chore : correct typos [no ci] (#20041)

* fix(docs): correct typos found during code review

Non-functional changes only:
- Fixed minor spelling mistakes in comments
- Corrected typos in user-facing strings
- No variables, logic, or functional code was modified.

Signed-off-by: Marcel Petrick <mail@marcelpetrick.it>

* Update docs/backend/CANN.md

Co-authored-by: Aaron Teo <taronaeo@gmail.com>

* Revert "Auxiliary commit to revert individual files from 846d1c301281178efbc6ce6060ad34c1ebe45af8"

This reverts commit 02fcf0c7db661d5ff3eff96b2b2db9fdb7213256.

* Update tests/test-backend-ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Update tests/test-backend-ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Signed-off-by: Marcel Petrick <mail@marcelpetrick.it>
Co-authored-by: Aaron Teo <taronaeo@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

src/llama-context.cpp

@@ -158,7 +158,7 @@ llama_context::llama_context(
     cparams.op_offload = params.op_offload;
     cparams.kv_unified = params.kv_unified;
 
-    // intialized later
+    // initialized later
     cparams.pipeline_parallel = false;
 
     {
@@ -1981,7 +1981,7 @@ ggml_cgraph * llama_context::graph_reserve(
 
     ggml_backend_sched_reset(sched.get());
 
-    // when the scheduler is reset, we cannnot reuse the old graph, so we reset the previous graph result to prevent that
+    // when the scheduler is reset, we cannot reuse the old graph, so we reset the previous graph result to prevent that
     gf_res_prev->reset();
 
     // store the n_outputs as it is, and restore it afterwards

src/llama-graph.cpp

@@ -1616,7 +1616,7 @@ ggml_tensor * llm_graph_context::build_inp_attn_scale() const {
 ggml_tensor * llm_graph_context::build_inp_out_ids() const {
     // note: when all tokens are output, we could skip this optimization to spare the ggml_get_rows() calls,
     //       but this would make the graph topology depend on the number of output tokens, which can interere with
-    //       features that require constant topology such as pipline parallelism
+    //       features that require constant topology such as pipeline parallelism
     //       ref: https://github.com/ggml-org/llama.cpp/pull/14275#issuecomment-2987424471
     //if (n_outputs < n_tokens) {
     //    return nullptr;
@@ -1779,7 +1779,7 @@ ggml_tensor * llm_graph_context::build_attn_mha(
         if (v_mla) {
 #if 0
             // v_mla can be applied as a matrix-vector multiplication with broadcasting across dimension 3 == n_tokens.
-            // However, the code is optimized for dimensions 0 and 1 being large, so this is ineffient.
+            // However, the code is optimized for dimensions 0 and 1 being large, so this is inefficient.
             cur = ggml_reshape_4d(ctx0, cur, v_mla->ne[0], 1, n_head, n_tokens);
             cur = ggml_mul_mat(ctx0, v_mla, cur);
 #else

src/llama-kv-cache.cpp

@@ -583,7 +583,7 @@ llama_kv_cache::slot_info_vec_t llama_kv_cache::prepare(const std::vector<llama_
             break;
         }
 
-        // remeber the position that we found
+        // remember the position that we found
         res.push_back(sinfo_new);
 
         // store the old state of the cells in the recovery stack
@@ -1293,7 +1293,7 @@ static void set_input_kq_mask_impl(const args_set_input_kq_mask & args, float *
     }
 
     for (uint32_t s = 0; s < n_stream; ++s) {
-        // bookeeping of the KQ mask cells that could change for other tokens of the same sequence
+        // bookkeeping of the KQ mask cells that could change for other tokens of the same sequence
         std::unordered_map<llama_seq_id, uint32_t>              seq_srct;
         std::unordered_map<llama_seq_id, std::vector<uint32_t>> seq_idxs;
 

src/llama-model.cpp

@@ -1524,7 +1524,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                 }
 
                 switch (hparams.n_layer) {
-                    // TODO: Jamba layers are a bit heterogenous, so naming this is hard.
+                    // TODO: Jamba layers are a bit heterogeneous, so naming this is hard.
                     case 12: // 900M  8x???M
                     case 32: // 51B  16x?B
                     default: type = LLM_TYPE_UNKNOWN;

src/llama-vocab.cpp

@@ -1833,7 +1833,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 const char * pc = (const char *) gguf_get_arr_data(ctx, precompiled_charsmap_keyidx);
                 precompiled_charsmap.assign(pc, pc + n_precompiled_charsmap);
 #if defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-                // correct endiannes of data in precompiled_charsmap binary blob
+                // correct endianness of data in precompiled_charsmap binary blob
                 uint32_t * xcda_blob_size = (uint32_t *) &precompiled_charsmap[0];
                 *xcda_blob_size = __builtin_bswap32(*xcda_blob_size);
                 assert(*xcda_blob_size + sizeof(uint32_t) < n_precompiled_charsmap);

src/models/deepseek2.cpp

@@ -146,7 +146,7 @@ llm_build_deepseek2::llm_build_deepseek2(const llama_model & model, const llm_gr
                     cb(Qcur, "Qcur_attn_temp_scaled", il);
                 }
 
-                // note: MLA with the absorption optimzation converts into MQA (ie: GQA with 1 group)
+                // note: MLA with the absorption optimization converts into MQA (ie: GQA with 1 group)
                 cur = build_attn(inp_attn_k,
                         model.layers[il].wo, NULL,
                         Qcur, Kcur, Vcur, nullptr, nullptr, model.layers[il].wv_b, kq_scale, il);

src/models/models.h

@@ -3,7 +3,7 @@
 #include "llama-model.h"
 #include "llama-graph.h"
 
-// note: almost all graphs require atleast sqrtf, so include cmath globally
+// note: almost all graphs require at least sqrtf, so include cmath globally
 #include <cmath>
 
 //

src/unicode.cpp

@@ -773,7 +773,7 @@ static std::vector<size_t> unicode_regex_split_custom(const std::string & text,
         // tiny_aya digit grouping pattern from tokenizer.json:
         //   {"type": "Split", "pattern": {"Regex": "\\d{1,3}(?=(?:\\d{3})*\\b)"}, "behavior": "Isolated"}
         // Splits digits into groups of 3 from the right (e.g., 1234567 -> 1, 234, 567)
-        // TODO: Revisit this regex, incase there are any subtle tokenization differences with the original regex.
+        // TODO: Revisit this regex, in case there are any subtle tokenization differences with the original regex.
         bpe_offsets = unicode_regex_split_custom_afmoe(text, offsets);
     }