[XPU][COSMOS3] removing cuda hardcode and make VLLM_VIDEO_SYNC_TIMEOUT a tunable config (#4360)

Signed-off-by: Chendi Xue <chendi.xue@intel.com>

docs/models/supported_models.md

@@ -33,7 +33,7 @@ th {
 | `ZImagePipeline` | Z-Image | `Tongyi-MAI/Z-Image-Turbo` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
 | `WanPipeline` | Wan2.1-T2V, Wan2.2-T2V, Wan2.2-TI2V | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers`, `Wan-AI/Wan2.1-T2V-14B-Diffusers`, `Wan-AI/Wan2.2-T2V-A14B-Diffusers`, `Wan-AI/Wan2.2-TI2V-5B-Diffusers` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
 | `WanImageToVideoPipeline` | Wan2.2-I2V | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
-| `Cosmos3OmniDiffusersPipeline` | Cosmos3 T2I, T2V, I2V, V2V, T2V with sound, action policy | `nvidia/Cosmos3-Nano` | ✅︎ | | | |
+| `Cosmos3OmniDiffusersPipeline` | Cosmos3 T2I, T2V, I2V, V2V, T2V with sound, action policy | `nvidia/Cosmos3-Nano` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
 | `WanSpeechToVideoPipeline` | Wan2.2-S2V | `Wan-AI/Wan2.2-S2V-14B` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
 | `Wan22VACEPipeline` | Wan2.1-VACE | `Wan-AI/Wan2.1-VACE-1.3B-diffusers`, `Wan-AI/Wan2.1-VACE-14B-diffusers` | ✅︎ | ✅︎ | ✅︎ | ✅︎ |
 | `LTX2Pipeline` | LTX-2-T2V | `Lightricks/LTX-2` | ✅︎ | ✅︎ | | |

vllm_omni/diffusion/models/cosmos3/transformer_cosmos3.py

@@ -34,6 +34,7 @@ from vllm_omni.diffusion.data import OmniDiffusionConfig
 from vllm_omni.diffusion.distributed.sp_plan import SequenceParallelInput, SequenceParallelOutput
 from vllm_omni.diffusion.forward_context import get_forward_context, is_forward_context_available
 from vllm_omni.diffusion.layers.norm import RMSNorm
+from vllm_omni.platforms import current_omni_platform
 
 logger = init_logger(__name__)
 
@@ -1419,8 +1420,8 @@ class Cosmos3VFMTransformer(nn.Module):
         # For I2V: only add to noisy tokens, not conditioned ones.
         # Conditioned frames are clean context and should not receive
         # the diffusion timestep signal.
-        with torch.autocast("cuda", enabled=True, dtype=torch.float32):
-            time_embed = self.time_embedder(timestep * self.timestep_scale)
+        with torch.autocast(current_omni_platform.device_type, enabled=False):
+            time_embed = self.time_embedder((timestep * self.timestep_scale).float())
         time_embed = time_embed.to(hidden_states.dtype)
 
         if noisy_frame_mask is not None:

vllm_omni/entrypoints/openai/api_server.py

@@ -2780,7 +2780,7 @@ async def _run_video_generation_job(
             os.unlink(reference_audio.path)
 
 
-VIDEO_SYNC_TIMEOUT_S = 600.0
+VIDEO_SYNC_TIMEOUT_S = float(os.environ.get("VLLM_OMNI_VIDEO_SYNC_TIMEOUT", 600.0))
 
 
 async def _parse_video_form(