Files
sglang/python/sglang/multimodal_gen/test/server/perf_baselines/5090.json
T

230 lines
7.9 KiB
JSON

{
"metadata": {
"model": "Diffusion Server",
"hardware": "CI RTX 5090 pool",
"description": "Reference numbers from RTX 5090 multimodal_gen CI runs 31935214180 and 33854592181.",
"last_updated": "2026-09-04"
},
"tolerances": {
"long_term": {
"e2e": 0.15,
"denoise_stage": 0.1,
"non_denoise_stage": 0.5,
"denoise_step": 0.25,
"denoise_agg": 0.15,
"load_peak_vram": 0.01,
"runtime_peak_vram": 0.02
},
"pr_test": {
"e2e": 0.25,
"denoise_stage": 0.25,
"non_denoise_stage": 0.8,
"denoise_step": 0.3,
"denoise_agg": 0.2,
"load_peak_vram": 0.01,
"runtime_peak_vram": 0.02
}
},
"improvement_reporting": {
"threshold": 0.2
},
"sampling": {
"step_fractions": [
0.0,
0.2,
0.4,
0.6,
0.8,
1.0
]
},
"scenarios": {
"flux_2_klein_base_image_t2i": {
"stages_ms": {
"DecodingStage": 18.19,
"DenoisingStage": 17612.9,
"ImageVAEEncodingStage": 0.01,
"InputValidationStage": 0.08,
"LatentPreparationStage": 1.67,
"TextEncodingStage": 104.46,
"TimestepPreparationStage": 142.18
},
"denoise_step_ms": {
"0": 95.07,
"10": 344.0,
"20": 345.85,
"29": 346.2,
"39": 346.9,
"49": 347.31
},
"expected_e2e_ms": 18255.39,
"expected_avg_denoise_ms": 338.78,
"expected_median_denoise_ms": 345.95,
"load_peak_vram_mb": 1126.0,
"runtime_peak_vram_mb": 13180.0,
"load_peak_allocated_mb": 1019.0,
"runtime_peak_allocated_mb": 8912.0,
"estimated_full_test_time_s": 94.0
},
"wan2_1_t2v_1.3b": {
"stages_ms": {
"DecodingStage": 495.01,
"DenoisingStage": 21452.59,
"InputValidationStage": 0.09,
"LatentPreparationStage": 0.23,
"TextEncodingStage": 1025.11,
"TimestepPreparationStage": 3.91
},
"denoise_step_ms": {
"0": 169.02,
"10": 426.11,
"20": 427.32,
"29": 428.09,
"39": 428.04,
"49": 422.31
},
"expected_e2e_ms": 23418.73,
"expected_avg_denoise_ms": 426.59,
"expected_median_denoise_ms": 427.94,
"load_peak_vram_mb": 4934.0,
"runtime_peak_vram_mb": 18246.0,
"load_peak_allocated_mb": 903.0,
"runtime_peak_allocated_mb": 6934.0,
"estimated_full_test_time_s": 160.9
},
"turbo_wan2_1_t2v_1.3b": {
"stages_ms": {
"InputValidationStage": 0.11,
"TextEncodingStage": 634.42,
"TimestepPreparationStage": 65.02,
"LatentPreparationStage": 0.23,
"DmdDenoisingStage": 577.26,
"DecodingStage": 493.29,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 91.5,
"1": 160.17,
"2": 161.71,
"3": 161.1
},
"expected_e2e_ms": 2155.59,
"expected_avg_denoise_ms": 143.62,
"expected_median_denoise_ms": 160.64,
"load_peak_vram_mb": 928.0,
"runtime_peak_vram_mb": 14216.0,
"load_peak_allocated_mb": 903.0,
"runtime_peak_allocated_mb": 6925.0,
"load_peak_host_anon_mb": 6759.33,
"runtime_peak_host_anon_mb": 6759.33,
"estimated_full_test_time_s": 200.3
},
"zimage_image_t2i": {
"stages_ms": {
"DecodingStage": 7.11,
"DenoisingStage": 2178.8,
"InputValidationStage": 0.04,
"LatentPreparationStage": 0.17,
"TextEncodingStage": 252.2,
"TimestepPreparationStage": 36.04
},
"denoise_step_ms": {
"0": 39.34,
"2": 270.3,
"3": 263.6,
"5": 264.28,
"6": 265.1,
"8": 265.56
},
"expected_e2e_ms": 2492.09,
"expected_avg_denoise_ms": 241.53,
"expected_median_denoise_ms": 265.1,
"load_peak_vram_mb": 13120.0,
"runtime_peak_vram_mb": 17776.0,
"load_peak_allocated_mb": 12852.0,
"runtime_peak_allocated_mb": 15133.0,
"estimated_full_test_time_s": 329.8
},
"flux_image_t2i_layerwise_cpu_offload_5090": {
"stages_ms": {
"InputValidationStage": 0.11,
"TextEncodingStage": 204.63,
"LatentPreparationStage": 0.72,
"TimestepPreparationStage": 71.6,
"DenoisingStage": 3603.59,
"DecodingStage": 37.95
},
"denoise_step_ms": {
"0": 400.73,
"1": 682.43,
"2": 876.57,
"3": 877.12
},
"expected_e2e_ms": 4604.15,
"expected_avg_denoise_ms": 709.21,
"expected_median_denoise_ms": 779.5,
"load_peak_vram_mb": 4916.0,
"runtime_peak_vram_mb": 19146.0,
"load_peak_allocated_mb": 4880.0,
"runtime_peak_allocated_mb": 14216.0,
"load_peak_host_anon_mb": 10559.66,
"runtime_peak_host_anon_mb": 10559.66,
"estimated_full_test_time_s": 172.8
},
"lingbot_video_moe_t2v": {
"stages_ms": {
"InputValidationStage": 0.1,
"LingBotVideoTextEncodingStage": 270.0,
"LatentPreparationStage": 1.2,
"TimestepPreparationStage": 0.2,
"DenoisingStage": 26414.5,
"DecodingStage": 628.3
},
"denoise_step_ms": {
"0": 2200.8,
"2": 2200.8,
"4": 2200.8,
"7": 2200.8,
"9": 2200.8,
"11": 2200.8
},
"expected_e2e_ms": 27740.0,
"expected_avg_denoise_ms": 2200.8,
"expected_median_denoise_ms": 2200.8,
"estimated_full_test_time_s": 90.0
},
"minimax_h3_t2va_consumer_budget_1gpu_5090": {
"stages_ms": {
"InputValidationStage": 0.06,
"MiniMaxH3PartitionAdmissionStage": 0.03,
"MiniMaxH3TextEncodingStage": 1825.86,
"MiniMaxH3VisualEncodingStage": 0.09,
"MiniMaxH3AudioEncodingStage": 0.06,
"MiniMaxH3LatentPreparationStage": 6.56,
"MiniMaxH3TimestepPreparationStage": 0.28,
"MiniMaxH3DenoisingStage": 34839.69,
"MiniMaxH3DecodingStage": 5257.87
},
"denoise_step_ms": {
"0": 4941.69,
"1": 4863.02,
"2": 4223.75,
"3": 4384.88,
"4": 6182.67,
"5": 5068.86,
"6": 4579.34
},
"expected_e2e_ms": 42763.29,
"expected_avg_denoise_ms": 4892.03,
"expected_median_denoise_ms": 4863.02,
"load_peak_vram_mb": 5052.0,
"runtime_peak_vram_mb": 12278.0,
"load_peak_allocated_mb": 4995.0,
"runtime_peak_allocated_mb": 11483.0,
"load_peak_host_anon_mb": 32768.0,
"runtime_peak_host_anon_mb": 32768.0,
"estimated_full_test_time_s": 283.0
}
}
}