diff --git a/python/sglang/srt/layers/attention/double_sparsity_backend.py b/python/sglang/srt/layers/attention/double_sparsity_backend.py
index 76a63a093..e8ee1538b 100644
--- a/python/sglang/srt/layers/attention/double_sparsity_backend.py
+++ b/python/sglang/srt/layers/attention/double_sparsity_backend.py
@@ -32,7 +32,7 @@ class DoubleSparseAttnBackend(AttentionBackend):
         self.heavy_token_num = model_runner.server_args.ds_heavy_token_num
 
         self.sorted_channels = model_runner.sorted_channels
-        self.sparse_decode_thresold = (
+        self.sparse_decode_threshold = (
             model_runner.server_args.ds_sparse_decode_threshold
         )
         self.att_out_approx: torch.Tensor = None
@@ -210,7 +210,7 @@ class DoubleSparseAttnBackend(AttentionBackend):
         #            and set a minimum value for sparse_decode
         if (
             min_seq_len < self.heavy_token_num
-            or max_seq_len < self.sparse_decode_thresold
+            or max_seq_len < self.sparse_decode_threshold
         ):
             self.decode_attention_fwd(
                 q.view(-1, layer.tp_q_head_num, layer.qk_head_dim),
diff --git a/python/sglang/srt/layers/quantization/fp8_kernel.py b/python/sglang/srt/layers/quantization/fp8_kernel.py
index 2022c3e8b..95d8ac46a 100644
--- a/python/sglang/srt/layers/quantization/fp8_kernel.py
+++ b/python/sglang/srt/layers/quantization/fp8_kernel.py
@@ -219,7 +219,7 @@ def _per_token_group_quant_8bit_raw(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         group_size: The group size used for quantization.
         eps: The minimum to avoid dividing zero.
         dtype: The dype of output tensor.
@@ -635,7 +635,7 @@ def static_quant_fp8(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         x_s: The quantization scale.
         repeat_scale: Whether to broadcast per-tensor scale to per-channel scale.
         dtype: The dype of output tensor.
diff --git a/python/sglang/srt/layers/quantization/int8_kernel.py b/python/sglang/srt/layers/quantization/int8_kernel.py
index 91cba1c32..9a122cad5 100644
--- a/python/sglang/srt/layers/quantization/int8_kernel.py
+++ b/python/sglang/srt/layers/quantization/int8_kernel.py
@@ -143,7 +143,7 @@ def per_token_group_quant_int8(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         group_size: The group size used for quantization.
         eps: The minimum to avoid dividing zero.
         dtype: The dype of output tensor. Note that only `torch.int8` is supported for now.
diff --git a/python/sglang/srt/layers/quantization/quark/utils.py b/python/sglang/srt/layers/quantization/quark/utils.py
index 6f7d25c73..e3a9cd6e6 100644
--- a/python/sglang/srt/layers/quantization/quark/utils.py
+++ b/python/sglang/srt/layers/quantization/quark/utils.py
@@ -128,10 +128,10 @@ def b_dynamic_mxfp4_quant(x):
     return x.view(h, b, d // 2), x_scales.view(h, b, d // 32)
 
 
-def mxfp4_to_f32(x, is_threed):
+def mxfp4_to_f32(x, is_3d):
     # 2 because we pack fp4 in uint8.
     x = x.repeat_interleave(2, dim=-1)
-    if is_threed:
+    if is_3d:
         x[..., ::2] = x[..., ::2] & 0xF
         x[..., 1::2] = x[..., 1::2] >> 4
     else:
diff --git a/python/sglang/srt/managers/io_struct.py b/python/sglang/srt/managers/io_struct.py
index c8074fa86..2135f5ecf 100644
--- a/python/sglang/srt/managers/io_struct.py
+++ b/python/sglang/srt/managers/io_struct.py
@@ -1271,7 +1271,7 @@ class UpdateWeightFromDiskReqInput(BaseReq):
     torch_empty_cache: bool = False
     # Whether to keep the scheduler paused after weight update
     keep_pause: bool = False
-    # Whether to recapture cuda graph after weight udpdate
+    # Whether to recapture cuda graph after weight update
     recapture_cuda_graph: bool = False
     # The trainer step id. Used to know which step's weights are used for sampling.
     token_step: int = 0
diff --git a/test/registered/rl/test_multi_instance_release_memory_occupation.py b/test/registered/rl/test_multi_instance_release_memory_occupation.py
index baf614508..da66b6ca6 100644
--- a/test/registered/rl/test_multi_instance_release_memory_occupation.py
+++ b/test/registered/rl/test_multi_instance_release_memory_occupation.py
@@ -32,7 +32,7 @@ TEST_SUITE = dict(
 
 class EngineWrapper:
     """
-    A wrapper around Sglang engine to mock multi instance cases such as RL traing.
+    A wrapper around Sglang engine to mock multi instance cases such as RL training.
 
     """
 
diff --git a/test/registered/utils/test_bench_typebaseddispatcher.py b/test/registered/utils/test_bench_typebaseddispatcher.py
index 2feeaf79c..02cccc4dc 100644
--- a/test/registered/utils/test_bench_typebaseddispatcher.py
+++ b/test/registered/utils/test_bench_typebaseddispatcher.py
@@ -198,17 +198,17 @@ def test_edge_case():
     assert result1 == result2
     print("Pass for normal test")
 
-    class UnkownType:
+    class UnknownType:
         pass
 
     try:
-        list_dispatcher(UnkownType())
+        list_dispatcher(UnknownType())
         print("exception was thrown from list version as expected")
     except ValueError:
         print("exception thrown from list version was processed...")
 
     try:
-        dict_dispatcher(UnkownType())
+        dict_dispatcher(UnknownType())
         print("exception was thrown from dict version as expected")
     except ValueError:
         print("exception thrown from dict version was processed...")