01, + "DMDTimestepPreparationStage": 2.07, + "RealtimeChunkLatentPreparationStage": 0.09, + "LingBotWorldCausalDMDDenoisingStage": 1684.12, + "CausalVaeDecodingStage": 217.75 + }, "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, + "expected_e2e_ms": 1912.29, "expected_avg_denoise_ms": 0.0, "expected_median_denoise_ms": 0.0, "load_peak_vram_mb": 42814.0, @@ -2933,11 +2966,17 @@ "estimated_full_test_time_s": 363.2 }, "qwen_image_t2i_cache_dit_scm_config_diffusers_1gpu": { - "stages_ms": {}, + "stages_ms": { + "DiffusersExecutionStage": 1075.16 + }, "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, + "expected_e2e_ms": 1076.57, "expected_avg_denoise_ms": 0.0, "expected_median_denoise_ms": 0.0, + "load_peak_vram_mb": 55304.0, + "runtime_peak_vram_mb": 56730.0, + "load_peak_allocated_mb": 55082.0, + "runtime_peak_allocated_mb": 56236.0, "estimated_full_test_time_s": 98.3 }, "cosmos3_nano_t2i": { @@ -3112,11 +3151,32 @@ "estimated_full_test_time_s": 211.8 }, "zimage_image_t2i_2_gpus_non_square": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.04, + "TextEncodingStage": 259.45, + "LatentPreparationStage": 0.1, + "TimestepPreparationStage": 23.65, + "DenoisingStage": 460.71, + "DecodingStage": 109.26 + }, + "denoise_step_ms": { + "0": 57.11, + "1": 29.15, + "2": 52.97, + "3": 49.26, + "4": 49.73, + "5": 63.78, + "6": 46.54, + "7": 52.38, + "8": 57.66 + }, + "expected_e2e_ms": 862.71, + "expected_avg_denoise_ms": 50.95, + "expected_median_denoise_ms": 52.38, + "load_peak_vram_mb": 13120.0, + "runtime_peak_vram_mb": 17604.0, + "load_peak_allocated_mb": 12852.0, + "runtime_peak_allocated_mb": 15130.0, "estimated_full_test_time_s": 40.5 }, "flux1_modelopt_fp8_t2i": { @@ -3144,11 +3204,35 @@ "estimated_full_test_time_s": 89.5 }, "hunyuanvideo_modelopt_fp8_t2v": { - "stages_ms": {}, - "denoise_step_ms": {}, - "expected_e2e_ms": 0.0, - "expected_avg_denoise_ms": 0.0, - "expected_median_denoise_ms": 0.0, + "stages_ms": { + "InputValidationStage": 0.05, + "TextEncodingStage": 33.67, + "TimestepPreparationStage": 44.99, + "LatentPreparationStage": 0.11, + "DenoisingStage": 1452.58, + "DecodingStage": 777.92 + }, + "denoise_step_ms": { + "0": 135.1, + "1": 80.04, + "2": 139.71, + "3": 112.9, + "4": 126.7, + "5": 115.62, + "6": 119.0, + "7": 118.17, + "8": 127.64, + "9": 127.04, + "10": 123.51, + "11": 125.43 + }, + "expected_e2e_ms": 2608.65, + "expected_avg_denoise_ms": 120.9, + "expected_median_denoise_ms": 124.47, + "load_peak_vram_mb": 16742.0, + "runtime_peak_vram_mb": 36348.0, + "load_peak_allocated_mb": 16063.0, + "runtime_peak_allocated_mb": 30436.0, "estimated_full_test_time_s": 64.3 }, "qwen_image_modelopt_fp8_t2i": {