[diffusion] CI: add cache-dit CI tests (#19213)

python/sglang/multimodal_gen/test/server/configs/cache_dit_scm_config.yaml

@@ -0,0 +1,13 @@
+cache_config:
+  max_warmup_steps: 2
+  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: 8
+  steps_computation_mask: "medium"
+  steps_computation_policy: dynamic

python/sglang/multimodal_gen/test/server/gpu_cases.py

@@ -1,3 +1,6 @@
+from dataclasses import replace
+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,
@@ -49,6 +52,8 @@ from sglang.multimodal_gen.test.test_utils import (
     DEFAULT_WAN_2_2_TI2V_5B_MODEL_NAME_FOR_TEST,
 )
 
+_CACHE_DIT_CONFIG_DIR = Path(__file__).parent / "configs"
+
 # All test cases with clean default values
 # To test different models, simply add more DiffusionCase entries
 ONE_GPU_CASES: list[DiffusionTestCase] = [
@@ -68,6 +73,28 @@ ONE_GPU_CASES: list[DiffusionTestCase] = [
         ),
         T2I_sampling_params,
     ),
+    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(_CACHE_DIT_CONFIG_DIR / "cache_dit_scm_config.yaml"),
+            ],
+        ),
+        replace(
+            T2I_sampling_params,
+            output_size="512x512",
+            extras={"num_inference_steps": 8, "seed": 0},
+        ),
+        run_perf_check=False,
+        run_consistency_check=False,
+        run_component_accuracy_check=False,
+        run_models_api_check=False,
+        run_t2v_input_reference_check=False,
+    ),
     DiffusionTestCase(
         "flux_image_t2i",
         DiffusionServerArgs(model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST),
@@ -521,6 +548,26 @@ TWO_GPU_CASES = [
         ),
         T2V_sampling_params,
     ),
+    DiffusionTestCase(
+        "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,
+            env_vars={"SGLANG_CACHE_DIT_WARMUP": "2"},
+        ),
+        replace(
+            T2V_sampling_params,
+            output_size="832x480",
+            num_frames=5,
+            extras={"num_inference_steps": 8, "seed": 0},
+        ),
+        run_perf_check=False,
+        run_consistency_check=False,
+        run_component_accuracy_check=False,
+        run_models_api_check=False,
+        run_t2v_input_reference_check=False,
+    ),
     DiffusionTestCase(
         "fsdp-inference",
         DiffusionServerArgs(