fix some typos (#24616)

Signed-off-by: co63oc <co63oc@users.noreply.github.com>

tests/compile/test_basic_correctness.py

@@ -23,7 +23,7 @@ class TestSetting:
     fullgraph: bool
 
 
-# we cannot afford testing the full Catesian product
+# we cannot afford testing the full Cartesian product
 # of all models and all levels
 @pytest.mark.parametrize(
     "test_setting",

tests/kernels/mamba/test_mamba_ssm_ssd.py

@@ -345,7 +345,7 @@ def test_mamba_chunk_scan_cont_batch_prefill_chunking(chunk_size, seqlens):
     # in the mamba2 ssd kernels, by comparing concatenation (in the sequence
     # dimension) of chunked results with the full sequence result.
     # It is different from test_mamba_chunk_scan_cont_batch by:
-    # 1. Not using the naive torch implementaion (ssd_minimal_discrete) to get
+    # 1. Not using the naive torch implementation (ssd_minimal_discrete) to get
     #    reference outputs. Instead, it compares chunked kernel outputs to full
     #    sequence kernel outputs. This is the most straightforward way to
     #    assert chunked prefill correctness.

vllm/model_executor/layers/fla/ops/cumsum.py

@@ -179,7 +179,7 @@ def chunk_local_cumsum_vector(
     def grid(meta):
         return (triton.cdiv(meta['S'], meta['BS']), NT, B * H)
 
-    # keep cummulative normalizer in fp32
+    # keep cumulative normalizer in fp32
     # this kernel is equivalent to
     # g = g.view(B, H, NT, BT, -1).cumsum(-2).view(B, H, T, -1)
     chunk_local_cumsum_vector_kernel[grid](g_org,

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

@@ -1322,7 +1322,7 @@ class MLACommonImpl(MLAAttentionImpl[M], Generic[M]):
         k_scale: torch.Tensor,
         dcp_world_size: int,
     ):
-        assert k_scale is None, "DCP not support sacled kvcache now."
+        assert k_scale is None, "DCP not support scaled kvcache now."
         assert attn_metadata.prefill is not None
         prefill_metadata = attn_metadata.prefill
         assert prefill_metadata.chunked_context is not None

vllm/v1/worker/block_table.py

@@ -112,9 +112,9 @@ class BlockTable:
             # tokens.
             virtual_block_offsets = positions % virtual_block_size
             mask = virtual_block_offsets % self.dcp_world_size == self.dcp_rank
-            # Calcuate local block_offsets
+            # Calculate local block_offsets
             block_offsets = virtual_block_offsets // self.dcp_world_size
-            # Calcuate slot_mapping
+            # Calculate slot_mapping
             slot_mapping = block_numbers * self.block_size + block_offsets
             # Write final slots, use -1 for not-local
             self.slot_mapping_np[:req_indices.shape[0]] = np.where(