Fix per file ruff ignores related to line length (#26262)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

vllm/model_executor/models/qwen3_next.py

@@ -916,8 +916,9 @@ class Qwen3NextDecoderLayer(nn.Module):
                 )
             else:
                 assert len(hidden_states.shape) == len(self.ffn_layer_scale.shape), (
-                    f"shape must be the same {len(hidden_states.shape)}, {len(self.ffn_layer_scale.shape)}"
-                )  # noqa: E501
+                    f"shape must be the same {len(hidden_states.shape)}, "
+                    f"{len(self.ffn_layer_scale.shape)}"
+                )
                 hidden_states = hidden_states * (
                     self.ffn_layer_scale.to(hidden_states.dtype) + 1
                 )

vllm/v1/attention/backends/mla/common.py

@@ -255,8 +255,8 @@ def is_rocm_aiter_fp8bmm_enabled() -> bool:
 
 
 if is_rocm_aiter_fp8bmm_enabled():
-    from aiter.ops.triton.batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant import (  # noqa: E501 # isort: skip
-        batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant as aiter_triton_fp8_bmm,
+    from aiter.ops.triton.batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant import (  # noqa: E501
+        batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant as aiter_triton_fp8_bmm,  # noqa: E501
     )
 
     def dynamic_per_batched_tensor_quant(
@@ -1284,8 +1284,10 @@ class MLACommonImpl(MLACommonBaseImpl[M], Generic[M]):
             actual_seq_lens_q=prefill.query_seq_lens.view(-1, 1, 1, 1),
             actual_seq_lens_kv=prefill.query_seq_lens.view(-1, 1, 1, 1),
             causal=True,
-            return_lse=True,  # do not support False for now
-            is_cuda_graph_compatible=True,  # Indicates actual_seq_lens are on GPU or CPU.
+            # Do not support False for now
+            return_lse=True,
+            # Indicates actual_seq_lens are on GPU or CPU.
+            is_cuda_graph_compatible=True,
         )
         if return_softmax_lse:
             return output, lse
@@ -1342,7 +1344,8 @@ class MLACommonImpl(MLACommonBaseImpl[M], Generic[M]):
             ),
             causal=False,
             return_lse=True,
-            is_cuda_graph_compatible=True,  # Indicates actual_seq_lens are on GPU or CPU.
+            # Indicates actual_seq_lens are on GPU or CPU.
+            is_cuda_graph_compatible=True,
         )
 
     def process_weights_after_loading(self, act_dtype: torch.dtype):

vllm/v1/engine/utils.py

@@ -872,10 +872,13 @@ def wait_for_engine_startup(
                 EngineHandshakeMetadata(
                     addresses=addresses,
                     parallel_config={
-                        "data_parallel_master_ip": parallel_config.data_parallel_master_ip,
-                        "data_parallel_master_port": parallel_config.data_parallel_master_port,
-                        "_data_parallel_master_port_list": parallel_config._data_parallel_master_port_list,
-                        "data_parallel_size": parallel_config.data_parallel_size,
+                        k: getattr(parallel_config, k)
+                        for k in (
+                            "data_parallel_master_ip",
+                            "data_parallel_master_port",
+                            "_data_parallel_master_port_list",
+                            "data_parallel_size",
+                        )
                     },
                 )
             )

vllm/v1/utils.py

@@ -345,13 +345,15 @@ def report_usage_stats(
 
     from vllm.model_executor.model_loader import get_architecture_class_name
 
+    parallel_config = vllm_config.parallel_config
+
     usage_message.report_usage(
         get_architecture_class_name(vllm_config.model_config),
         usage_context,
         extra_kvs={
             # Common configuration
             "dtype": str(vllm_config.model_config.dtype),
-            "tensor_parallel_size": vllm_config.parallel_config.tensor_parallel_size,
+            "tensor_parallel_size": parallel_config.tensor_parallel_size,
             "block_size": vllm_config.cache_config.block_size,
             "gpu_memory_utilization": vllm_config.cache_config.gpu_memory_utilization,
             "kv_cache_memory_bytes": vllm_config.cache_config.kv_cache_memory_bytes,
@@ -362,7 +364,7 @@ def report_usage_stats(
             "enable_lora": bool(vllm_config.lora_config),
             "enable_prefix_caching": vllm_config.cache_config.enable_prefix_caching,
             "enforce_eager": vllm_config.model_config.enforce_eager,
-            "disable_custom_all_reduce": vllm_config.parallel_config.disable_custom_all_reduce,
+            "disable_custom_all_reduce": parallel_config.disable_custom_all_reduce,
         },
     )
 

vllm/v1/worker/gpu_model_runner.py

@@ -3391,7 +3391,8 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
                                 attn_metadata[ubid][layer_name] = attn_metadata_i
                     else:
                         assert type(attn_metadata) is dict
-                        attn_metadata_i = attn_group.get_metadata_builder().build_for_cudagraph_capture(
+                        metadata_builder = attn_group.get_metadata_builder()
+                        attn_metadata_i = metadata_builder.build_for_cudagraph_capture(
                             common_attn_metadata
                         )
                         for layer_name in attn_group.layer_names: