update pre-commit config (#18860)

test/manual/test_vlm_accuracy.py

@@ -1,5 +1,4 @@
-"""
-"""
+""" """
 
 import unittest
 from typing import List, Optional
@@ -195,7 +194,7 @@ class TestMiniCPMV2_6Logits(VisionLLMLogitsBase):
                 "pixel_values": inputs.pixel_values,
                 "tgt_sizes": inputs.tgt_sizes,
             }
-            (hf_output, _) = self.hf_model.get_vllm_embedding(
+            hf_output, _ = self.hf_model.get_vllm_embedding(
                 model_inputs,
             )
             hf_output = hf_output.squeeze(0)

test/registered/attention/test_triton_sliding_window.py

@@ -43,12 +43,9 @@ class TestSlidingWindowAttentionTriton(CustomTestCase):
         cls.short_context_prompt = "The capital of France is"
 
         # Test prompt longer than window size
-        cls.long_context_prompt = (
-            """
+        cls.long_context_prompt = """
         Once upon a time, there was a mountain. In the mountain, there was a temple. In the temple, there was an old monk telling a story. The story was:
-        """
-            * 100
-        )
+        """ * 100
         cls.long_context_prompt += "\nNow, summarize the story in one sentence:"
 
     def _test_mmlu(self):

test/registered/layers/test_fla_layernorm_guard.py

@@ -10,7 +10,10 @@ import torch.nn.functional as F
 from sglang.srt.layers.attention.fla.layernorm_gated import (
     _layer_norm_fwd as layer_norm_fwd,
 )
-from sglang.srt.layers.attention.fla.layernorm_gated import layernorm_fn, rms_norm_ref
+from sglang.srt.layers.attention.fla.layernorm_gated import (
+    layernorm_fn,
+    rms_norm_ref,
+)
 from sglang.test.ci.ci_register import register_cuda_ci
 
 register_cuda_ci(

test/registered/quant/test_awq_dequant.py

@@ -7,6 +7,7 @@ unittest version of the AWQ Triton kernel tests.
 Run with:
     python -m unittest test_awq_dequant.py
 """
+
 import unittest
 
 import torch

test/srt/cpu/test_norm.py

@@ -222,7 +222,7 @@ class TestLayerNorm(CustomTestCase):
             x = x + residual.to(torch.float32)
             residual = x.to(orig_dtype)
 
-        (variance, mean) = torch.var_mean(x, dim=-1, keepdim=True, correction=0)
+        variance, mean = torch.var_mean(x, dim=-1, keepdim=True, correction=0)
         x = (x - mean) * torch.rsqrt(variance + variance_epsilon)
         x = x.to(orig_dtype) * weight
         if residual is None:

test/srt/cpu/test_qwen3.py

@@ -39,8 +39,8 @@ def fix_query_key_value_ordering_reshape_cat(
     ]
     # [b, sq, ng, (hn + hn + np/ng * hn + np/ng + np/ng)]
     # --> [b, sq, ng, hn], [b, sq, ng, hn], [b, sq, ng, np/ng * hn], [b, sq, ng, np/ng * hn], [b, sq, ng, np/ng], [b, sq, ng, np/ng]
-    (query, key, value, z) = torch.split(mixed_qkvz, split_arg_list_qkvz, dim=2)
-    (b, a) = torch.split(mixed_ba, split_arg_list_ba, dim=2)
+    query, key, value, z = torch.split(mixed_qkvz, split_arg_list_qkvz, dim=2)
+    b, a = torch.split(mixed_ba, split_arg_list_ba, dim=2)
 
     # [b, sq, ng, np/ng * hn] -> [b, sq, np, hn]
     value = value.reshape(value.size(0), -1, head_v_dim)