[MISC] fix format check error (#654)

This pr makes format.sh works as expect.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm_ascend/attention/attention.py

@@ -1080,7 +1080,7 @@ class AscendMLAAttentionBackendImpl(MLAAttentionImpl):
             if len(kv_cache) > 0 and kv_cache[0].numel(
             ) > 0 and attn_metadata.num_prefills > 0:
                 slots = attn_metadata.slot_mapping
-                # NOTE: Seperate the kv cache in advance to avoid OOM or other issues
+                # NOTE: Separate the kv cache in advance to avoid OOM or other issues
                 torch_npu._npu_reshape_and_cache(key=kv_c_normed.view(
                     num_tokens, self.num_kv_heads, -1),
                                                  value=k_pe,

vllm_ascend/core/schedule_config.py

@@ -60,7 +60,7 @@ class AscendSchedulerConfig(SchedulerConfig):
             )
         if self.is_multimodal_model:
             raise NotImplementedError(
-                "currently AscendScheduler only supports LLM modles.")
+                "currently AscendScheduler only supports LLM models.")
         if self.num_scheduler_steps > 1:
             raise NotImplementedError(
                 "currently AscendScheduler doesn't support multi-step.")

vllm_ascend/distributed/llmdatadist_connector.py

@@ -57,8 +57,10 @@ def get_device_ips():
                               universal_newlines=True)
     if npu_info.returncode != 0 or not os.path.exists(HCCN_TOOL_PATH):
         raise RuntimeError("No npu-smi/hccn_tool tools provided for NPU.")
-    npu_start_idx = int(
-        re.match(r'.*\n\t([0-9]+).*', npu_info.stdout).group(1))
+    re_result = re.match(r'.*\n\t([0-9]+).*', npu_info.stdout)
+    if re_result is None:
+        raise RuntimeError("Can't find npu start index")
+    npu_start_idx = int(re_result.group(1))
     device_ip_list = []
     for ip_offset in range(world_size):
         cmd = [
@@ -68,7 +70,10 @@ def get_device_ips():
                                         stdout=subprocess.PIPE,
                                         stderr=subprocess.PIPE,
                                         universal_newlines=True)
-        device_ip = re.match(r'ipaddr:(.*)\n', device_ip_info.stdout).group(1)
+        re_result = re.match(r'ipaddr:(.*)\n', device_ip_info.stdout)
+        if re_result is None:
+            raise RuntimeError("Can't find npu ip")
+        device_ip = re_result.group(1)
         device_ip_list.append(device_ip)
     return device_ip_list
 

vllm_ascend/distributed/parallel_state.py

@@ -7,7 +7,7 @@ from vllm.distributed.parallel_state import (GroupCoordinator, get_world_group,
 # vllm-ascend will maintain its own EP GroupCoordinator and ETP GroupCoordinator for
 # customize parallel solution
 _EP: Optional[GroupCoordinator] = None
-_ETP: Optional[list[GroupCoordinator]] = None
+_ETP: Optional[GroupCoordinator] = None
 
 
 def get_ep_group() -> GroupCoordinator:
@@ -69,4 +69,4 @@ def destory_ascend_model_parallel():
     global _ETP
     if _ETP:
         _ETP.destroy()
-    _ETP = None
+    _ETP = None

vllm_ascend/ops/fused_moe.py

@@ -278,7 +278,7 @@ def fused_experts(
                                           dtype=dtype)
 
         # TODO: npu_grouped_matmul output random values at [num_valid_tokens:, ...]
-        # This created multiple NaN and index_add_ will mix them up which harms accracy
+        # This created multiple NaN and index_add_ will mix them up which harms accuracy
         # remove this mask and filter after it being fixed
         num_valid_tokens = mask.sum()
         valid_token_mask = torch.arange(

vllm_ascend/ops/rotary_embedding.py

@@ -227,7 +227,7 @@ def _set_cos_sin_cache(self, seq_len, device, dtype):
                          persistent=False)
 
 
-# TODO: Patch when aclnn ops avaiable
+# TODO: Patch when aclnn ops available
 RotaryEmbedding.forward_oot = rope_forward_oot
 DeepseekScalingRotaryEmbedding.forward = native_rope_deepseek_forward
 DeepseekScalingRotaryEmbedding._set_cos_sin_cache = _set_cos_sin_cache