[Frontend][Model] Support batch request with refined OmniDiffusionReq… (#797)

Signed-off-by: Huang, Zeyu <11222265+fhfuih@users.noreply.github.com>

tests/e2e/offline_inference/test_cache_dit.py

@@ -15,6 +15,8 @@ from pathlib import Path
 import pytest
 import torch
 
+from vllm_omni.inputs.data import OmniDiffusionSamplingParams
+
 # ruff: noqa: E402
 REPO_ROOT = Path(__file__).resolve().parents[2]
 if str(REPO_ROOT) not in sys.path:
@@ -55,12 +57,14 @@ def test_cache_dit(model_name: str):
 
         outputs = m.generate(
             "a photo of a cat sitting on a laptop keyboard",
-            height=height,
-            width=width,
-            num_inference_steps=num_inference_steps,
-            guidance_scale=0.0,
-            generator=torch.Generator("cuda").manual_seed(42),
-            num_outputs_per_prompt=1,  # Single output for speed
+            OmniDiffusionSamplingParams(
+                height=height,
+                width=width,
+                num_inference_steps=num_inference_steps,
+                guidance_scale=0.0,
+                generator=torch.Generator("cuda").manual_seed(42),
+                num_outputs_per_prompt=1,  # Single output for speed
+            ),
         )
         # Extract images from request_output[0]['images']
         first_output = outputs[0]