[diffusion] feat: support default 4-step inference for Flux2-Klein distilled models (#17225)

Signed-off-by: Lancer <maruixiang6688@gmail.com>

python/sglang/multimodal_gen/registry.py

@@ -61,7 +61,10 @@ from sglang.multimodal_gen.configs.pipeline_configs.wan import (
     Wan2_2_T2V_A14B_Config,
     Wan2_2_TI2V_5B_Config,
 )
-from sglang.multimodal_gen.configs.sample.flux import FluxSamplingParams
+from sglang.multimodal_gen.configs.sample.flux import (
+    Flux2KleinSamplingParams,
+    FluxSamplingParams,
+)
 from sglang.multimodal_gen.configs.sample.glmimage import GlmImageSamplingParams
 from sglang.multimodal_gen.configs.sample.hunyuan import (
     FastHunyuanSamplingParam,
@@ -538,7 +541,7 @@ def _register_configs():
         model_detectors=[lambda hf_id: "flux.1" in hf_id.lower()],
     )
     register_configs(
-        sampling_param_cls=FluxSamplingParams,
+        sampling_param_cls=Flux2KleinSamplingParams,
         pipeline_config_cls=Flux2KleinPipelineConfig,
         hf_model_paths=[
             "black-forest-labs/FLUX.2-klein-4B",