chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[diffusion] CI: deprecate WarmupRunner in CI (#18038)

Browse Source 
Co-authored-by: Mick <mickjagger19@icloud.com>
This commit is contained in:
陈一涵
2026-02-02 15:16:08 +08:00
committed by GitHub
co-authored by Mick
parent 522e13b4d2
commit 86117dfe0e
6 changed files with 123 additions and 273 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/test/server/testcase_configs.py
+36 -38
View File
@@ -98,7 +98,6 @@ class BaselineConfig:
scenarios: dict[str, ScenarioConfig]
step_fractions: Sequence[float]
warmup_defaults: dict[str, int]
tolerances: ToleranceConfig
improvement_threshold: float
@@ -127,7 +126,6 @@ class BaselineConfig:
return cls(
scenarios=scenarios,
step_fractions=tuple(data["sampling"]["step_fractions"]),
warmup_defaults=data["sampling"].get("warmup_requests", {}),
tolerances=tolerances,
improvement_threshold=data.get("improvement_reporting", {}).get(
"threshold", 0.2
@@ -142,7 +140,6 @@ class DiffusionServerArgs:
model_path: str # HF repo or local path
modality: str = "image" # "image" or "video" or "3d"
warmup: int = 1 # number of warmup requests
custom_validator: str | None = None # optional custom validator name
# resources
num_gpus: int = 1
@@ -161,7 +158,7 @@ class DiffusionServerArgs:
None # Second LoRA adapter path for multi-LoRA testing
)
# misc
enable_warmup: bool = False
warmup: bool = False
dit_layerwise_offload: bool = False
dit_offload_prefetch_size: int | float | None = None
@@ -325,6 +322,7 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Qwen/Qwen-Image",
modality="image",
warmup=True,
),
T2I_sampling_params,
),
@@ -334,22 +332,21 @@ 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",
model_path="black-forest-labs/FLUX.1-dev", modality="image", warmup=True
),
T2I_sampling_params,
),
DiffusionTestCase(
"flux_2_image_t2i",
DiffusionServerArgs(
model_path="black-forest-labs/FLUX.2-dev",
modality="image",
model_path="black-forest-labs/FLUX.2-dev", modality="image", warmup=True
),
T2I_sampling_params,
),
@@ -358,6 +355,7 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="black-forest-labs/FLUX.2-klein-4B",
modality="image",
warmup=True,
),
T2I_sampling_params,
),
@@ -371,21 +369,21 @@ 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",
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", enable_warmup=True
model_path="Tongyi-MAI/Z-Image-Turbo", modality="image", warmup=True
),
T2I_sampling_params,
),
@@ -397,6 +395,7 @@ 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,
),
@@ -404,32 +403,28 @@ ONE_GPU_CASES_A: list[DiffusionTestCase] = [
DiffusionTestCase(
"qwen_image_edit_ti2i",
DiffusionServerArgs(
model_path="Qwen/Qwen-Image-Edit",
modality="image",
model_path="Qwen/Qwen-Image-Edit", modality="image", warmup=True
),
TI2I_sampling_params,
),
DiffusionTestCase(
"qwen_image_edit_2509_ti2i",
DiffusionServerArgs(
model_path="Qwen/Qwen-Image-Edit-2509",
modality="image",
model_path="Qwen/Qwen-Image-Edit-2509", modality="image", warmup=True
),
MULTI_IMAGE_TI2I_sampling_params,
),
DiffusionTestCase(
"qwen_image_edit_2511_ti2i",
DiffusionServerArgs(
model_path="Qwen/Qwen-Image-Edit-2511",
modality="image",
model_path="Qwen/Qwen-Image-Edit-2511", modality="image", warmup=True
),
TI2I_sampling_params,
),
DiffusionTestCase(
"qwen_image_layered_i2i",
DiffusionServerArgs(
model_path="Qwen/Qwen-Image-Layered",
modality="image",
model_path="Qwen/Qwen-Image-Layered", modality="image", warmup=True
),
MULTI_FRAME_I2I_sampling_params,
),
@@ -442,8 +437,8 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -454,9 +449,9 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
text_encoder_cpu_offload=True,
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -468,8 +463,8 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -484,10 +479,10 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
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",
@@ -508,8 +503,7 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionTestCase(
"flux_2_ti2i",
DiffusionServerArgs(
model_path="black-forest-labs/FLUX.2-dev",
modality="image",
model_path="black-forest-labs/FLUX.2-dev", modality="image", warmup=True
),
TI2I_sampling_params,
),
@@ -518,8 +512,8 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="FastVideo/FastHunyuan-diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -531,8 +525,8 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.2-TI2V-5B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
TI2V_sampling_params,
),
@@ -541,8 +535,8 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
DiffusionServerArgs(
model_path="FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
TI2V_sampling_params,
),
@@ -556,8 +550,8 @@ if not current_platform.is_hip():
DiffusionServerArgs(
model_path="IPostYellow/TurboWan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -571,8 +565,8 @@ TWO_GPU_CASES_A = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.2-I2V-A14B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
warmup=True,
),
TI2V_sampling_params,
),
@@ -581,9 +575,9 @@ TWO_GPU_CASES_A = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.2-T2V-A14B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
num_gpus=2,
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -595,10 +589,10 @@ TWO_GPU_CASES_A = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.2-T2V-A14B-Diffusers",
modality="video",
warmup=0,
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",
@@ -608,10 +602,10 @@ TWO_GPU_CASES_A = [
"wan2_1_t2v_14b_2gpu",
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-14B-Diffusers",
warmup=0,
modality="video",
num_gpus=2,
custom_validator="video",
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -623,10 +617,10 @@ TWO_GPU_CASES_A = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
num_gpus=2,
cfg_parallel=True,
warmup=True,
),
DiffusionSamplingParams(
prompt=T2V_PROMPT,
@@ -642,10 +636,10 @@ if not current_platform.is_hip():
DiffusionServerArgs(
model_path="IPostYellow/TurboWan2.2-I2V-A14B-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
num_gpus=2,
tp_size=2,
warmup=True,
),
TURBOWAN_I2V_sampling_params,
)
@@ -656,10 +650,10 @@ TWO_GPU_CASES_B = [
"wan2_1_i2v_14b_480P_2gpu",
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-I2V-14B-480P-Diffusers",
warmup=0,
modality="video",
custom_validator="video",
num_gpus=2,
warmup=True,
),
TI2V_sampling_params,
),
@@ -669,10 +663,10 @@ TWO_GPU_CASES_B = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-I2V-14B-720P-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
num_gpus=2,
lora_path="starsfriday/Wan2.1-Divine-Power-LoRA",
warmup=True,
),
TI2V_sampling_params,
),
@@ -681,9 +675,9 @@ TWO_GPU_CASES_B = [
DiffusionServerArgs(
model_path="Wan-AI/Wan2.1-I2V-14B-720P-Diffusers",
modality="video",
warmup=0,
custom_validator="video",
num_gpus=2,
warmup=True,
),
TI2V_sampling_params,
),
@@ -696,6 +690,7 @@ TWO_GPU_CASES_B = [
# test ring attn
ulysses_degree=1,
ring_degree=2,
warmup=True,
),
T2I_sampling_params,
),
@@ -706,6 +701,7 @@ TWO_GPU_CASES_B = [
modality="image",
num_gpus=2,
ulysses_degree=2,
warmup=True,
),
T2I_sampling_params,
),
@@ -715,6 +711,7 @@ TWO_GPU_CASES_B = [
model_path="black-forest-labs/FLUX.1-dev",
modality="image",
num_gpus=2,
warmup=True,
),
T2I_sampling_params,
),
@@ -725,6 +722,7 @@ TWO_GPU_CASES_B = [
modality="image",
num_gpus=2,
tp_size=2,
warmup=True,
),
T2I_sampling_params,
),