Skip to content

[diffusion] Add cache-dit CI tests #19213

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mickqian merged 11 commits into from
May 10, 2026
Merged

[diffusion] Add cache-dit CI tests #19213

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
0593ced
add
qimcis Feb 24, 2026
5c45e52
fix
qimcis Feb 24, 2026
10ffede
add
qimcis Feb 24, 2026
2ff693b
fix
qimcis Feb 24, 2026
018ba0b
scm
qimcis Feb 25, 2026
57bbc8b
fix baseline
qimcis Feb 25, 2026
6bad7e2
configs
qimcis Mar 10, 2026
0c272f8
add
qimcis May 9, 2026
1819297
fix
qimcis May 9, 2026
17b1be2
fix
qimcis May 9, 2026
c4cbdae
Merge branch 'main' into cachedit-test
qimcis May 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions python/sglang/multimodal_gen/test/server/configs/cache_dit_config_1gpu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
cache_config:

@mickqian mickqian Mar 10, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

gather to a configs folder?

max_warmup_steps: 8
warmup_interval: 2
max_cached_steps: -1
max_continuous_cached_steps: 2
Fn_compute_blocks: 1
Bn_compute_blocks: 0
residual_diff_threshold: 0.12
enable_taylorseer: true
taylorseer_order: 1
13 changes: 13 additions & 0 deletions python/sglang/multimodal_gen/test/server/configs/cache_dit_scm_config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
cache_config:
max_warmup_steps: 8
warmup_interval: 2
max_cached_steps: -1
max_continuous_cached_steps: 2
Fn_compute_blocks: 1
Bn_compute_blocks: 0
residual_diff_threshold: 0.12
enable_taylorseer: true
taylorseer_order: 1
num_inference_steps: 50
steps_computation_mask: "medium"
steps_computation_policy: dynamic
43 changes: 43 additions & 0 deletions python/sglang/multimodal_gen/test/server/gpu_cases.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
from pathlib import Path

from sglang.multimodal_gen.runtime.platforms import current_platform
from sglang.multimodal_gen.test.server.testcase_configs import (
MODELOPT_FLUX1_FP8_TRANSFORMER,
Expand Down Expand Up @@ -68,6 +70,36 @@
),
T2I_sampling_params,
),
DiffusionTestCase(
"qwen_image_t2i_cache_dit_config_diffusers_1gpu",
DiffusionServerArgs(
model_path=DEFAULT_QWEN_IMAGE_MODEL_NAME_FOR_TEST,
extras=[
"--backend",
"diffusers",
"--cache-dit-config",
str(Path(__file__).parent / "configs" / "cache_dit_config_1gpu.yaml"),
],
),
T2I_sampling_params,
run_consistency_check=False,
run_component_accuracy_check=False,
),
DiffusionTestCase(
"qwen_image_t2i_cache_dit_scm_config_diffusers_1gpu",
DiffusionServerArgs(
model_path=DEFAULT_QWEN_IMAGE_MODEL_NAME_FOR_TEST,
extras=[
"--backend",
"diffusers",
"--cache-dit-config",
str(Path(__file__).parent / "configs" / "cache_dit_scm_config.yaml"),
],
),
T2I_sampling_params,
run_consistency_check=False,
run_component_accuracy_check=False,
),
DiffusionTestCase(
"flux_image_t2i",
DiffusionServerArgs(model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST),
Expand Down Expand Up @@ -521,6 +553,17 @@
),
T2V_sampling_params,
),
DiffusionTestCase(

@mickqian mickqian May 9, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

these tests are heavy, could we try to make it lightweight

@qimcis qimcis May 9, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i removed a redundant test, and made some changes, is this ok? passes locally for me @mickqian

"wan2_1_t2v_1_3b_cache_dit_sp_only_2gpu",
DiffusionServerArgs(
model_path=DEFAULT_WAN_2_1_T2V_1_3B_MODEL_NAME_FOR_TEST,
ulysses_degree=2,
enable_cache_dit=True,
),
T2V_sampling_params,
run_consistency_check=False,
run_component_accuracy_check=False,
),
DiffusionTestCase(
"fsdp-inference",
DiffusionServerArgs(
Expand Down
87 changes: 87 additions & 0 deletions python/sglang/multimodal_gen/test/server/perf_baselines.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1073,6 +1073,73 @@
"expected_median_denoise_ms": 100.05,
"estimated_full_test_time_s": 127.6
},
"wan2_1_t2v_1_3b_cache_dit_sp_only_2gpu": {
"stages_ms": {
"InputValidationStage": 0.08,
"TextEncodingStage": 853.63,
"LatentPreparationStage": 0.09,
"TimestepPreparationStage": 1.51,
"DenoisingStage": 2102.64,
"DecodingStage": 438.71,
"per_frame_generation": null
},
"denoise_step_ms": {
"0": 107.59,
"1": 103.11,
"2": 98.13,
"3": 94.12,
"4": 45.26,
"5": 8.66,
"6": 8.87,
"7": 105.03,
"8": 17.53,
"9": 9.26,
"10": 15.2,
"11": 93.17,
"12": 27.27,
"13": 7.78,
"14": 8.8,
"15": 96.54,
"16": 24.72,
"17": 9.27,
"18": 8.53,
"19": 106.58,
"20": 13.47,
"21": 10.13,
"22": 10.23,
"23": 96.49,
"24": 21.93,
"25": 8.73,
"26": 9.72,
"27": 103.29,
"28": 14.4,
"29": 9.07,
"30": 8.71,
"31": 105.88,
"32": 16.93,
"33": 9.15,
"34": 9.83,
"35": 100.03,
"36": 21.5,
"37": 9.41,
"38": 9.18,
"39": 105.14,
"40": 12.36,
"41": 10.1,
"42": 8.29,
"43": 102.88,
"44": 18.36,
"45": 7.67,
"46": 10.27,
"47": 103.19,
"48": 25.53,
"49": 106.23
},
"expected_e2e_ms": 3402.6,
"expected_avg_denoise_ms": 41.87,
"expected_median_denoise_ms": 16.07,
"estimated_full_test_time_s": 94.0
},
"turbo_wan2_1_t2v_1.3b": {
"stages_ms": {
"InputValidationStage": 0.06,
Expand Down Expand Up @@ -2042,6 +2109,26 @@
"expected_median_denoise_ms": 268.51,
"estimated_full_test_time_s": 122.7
},
"qwen_image_t2i_cache_dit_config_diffusers_1gpu": {
"stages_ms": {
"DiffusersExecutionStage": 64520.18
},
"denoise_step_ms": {},
"expected_e2e_ms": 64523.93,
"expected_avg_denoise_ms": 0.0,
"expected_median_denoise_ms": 0.0,
"estimated_full_test_time_s": 142.2
},
"qwen_image_t2i_cache_dit_scm_config_diffusers_1gpu": {
"stages_ms": {
"DiffusersExecutionStage": 61106.66
},
"denoise_step_ms": {},
"expected_e2e_ms": 61110.0,
"expected_avg_denoise_ms": 0.0,
"expected_median_denoise_ms": 0.0,
"estimated_full_test_time_s": 135.3
},
"hunyuan3d_shape_gen": {
"stages_ms": {
"Hunyuan3DShapeBeforeDenoisingStage": 544.59,
Expand Down
3 changes: 2 additions & 1 deletion python/sglang/multimodal_gen/test/server/test_server_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -943,7 +943,8 @@ def generate_image(case_id, client) -> tuple[str, bytes]:

# Build extra_body for optional features
extra_body = dict(sampling_params.extras)

if sampling_params.diffusers_kwargs:
extra_body["diffusers_kwargs"] = sampling_params.diffusers_kwargs
response = client.images.with_raw_response.generate(
model=model_path,
prompt=sampling_params.prompt,
Expand Down
2 changes: 2 additions & 0 deletions python/sglang/multimodal_gen/test/server/testcase_configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,6 +247,8 @@ class DiffusionSamplingParams:
# merged directly into the OpenAI extra_body dict.
extras: dict = field(default_factory=dict)

diffusers_kwargs: dict | None = None


@dataclass(frozen=True)
class DiffusionTestCase:
Expand Down
Loading