[diffusion] model: Support TurboWan2.2-I2V SLA && add CI test for TurboWan (#16536)

python/sglang/multimodal_gen/test/server/perf_baselines.json

@@ -804,6 +804,27 @@
             "expected_avg_denoise_ms": 260.76,
             "expected_median_denoise_ms": 247.84
         },
+      "turbo_wan2_1_t2v_1.3b": {
+            "stages_ms": {
+                "InputValidationStage": 0.06,
+                "TextEncodingStage": 2508.95,
+                "ConditioningStage": 0.04,
+                "TimestepPreparationStage": 73.51,
+                "LatentPreparationStage": 1.34,
+                "DmdDenoisingStage": 1285.25,
+                "DecodingStage": 805.04,
+                "per_frame_generation": null
+            },
+            "denoise_step_ms": {
+                "0": 897.62,
+                "1": 126.04,
+                "2": 126.52,
+                "3": 128.26
+            },
+            "expected_e2e_ms": 4686.66,
+            "expected_avg_denoise_ms": 319.61,
+            "expected_median_denoise_ms": 127.39
+        },
         "wan2_2_ti2v_5b": {
             "stages_ms": {
                 "InputValidationStage": 96.27,
@@ -1092,6 +1113,28 @@
             "expected_avg_denoise_ms": 2831.00,
             "expected_median_denoise_ms": 1600.09
         },
+        "turbo_wan2_2_i2v_a14b_2gpu": {
+            "stages_ms": {
+                "InputValidationStage": 25.01,
+                "TextEncodingStage": 5198.6,
+                "ConditioningStage": 0.04,
+                "TimestepPreparationStage": 56.26,
+                "LatentPreparationStage": 1.4,
+                "ImageVAEEncodingStage": 1001.89,
+                "DmdDenoisingStage": 4487.79,
+                "DecodingStage": 821.01,
+                "per_frame_generation": null
+            },
+            "denoise_step_ms": {
+                "0": 3042.56,
+                "1": 485.88,
+                "2": 477.59,
+                "3": 475.58
+            },
+            "expected_e2e_ms": 11605.97,
+            "expected_avg_denoise_ms": 1120.4,
+            "expected_median_denoise_ms": 481.74
+        },
         "wan2_1_i2v_14b_480P_2gpu": {
             "stages_ms": {
                 "InputValidationStage": 38.23,

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

@@ -1078,7 +1078,11 @@ def get_generate_fn(
             prompt=sampling_params.prompt,
             size=sampling_params.output_size,
             seconds=video_seconds,
-            extra_body={"reference_url": sampling_params.image_path},
+            extra_body={
+                "reference_url": sampling_params.image_path,
+                "fps": sampling_params.fps,
+                "num_frames": sampling_params.num_frames,
+            },
         )
 
     def generate_text_image_to_video(case_id, client) -> str:
@@ -1102,6 +1106,10 @@ def get_generate_fn(
                 size=output_size,
                 seconds=video_seconds,
                 input_reference=fh,
+                extra_body={
+                    "fps": sampling_params.fps,
+                    "num_frames": sampling_params.num_frames,
+                },
             )
 
     if modality == "video":

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

@@ -25,6 +25,7 @@ from dataclasses import dataclass
 from pathlib import Path
 from typing import Sequence
 
+from sglang.multimodal_gen.runtime.platforms import current_platform
 from sglang.multimodal_gen.runtime.utils.perf_logger import RequestPerfRecord
 
 
@@ -300,6 +301,15 @@ TI2V_sampling_params = DiffusionSamplingParams(
     direct_url_test=True,
 )
 
+TURBOWAN_I2V_sampling_params = DiffusionSamplingParams(
+    prompt="The man in the picture slowly turns his head, his expression enigmatic and otherworldly. The camera performs a slow, cinematic dolly out, focusing on his face. Moody lighting, neon signs glowing in the background, shallow depth of field.",
+    image_path="https://is1-ssl.mzstatic.com/image/thumb/Music114/v4/5f/fa/56/5ffa56c2-ea1f-7a17-6bad-192ff9b6476d/825646124206.jpg/600x600bb.jpg",
+    direct_url_test=True,
+    output_size="960x960",
+    num_frames=4,
+    fps=4,
+)
+
 # All test cases with clean default values
 # To test different models, simply add more DiffusionCase entries
 ONE_GPU_CASES_A: list[DiffusionTestCase] = [
@@ -496,6 +506,23 @@ ONE_GPU_CASES_B: list[DiffusionTestCase] = [
     ),
 ]
 
+# Skip turbowan because Triton requires 81920 shared memory, but AMD only has 65536.
+if not current_platform.is_hip():
+    ONE_GPU_CASES_B.append(
+        DiffusionTestCase(
+            "turbo_wan2_1_t2v_1.3b",
+            DiffusionServerArgs(
+                model_path="IPostYellow/TurboWan2.1-T2V-1.3B-Diffusers",
+                modality="video",
+                warmup=0,
+                custom_validator="video",
+            ),
+            DiffusionSamplingParams(
+                prompt=T2V_PROMPT,
+            ),
+        )
+    )
+
 TWO_GPU_CASES_A = [
     DiffusionTestCase(
         "wan2_2_i2v_a14b_2gpu",
@@ -551,6 +578,23 @@ TWO_GPU_CASES_A = [
     ),
 ]
 
+# Skip turbowan because Triton requires 81920 shared memory, but AMD only has 65536.
+if not current_platform.is_hip():
+    TWO_GPU_CASES_A.append(
+        DiffusionTestCase(
+            "turbo_wan2_2_i2v_a14b_2gpu",
+            DiffusionServerArgs(
+                model_path="IPostYellow/TurboWan2.2-I2V-A14B-Diffusers",
+                modality="video",
+                warmup=0,
+                custom_validator="video",
+                num_gpus=2,
+                tp_size=2,
+            ),
+            TURBOWAN_I2V_sampling_params,
+        )
+    )
+
 TWO_GPU_CASES_B = [
     DiffusionTestCase(
         "wan2_1_i2v_14b_480P_2gpu",