diff --git a/python/sglang/multimodal_gen/test/scripts/gen_perf_baselines.py b/python/sglang/multimodal_gen/test/scripts/gen_perf_baselines.py index 45d0c6edc..4a3a385d0 100644 --- a/python/sglang/multimodal_gen/test/scripts/gen_perf_baselines.py +++ b/python/sglang/multimodal_gen/test/scripts/gen_perf_baselines.py @@ -65,8 +65,9 @@ def _build_server_extra_args(case: DiffusionTestCase) -> str: a += f" --ring-degree {server_args.ring_degree}" if server_args.lora_path: a += f" --lora-path {server_args.lora_path}" - if server_args.warmup: - a += " --warmup" + + # default warmup + a += " --warmup" for extra_arg in server_args.extras: a += f" {extra_arg}" diff --git a/python/sglang/multimodal_gen/test/server/ascend/testcase_configs_npu.py b/python/sglang/multimodal_gen/test/server/ascend/testcase_configs_npu.py index 5a3710733..c565aa4f8 100644 --- a/python/sglang/multimodal_gen/test/server/ascend/testcase_configs_npu.py +++ b/python/sglang/multimodal_gen/test/server/ascend/testcase_configs_npu.py @@ -12,7 +12,6 @@ ONE_NPU_CASES: list[DiffusionTestCase] = [ DiffusionServerArgs( model_path="/root/.cache/modelscope/hub/models/Wan-AI/Wan2.1-T2V-1.3B-Diffusers", modality="video", - warmup=True, custom_validator="video", ), DiffusionSamplingParams( diff --git a/python/sglang/multimodal_gen/test/server/perf_baselines.json b/python/sglang/multimodal_gen/test/server/perf_baselines.json index dcef0288a..2543fa547 100644 --- a/python/sglang/multimodal_gen/test/server/perf_baselines.json +++ b/python/sglang/multimodal_gen/test/server/perf_baselines.json @@ -523,28 +523,28 @@ }, "zimage_image_t2i": { "stages_ms": { - "ConditioningStage": 0.01, - "DecodingStage": 33.93, - "TextEncodingStage": 409.74, "InputValidationStage": 0.03, - "DenoisingStage": 808.69, - "LatentPreparationStage": 0.14, - "TimestepPreparationStage": 1.33 + "TextEncodingStage": 403.47, + "ConditioningStage": 0.01, + "TimestepPreparationStage": 1.41, + "LatentPreparationStage": 0.11, + "DenoisingStage": 756.21, + "DecodingStage": 29.41 }, "denoise_step_ms": { - "0": 23.4, - "1": 28.26, - "2": 100.72, - "3": 100.48, - "4": 99.62, - "5": 99.83, - "6": 99.67, - "7": 99.3, - "8": 100.83 + "0": 22.29, + "1": 75.04, + "2": 93.82, + "3": 93.34, + "4": 93.38, + "5": 93.58, + "6": 94.01, + "7": 93.97, + "8": 94.32 }, - "expected_e2e_ms": 1461.08, - "expected_avg_denoise_ms": 89.36, - "expected_median_denoise_ms": 99.67 + "expected_e2e_ms": 1192.92, + "expected_avg_denoise_ms": 83.75, + "expected_median_denoise_ms": 93.58 }, "zimage_image_t2i_multi_lora": { "stages_ms": { @@ -596,31 +596,6 @@ "expected_avg_denoise_ms": 144.49, "expected_median_denoise_ms": 156.16 }, - "zimage_image_t2i_warmup": { - "stages_ms": { - "InputValidationStage": 0.02, - "TextEncodingStage": 456.16, - "ConditioningStage": 0.01, - "TimestepPreparationStage": 0.98, - "LatentPreparationStage": 0.06, - "DenoisingStage": 889.42, - "DecodingStage": 37.81 - }, - "denoise_step_ms": { - "0": 16.49, - "1": 94.63, - "2": 109.65, - "3": 110.05, - "4": 109.39, - "5": 110.58, - "6": 109.52, - "7": 110.54, - "8": 115.24 - }, - "expected_e2e_ms": 1184.45, - "expected_avg_denoise_ms": 98.46, - "expected_median_denoise_ms": 109.65 - }, "qwen_image_edit_ti2i": { "stages_ms": { "LatentPreparationStage": 0.16, diff --git a/python/sglang/multimodal_gen/test/server/test_server_common.py b/python/sglang/multimodal_gen/test/server/test_server_common.py index 046495a66..7d6244a8f 100644 --- a/python/sglang/multimodal_gen/test/server/test_server_common.py +++ b/python/sglang/multimodal_gen/test/server/test_server_common.py @@ -87,8 +87,8 @@ def diffusion_server(case: DiffusionTestCase) -> ServerContext: if server_args.lora_path: extra_args += f" --lora-path {server_args.lora_path}" - if server_args.warmup: - extra_args += f" --warmup" + # default warmup + extra_args += f" --warmup" for arg in server_args.extras: extra_args += f" {arg}" diff --git a/python/sglang/multimodal_gen/test/server/testcase_configs.py b/python/sglang/multimodal_gen/test/server/testcase_configs.py index f66b8801c..c1920594b 100644 --- a/python/sglang/multimodal_gen/test/server/testcase_configs.py +++ b/python/sglang/multimodal_gen/test/server/testcase_configs.py @@ -177,8 +177,6 @@ class DiffusionServerArgs: second_lora_path: str | None = ( None # Second LoRA adapter path for multi-LoRA testing ) - # misc - warmup: bool = False dit_layerwise_offload: bool = False dit_offload_prefetch_size: int | float | None = None @@ -350,7 +348,6 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [ DiffusionServerArgs( model_path="Qwen/Qwen-Image", modality="image", - warmup=True, ), T2I_sampling_params, ), @@ -360,21 +357,20 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [ model_path="Qwen/Qwen-Image", modality="image", enable_cache_dit=True, - warmup=True, ), T2I_sampling_params, ), DiffusionTestCase( "flux_image_t2i", DiffusionServerArgs( - model_path="black-forest-labs/FLUX.1-dev", modality="image", warmup=True + model_path="black-forest-labs/FLUX.1-dev", modality="image" ), T2I_sampling_params, ), DiffusionTestCase( "flux_2_image_t2i", DiffusionServerArgs( - model_path="black-forest-labs/FLUX.2-dev", modality="image", warmup=True + model_path="black-forest-labs/FLUX.2-dev", modality="image" ), T2I_sampling_params, ), @@ -383,7 +379,6 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [ DiffusionServerArgs( model_path="black-forest-labs/FLUX.2-klein-4B", modality="image", - warmup=True, ), T2I_sampling_params, ), @@ -397,22 +392,12 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [ modality="image", dit_layerwise_offload=True, dit_offload_prefetch_size=2, - warmup=True, ), T2I_sampling_params, ), DiffusionTestCase( "zimage_image_t2i", - DiffusionServerArgs( - model_path="Tongyi-MAI/Z-Image-Turbo", modality="image", warmup=True - ), - T2I_sampling_params, - ), - DiffusionTestCase( - "zimage_image_t2i_warmup", - DiffusionServerArgs( - model_path="Tongyi-MAI/Z-Image-Turbo", modality="image", warmup=True - ), + DiffusionServerArgs(model_path="Tongyi-MAI/Z-Image-Turbo", modality="image"), T2I_sampling_params, ), # Multi-LoRA test case for Z-Image-Turbo @@ -423,37 +408,28 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [ modality="image", lora_path="reverentelusarca/elusarca-anime-style-lora-z-image-turbo", second_lora_path="tarn59/pixel_art_style_lora_z_image_turbo", - warmup=True, ), T2I_sampling_params, ), # === Text and Image to Image (TI2I) === DiffusionTestCase( "qwen_image_edit_ti2i", - DiffusionServerArgs( - model_path="Qwen/Qwen-Image-Edit", modality="image", warmup=True - ), + DiffusionServerArgs(model_path="Qwen/Qwen-Image-Edit", modality="image"), TI2I_sampling_params, ), DiffusionTestCase( "qwen_image_edit_2509_ti2i", - DiffusionServerArgs( - model_path="Qwen/Qwen-Image-Edit-2509", modality="image", warmup=True - ), + DiffusionServerArgs(model_path="Qwen/Qwen-Image-Edit-2509", modality="image"), MULTI_IMAGE_TI2I_sampling_params, ), DiffusionTestCase( "qwen_image_edit_2511_ti2i", - DiffusionServerArgs( - model_path="Qwen/Qwen-Image-Edit-2511", modality="image", warmup=True - ), + DiffusionServerArgs(model_path="Qwen/Qwen-Image-Edit-2511", modality="image"), TI2I_sampling_params, ), DiffusionTestCase( "qwen_image_layered_i2i", - DiffusionServerArgs( - model_path="Qwen/Qwen-Image-Layered", modality="image", warmup=True - ), + DiffusionServerArgs(model_path="Qwen/Qwen-Image-Layered", modality="image"), MULTI_FRAME_I2I_sampling_params, ), ] @@ -466,7 +442,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers", modality="video", custom_validator="video", - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -479,7 +454,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ modality="video", custom_validator="video", text_encoder_cpu_offload=True, - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -492,7 +466,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers", modality="video", custom_validator="video", - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -510,7 +483,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ custom_validator="video", num_gpus=1, dynamic_lora_path="Cseti/Wan-LoRA-Arcane-Jinx-v1", - warmup=True, ), DiffusionSamplingParams( prompt="csetiarcane Nfj1nx with blue hair, a woman walking in a cyberpunk city at night", @@ -531,7 +503,7 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ DiffusionTestCase( "flux_2_ti2i", DiffusionServerArgs( - model_path="black-forest-labs/FLUX.2-dev", modality="image", warmup=True + model_path="black-forest-labs/FLUX.2-dev", modality="image" ), TI2I_sampling_params, ), @@ -541,7 +513,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ model_path="FastVideo/FastHunyuan-diffusers", modality="video", custom_validator="video", - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -554,7 +525,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ model_path="Wan-AI/Wan2.2-TI2V-5B-Diffusers", modality="video", custom_validator="video", - warmup=True, ), TI2V_sampling_params, ), @@ -564,7 +534,6 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [ model_path="FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers", modality="video", custom_validator="video", - warmup=True, ), TI2V_sampling_params, ), @@ -579,7 +548,6 @@ if not current_platform.is_hip(): model_path="IPostYellow/TurboWan2.1-T2V-1.3B-Diffusers", modality="video", custom_validator="video", - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -594,7 +562,6 @@ TWO_GPU_CASES_A = [ model_path="Wan-AI/Wan2.2-I2V-A14B-Diffusers", modality="video", custom_validator="video", - warmup=True, ), TI2V_sampling_params, ), @@ -605,7 +572,6 @@ TWO_GPU_CASES_A = [ modality="video", custom_validator="video", num_gpus=2, - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -620,7 +586,6 @@ TWO_GPU_CASES_A = [ custom_validator="video", num_gpus=2, lora_path="Cseti/wan2.2-14B-Arcane_Jinx-lora-v1", - warmup=True, ), DiffusionSamplingParams( prompt="Nfj1nx with blue hair, a woman walking in a cyberpunk city at night", @@ -633,7 +598,6 @@ TWO_GPU_CASES_A = [ modality="video", num_gpus=2, custom_validator="video", - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -648,7 +612,6 @@ TWO_GPU_CASES_A = [ custom_validator="video", num_gpus=2, cfg_parallel=True, - warmup=True, ), DiffusionSamplingParams( prompt=T2V_PROMPT, @@ -660,7 +623,6 @@ TWO_GPU_CASES_A = [ model_path=DEFAULT_SMALL_MODEL, modality="image", num_gpus=2, - warmup=True, extras=["--use-fsdp-inference"], ), T2I_sampling_params, @@ -678,7 +640,6 @@ if not current_platform.is_hip(): custom_validator="video", num_gpus=2, tp_size=2, - warmup=True, ), TURBOWAN_I2V_sampling_params, ) @@ -692,7 +653,6 @@ TWO_GPU_CASES_B = [ modality="video", custom_validator="video", num_gpus=2, - warmup=True, ), TI2V_sampling_params, ), @@ -705,7 +665,6 @@ TWO_GPU_CASES_B = [ custom_validator="video", num_gpus=2, lora_path="starsfriday/Wan2.1-Divine-Power-LoRA", - warmup=True, ), TI2V_sampling_params, ), @@ -716,7 +675,6 @@ TWO_GPU_CASES_B = [ modality="video", custom_validator="video", num_gpus=2, - warmup=True, ), TI2V_sampling_params, ), @@ -729,7 +687,6 @@ TWO_GPU_CASES_B = [ # test ring attn ulysses_degree=1, ring_degree=2, - warmup=True, ), T2I_sampling_params, ), @@ -740,7 +697,6 @@ TWO_GPU_CASES_B = [ modality="image", num_gpus=2, ulysses_degree=2, - warmup=True, ), T2I_sampling_params, ), @@ -750,7 +706,6 @@ TWO_GPU_CASES_B = [ model_path="black-forest-labs/FLUX.1-dev", modality="image", num_gpus=2, - warmup=True, ), T2I_sampling_params, ), @@ -761,7 +716,6 @@ TWO_GPU_CASES_B = [ modality="image", num_gpus=2, tp_size=2, - warmup=True, ), T2I_sampling_params, ),