pyg-team · akihironitta · Sep 20, 2025 · Sep 13, 2025
diff --git a/pyg_lib/csrc/ops/cpu/softmax_kernel.cpp b/pyg_lib/csrc/ops/cpu/softmax_kernel.cpp
@@ -89,7 +89,7 @@ at::Tensor softmax_csr_forward_kernel_impl(const at::Tensor& src,
                     range *= src.size(i);
                   return range;
                 }();
-                // inner_range says how many contigous elements we can visit
+                // inner_range says how many contiguous elements we can visit
                 const auto inner_range = local_dim_size * dim_stride;
                 const auto inout_offset = dim_beg * dim_stride;
                 const auto aux_offset = group_id * n_heads;
@@ -189,7 +189,7 @@ at::Tensor softmax_csr_backward_kernel_impl(const at::Tensor& out,
                     range *= out.size(i);
                   return range;
                 }();
-                // inner_range says how many contigous elements we can visit
+                // inner_range says how many contiguous elements we can visit
                 const auto inner_range = local_dim_size * dim_stride;
                 const auto inout_offset = dim_beg * dim_stride;
                 const auto sum_offset = group_id * n_heads;

diff --git a/pyg_lib/csrc/random/cpu/biased_sampling.h b/pyg_lib/csrc/random/cpu/biased_sampling.h
@@ -24,7 +24,7 @@ namespace random {
  *
  * @param bias the pointer to the start of the bias array
  *
- * @param len the length of the index/bias arraay
+ * @param len the length of the index/bias array
  *
  * @returns the chosen index
  *
@@ -141,7 +141,7 @@ index_t biased_random_alias(const index_t* idx,
  *
  * @returns (optional) the cdf array which is grouped by the neighbors of each
  * node. For each group of neighbors, the weight is exclusively summed to form a
- * cdf array. The sum of each group will be guranteed be equal to 1.
+ * cdf array. The sum of each group will be guaranteed be equal to 1.
  *
  * Example:
  *
@@ -153,7 +153,7 @@ c10::optional<at::Tensor> biased_to_cdf(const at::Tensor& rowptr,
                                         at::Tensor& bias,
                                         bool inplace);
 
-// The implementation of coverting to CDF representation for biased sampling.
+// The implementation of converting to CDF representation for biased sampling.
 template <typename scalar_t>
 void biased_to_cdf_helper(int64_t* rowptr_data,
                           size_t rowptr_size,
@@ -190,7 +190,7 @@ void biased_to_cdf_helper(int64_t* rowptr_data,
 std::pair<at::Tensor, at::Tensor> biased_to_alias(at::Tensor rowptr,
                                                   at::Tensor bias);
 
-// The implementation of coverting to alias table for biased sampling.
+// The implementation of converting to alias table for biased sampling.
 template <typename scalar_t>
 void biased_to_alias_helper(int64_t* rowptr_data,
                             size_t rowptr_size,

diff --git a/pyg_lib/csrc/random/cpu/rand_engine.h b/pyg_lib/csrc/random/cpu/rand_engine.h
@@ -42,7 +42,7 @@ class PrefetchedRandint {
   T next(T range) {
     unsigned needed = 64;
 
-    // Mutiple levels of range to save prefetched bits.
+    // Multiple levels of range to save prefetched bits.
     if (range < (1 << 16)) {
       needed = 16;
     } else if (range < (1UL << 32)) {

diff --git a/pyg_lib/csrc/sampler/cpu/dist_merge_outputs_kernel.cpp b/pyg_lib/csrc/sampler/cpu/dist_merge_outputs_kernel.cpp
@@ -109,7 +109,7 @@ merge_outputs(
       }
     });
 
-    // Remove auxilary -1 numbers:
+    // Remove auxiliary -1 numbers:
     auto neg =
         std::remove(sampled_node_ids.begin(), sampled_node_ids.end(), -1);
     sampled_node_ids.erase(neg, sampled_node_ids.end());

diff --git a/pyg_lib/csrc/sampler/cpu/neighbor_kernel.cpp b/pyg_lib/csrc/sampler/cpu/neighbor_kernel.cpp
@@ -265,7 +265,7 @@ class NeighborSampler {
     else {
       at::Tensor index;
       if (replace) {
-        // at::multinomial only has good perfomance for `replace=true`, see:
+        // at::multinomial only has good performance for `replace=true`, see:
         // https://github.com/pytorch/pytorch/issues/11931
         index = at::multinomial(weight, count, replace);
       } else {
@@ -707,7 +707,7 @@ sample(const std::vector<node_type>& node_types,
       phmap::flat_hash_map<node_type, std::vector<node_t>>
           dst_sampled_nodes_dict;
       if (parallel) {
-        for (const auto& k : threads_edge_types) {  // Intialize empty vectors.
+        for (const auto& k : threads_edge_types) {  // Initialize empty vectors.
           dst_sampled_nodes_dict[!csc ? std::get<2>(k[0]) : std::get<0>(k[0])];
         }
       }

diff --git a/pyg_lib/csrc/sampler/dist_merge_outputs.h b/pyg_lib/csrc/sampler/dist_merge_outputs.h
@@ -10,10 +10,10 @@ namespace sampler {
 // For distributed training purposes. Merges sampler outputs from different
 // partitions, so that they are sorted according to the sampling order.
 // Removes seed nodes from sampled nodes and calculates how many neighbors
-// were sampled by each source node based on the cummulative sum of sampled
+// were sampled by each source node based on the cumulative sum of sampled
 // neighbors for each input node.
 // Returns the unified node, edge and batch indices as well as the merged
-// cummulative sum of sampled neighbors.
+// cumulative sum of sampled neighbors.
 PYG_API
 std::tuple<at::Tensor,
            at::Tensor,