Fix spelling errors (#33978)

tests/kernels/moe/test_cutedsl_moe.py

@@ -139,7 +139,7 @@ def prepare_inputs(
         masked_m.append(mask.sum())
 
     masked_m = torch.tensor(masked_m, dtype=torch.int32)
-    # Intialize the hidden_states_3d with ones instead of empty to avoid nan
+    # Initialize the hidden_states_3d with ones instead of empty to avoid nan
     # issue.
     hidden_states_3d = torch.ones(
         (num_experts, max(masked_m), hidden_states.shape[1]), dtype=hidden_states.dtype

tests/kernels/moe/test_moe_align_block_size.py

@@ -334,7 +334,7 @@ def test_batched_moe_align_block_size(
         ref_expert_ids = torch.empty((Msum // block_size,), dtype=torch.int32)
         ref_num_tokens_post_pad = torch.empty((1,), dtype=torch.int32)
 
-        # Intialize
+        # Initialize
         sentinel = E * max_tokens_per_batch
         ref_sorted_ids.fill_(sentinel)
         ref_expert_ids.fill_(-1)

tests/reasoning/test_hunyuan_reasoning_parser.py

@@ -12,7 +12,7 @@ START_REASONING = "<think>\n"
 START_RESPONSE = "\n</think>\n<answer>\n"
 END_RESPONSE = "\n</answer>"
 
-NO_REASONING_QUICK_THROUGHT = {
+NO_REASONING_QUICK_THOUGHT = {
     "output": f"{START_REASONING}{START_RESPONSE}This is the rest{END_RESPONSE}",  # noqa: E501
     "reasoning": None,
     "content": "This is the rest",
@@ -81,7 +81,7 @@ TEST_CASES = [
         NO_REASONING,
         id="no_reasoning",
     ),
-    pytest.param(False, NO_REASONING_QUICK_THROUGHT, id="no_reasoning_quick"),
+    pytest.param(False, NO_REASONING_QUICK_THOUGHT, id="no_reasoning_quick"),
     pytest.param(
         False,
         MULTIPLE_LINES,
@@ -117,7 +117,7 @@ TEST_CASES = [
         NO_REASONING,
         id="no_reasoning_streaming",
     ),
-    pytest.param(True, NO_REASONING_QUICK_THROUGHT, id="no_reasoning_quick_stream"),
+    pytest.param(True, NO_REASONING_QUICK_THOUGHT, id="no_reasoning_quick_stream"),
     pytest.param(
         True,
         MULTIPLE_LINES,

vllm/model_executor/layers/fused_moe/oracle/fp8.py

@@ -333,7 +333,7 @@ def select_fp8_moe_backend(
 
     # TODO(rob): per discussion with TPU team, we need a way to register
     # MoE backends by OOT plugins, rather than having an explicit list
-    # of AVAILBLE_BACKENDS. Enabling returning `Fp8MoeBackend.NONE` is
+    # of AVAILABLE_BACKENDS. Enabling returning `Fp8MoeBackend.NONE` is
     # a temporary measure until these register APIs are complete.
     if current_platform.is_cuda() or current_platform.is_rocm():
         raise NotImplementedError(

vllm/v1/attention/ops/flashmla.py

@@ -52,8 +52,8 @@ def is_flashmla_dense_supported() -> tuple[bool, str | None]:
     """
     Return: is_supported_flag, unsupported_reason (optional).
     """
-    is_availble, maybe_reason = _is_flashmla_available()
-    if not is_availble:
+    is_available, maybe_reason = _is_flashmla_available()
+    if not is_available:
         return False, maybe_reason
     if not current_platform.is_device_capability_family(90):
         return False, "FlashMLA Dense is only supported on Hopper devices."
@@ -64,8 +64,8 @@ def is_flashmla_sparse_supported() -> tuple[bool, str | None]:
     """
     Return: is_supported_flag, unsupported_reason (optional).
     """
-    is_availble, maybe_reason = _is_flashmla_available()
-    if not is_availble:
+    is_available, maybe_reason = _is_flashmla_available()
+    if not is_available:
         return False, maybe_reason
     if not (
         current_platform.is_device_capability_family(90)