Lint

Author: gshtras

ROCm_performance.md

@@ -17,4 +17,3 @@ The custom PagedAttention kernel is enabled for dtype: bf16, fp16, block-size=16
 ## NCCL Performance environment variable
 
 For MI300x, setting environment variable NCCL_MIN_NCHANNELS=112 is expected to improve performance.
-

benchmarks/profiling/README.md

@@ -55,5 +55,3 @@ python3 rocmProfileData/tools/rpd2tracing.py trace.rpd trace.json
 ```
 
 Once the trace is converted, open the `.json` file in [Chrome](chrome://tracing/) or [Perfetto](https://ui.perfetto.dev/) for visualization.
-
-

vllm/model_executor/models/deepseek_v2.py

@@ -165,7 +165,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
                 router_logits=router_logits) * self.routed_scaling_factor
         else:
             final_hidden_states = self.experts(hidden_states=hidden_states,
-                                            router_logits=router_logits)
+                                               router_logits=router_logits)
         if shared_output is not None:
             if hidden_states.dtype != torch.float16:
                 final_hidden_states = final_hidden_states + shared_output