[diffusion] quant: update modelopt quantization docs and CI coverage (#22772)

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

@@ -20,7 +20,7 @@ logger = init_logger(__name__)
 
 
 class TestDiffusionServerOneGpuB200(DiffusionServerBase):
-    """B200-targeted smoke tests for 1-GPU diffusion cases."""
+    """B200-targeted CI tests for 1-GPU ModelOpt diffusion cases."""
 
     @pytest.fixture(params=ONE_GPU_CASES_C, ids=lambda c: c.id)
     def case(self, request) -> DiffusionTestCase:

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

@@ -78,6 +78,7 @@ def diffusion_server(case: DiffusionTestCase) -> ServerContext:
     port = int(os.environ.get("SGLANG_TEST_SERVER_PORT", default_port))
     sampling_params = case.sampling_params
     extra_args = os.environ.get("SGLANG_TEST_SERVE_ARGS", "")
+    extra_args = f"--model-type diffusion {extra_args}".strip()
 
     extra_args += f" --num-gpus {server_args.num_gpus}"
 
@@ -122,6 +123,7 @@ def diffusion_server(case: DiffusionTestCase) -> ServerContext:
     env_vars = {}
     if server_args.enable_cache_dit:
         env_vars["SGLANG_CACHE_DIT_ENABLED"] = "true"
+    env_vars.update(server_args.env_vars)
 
     # start server
     wait_deadline = float(os.environ.get("SGLANG_TEST_WAIT_SECS", "1200"))

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

@@ -209,6 +209,7 @@ class DiffusionServerArgs:
     enable_warmup: bool = True
 
     extras: list[str] = field(default_factory=lambda: [])
+    env_vars: dict[str, str] = field(default_factory=dict)
 
     def __post_init__(self):
         if self.modality is None:
@@ -380,6 +381,12 @@ T2I_sampling_params = DiffusionSamplingParams(
     output_size="1024x1024",
 )
 
+MODELOPT_T2I_CI_sampling_params = DiffusionSamplingParams(
+    prompt="Doraemon is eating dorayaki",
+    output_size="768x768",
+    extras={"num_inference_steps": 12},
+)
+
 TI2I_sampling_params = DiffusionSamplingParams(
     prompt="Convert 2D style to 3D style",
     image_path="https://github.com/lm-sys/lm-sys.github.io/releases/download/test/TI2I_Qwen_Image_Edit_Input.jpg",
@@ -416,6 +423,13 @@ T2V_sampling_params = DiffusionSamplingParams(
     prompt=T2V_PROMPT,
 )
 
+MODELOPT_T2V_CI_sampling_params = DiffusionSamplingParams(
+    prompt=T2V_PROMPT,
+    output_size="640x384",
+    num_frames=17,
+    extras={"num_inference_steps": 12},
+)
+
 TI2V_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",
@@ -754,15 +768,87 @@ if not current_platform.is_hip():
         )
     )
 
-# TODO: enable on 4090/5090
-ONE_GPU_CASES_C = [
-    DiffusionTestCase(
-        "flux_2_nvfp4_t2i",
+MODELOPT_FLUX1_FP8_TRANSFORMER = "BBuf/flux1-dev-modelopt-fp8-sglang-transformer"
+MODELOPT_FLUX2_FP8_TRANSFORMER = "BBuf/flux2-dev-modelopt-fp8-sglang-transformer"
+MODELOPT_WAN22_FP8_TRANSFORMER = "BBuf/wan22-t2v-a14b-modelopt-fp8-sglang-transformer"
+MODELOPT_FLUX1_NVFP4_TRANSFORMER = "BBuf/flux1-dev-modelopt-nvfp4-sglang-transformer"
+MODELOPT_FLUX2_NVFP4_MODEL = "black-forest-labs/FLUX.2-dev-NVFP4"
+MODELOPT_WAN22_NVFP4_TRANSFORMER = (
+    "BBuf/wan22-t2v-a14b-modelopt-nvfp4-sglang-transformer"
+)
+MODELOPT_NVFP4_B200_ENV_VARS = {"SGLANG_DIFFUSION_FLASHINFER_FP4_GEMM_BACKEND": "cudnn"}
+
+
+def _make_modelopt_ci_case(
+    case_id: str,
+    *,
+    model_path: str,
+    modality: str,
+    sampling_params: DiffusionSamplingParams,
+    extras: list[str],
+    env_vars: dict[str, str] | None = None,
+) -> DiffusionTestCase:
+    return DiffusionTestCase(
+        case_id,
         DiffusionServerArgs(
-            model_path="black-forest-labs/FLUX.2-dev-NVFP4",
+            model_path=model_path,
+            modality=modality,
+            enable_warmup=False,
+            extras=extras,
+            env_vars=env_vars or {},
         ),
-        T2I_sampling_params,
+        sampling_params,
+        run_perf_check=False,
+        run_consistency_check=False,
     )
+
+
+ONE_GPU_CASES_C = [
+    _make_modelopt_ci_case(
+        "flux1_modelopt_fp8_t2i",
+        model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST,
+        modality="image",
+        sampling_params=MODELOPT_T2I_CI_sampling_params,
+        extras=["--transformer-path", MODELOPT_FLUX1_FP8_TRANSFORMER],
+    ),
+    _make_modelopt_ci_case(
+        "flux2_modelopt_fp8_t2i",
+        model_path=DEFAULT_FLUX_2_DEV_MODEL_NAME_FOR_TEST,
+        modality="image",
+        sampling_params=MODELOPT_T2I_CI_sampling_params,
+        extras=["--transformer-path", MODELOPT_FLUX2_FP8_TRANSFORMER],
+    ),
+    _make_modelopt_ci_case(
+        "wan22_modelopt_fp8_t2v",
+        model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST,
+        modality="video",
+        sampling_params=MODELOPT_T2V_CI_sampling_params,
+        extras=["--transformer-path", MODELOPT_WAN22_FP8_TRANSFORMER],
+    ),
+    _make_modelopt_ci_case(
+        "flux1_modelopt_nvfp4_t2i",
+        model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST,
+        modality="image",
+        sampling_params=MODELOPT_T2I_CI_sampling_params,
+        extras=["--transformer-path", MODELOPT_FLUX1_NVFP4_TRANSFORMER],
+        env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
+    ),
+    _make_modelopt_ci_case(
+        "flux2_modelopt_nvfp4_t2i",
+        model_path=MODELOPT_FLUX2_NVFP4_MODEL,
+        modality="image",
+        sampling_params=MODELOPT_T2I_CI_sampling_params,
+        extras=[],
+        env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
+    ),
+    _make_modelopt_ci_case(
+        "wan22_modelopt_nvfp4_t2v",
+        model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST,
+        modality="video",
+        sampling_params=MODELOPT_T2V_CI_sampling_params,
+        extras=["--transformer-path", MODELOPT_WAN22_NVFP4_TRANSFORMER],
+        env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
+    ),
 ]
 
 TWO_GPU_CASES_A = [
@@ -781,7 +867,7 @@ TWO_GPU_CASES_A = [
         ),
         T2V_sampling_params,
     ),
-    # TeaCache smoke test for Wan2.2 T2V A14B — verifies enable_teacache=True
+    # TeaCache bring-up test for Wan2.2 T2V A14B — verifies enable_teacache=True
     # doesn't crash. Perf check disabled because Wan2.2-specific TeaCache
     # coefficients are not yet calibrated (teacache_params=None, so no speedup).
     DiffusionTestCase(