fix optional and profling

vllm/compilation/decorators.py

@@ -32,6 +32,7 @@ from vllm.utils.torch_utils import supports_dynamo
 
 from .monitor import start_monitoring_torch_compile
 from vllm.config import VllmConfig
+from vllm.forward_context import get_profilling
 
 logger = init_logger(__name__)
 

vllm/model_executor/models/deepseek_v2.py

@@ -232,9 +232,9 @@ class DeepseekV2MLP(nn.Module):
         self.act_fn = SiluAndMul()
 
     def forward(self, x,
-                rms_weight: Optional[torch.Tensor] = None,
-                residual: Optional[torch.Tensor] = None,
-                update_hd: Optional[bool] = False
+                rms_weight: torch.Tensor | None = None,
+                residual: torch.Tensor | None = None,
+                update_hd: bool | None = False
                 ):
         if envs.USE_FUSED_RMS_QUANT:
             gate_up, new_resi, _  = self.gate_up_proj(x, rms_weight, residual, update_hd=update_hd)
@@ -356,8 +356,8 @@ class DeepseekV2MoE(nn.Module):
         )
 
     def forward(self, hidden_states: torch.Tensor,
-                rms_weight: Optional[torch.Tensor] = None,
-                residual: Optional[torch.Tensor] = None
+                rms_weight: torch.Tensor | None = None,
+                residual: torch.Tensor | None = None
                 ) -> torch.Tensor:
         num_tokens, hidden_dim = hidden_states.shape
         hidden_states = hidden_states.view(-1, hidden_dim)