[diffusion] CI: add testcase for cfg parallel (#17056)

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

@@ -804,6 +804,73 @@
             "expected_avg_denoise_ms": 260.76,
             "expected_median_denoise_ms": 247.84
         },
+        "wan2_1_t2v_1.3b_cfg_parallel": {
+            "stages_ms": {
+                "InputValidationStage": 0.09,
+                "TextEncodingStage": 1834.21,
+                "ConditioningStage": 0.04,
+                "TimestepPreparationStage": 3.18,
+                "LatentPreparationStage": 1.27,
+                "DenoisingStage": 6807.91,
+                "DecodingStage": 891.98,
+                "per_frame_generation": null
+            },
+            "denoise_step_ms": {
+                "0": 847.76,
+                "1": 120.56,
+                "2": 173.14,
+                "3": 147.17,
+                "4": 118.31,
+                "5": 107.02,
+                "6": 111.75,
+                "7": 107.35,
+                "8": 108.29,
+                "9": 106.22,
+                "10": 108.6,
+                "11": 105.37,
+                "12": 111.32,
+                "13": 104.72,
+                "14": 104.12,
+                "15": 106.77,
+                "16": 104.02,
+                "17": 107.16,
+                "18": 106.08,
+                "19": 105.54,
+                "20": 105.53,
+                "21": 106.69,
+                "22": 106.55,
+                "23": 106.72,
+                "24": 106.66,
+                "25": 107.57,
+                "26": 106.68,
+                "27": 110.17,
+                "28": 108.17,
+                "29": 107.83,
+                "30": 106.11,
+                "31": 109.55,
+                "32": 134.08,
+                "33": 132.01,
+                "34": 107.83,
+                "35": 106.15,
+                "36": 106.1,
+                "37": 106.32,
+                "38": 106.08,
+                "39": 106.71,
+                "40": 109.31,
+                "41": 110.77,
+                "42": 107.72,
+                "43": 106.32,
+                "44": 106.3,
+                "45": 108.05,
+                "46": 106.09,
+                "47": 107.59,
+                "48": 106.98,
+                "49": 105.91
+            },
+            "expected_e2e_ms": 9549.87,
+            "expected_avg_denoise_ms": 125.6,
+            "expected_median_denoise_ms": 107.0
+        },
         "turbo_wan2_1_t2v_1.3b": {
             "stages_ms": {
                 "InputValidationStage": 0.06,

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

@@ -149,6 +149,7 @@ class DiffusionServerArgs:
     tp_size: int | None = None
     ulysses_degree: int | None = None
     ring_degree: int | None = None
+    cfg_parallel: bool | None = None
     # LoRA
     lora_path: str | None = (
         None  # LoRA adapter path (HF repo or local path, loaded at startup)
@@ -576,6 +577,20 @@ TWO_GPU_CASES_A = [
             output_size="832x480",
         ),
     ),
+    DiffusionTestCase(
+        "wan2_1_t2v_1.3b_cfg_parallel",
+        DiffusionServerArgs(
+            model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+            modality="video",
+            warmup=0,
+            custom_validator="video",
+            num_gpus=2,
+            cfg_parallel=True,
+        ),
+        DiffusionSamplingParams(
+            prompt=T2V_PROMPT,
+        ),
+    ),
 ]
 
 # Skip turbowan because Triton requires 81920 shared memory, but AMD only has 65536.