llama: Fix typos in multiple files

common/chat.cpp

@@ -9,7 +9,7 @@
 typedef minja::chat_template common_chat_template;
 
 struct common_chat_templates {
-    bool has_explicit_template; // Model had builtin template or template overridde was specified.
+    bool has_explicit_template; // Model had builtin template or template overridden was specified.
     std::unique_ptr<common_chat_template> template_default; // always set (defaults to chatml)
     std::unique_ptr<common_chat_template> template_tool_use;
 };
@@ -644,7 +644,7 @@ static std::string apply(
     // tmpl_inputs.now = std::chrono::system_clock::now();
 
     minja::chat_template_options tmpl_opts;
-    // To avoid double BOS / EOS tokens, we're manually removing begining / trailing tokens
+    // To avoid double BOS / EOS tokens, we're manually removing beginning / trailing tokens
     // instead of using `chat_template_options.use_bos_token = false`, since these tokens
     // may be needed inside the template / between messages too.
     auto result = tmpl.apply(tmpl_inputs, tmpl_opts);

common/common.h

@@ -321,7 +321,7 @@ struct common_params {
     bool cont_batching     = true;  // insert new sequences for decoding on-the-fly
     bool flash_attn        = false; // flash attention
     bool no_perf           = false; // disable performance metrics
-    bool ctx_shift         = true;  // context shift on inifinite text generation
+    bool ctx_shift         = true;  // context shift on infinite text generation
 
     bool input_prefix_bos  = false; // prefix BOS to user inputs, preceding input_prefix
     bool logits_all        = false; // return logits for all tokens in the batch
@@ -628,7 +628,7 @@ void common_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_si
 // Embedding utils
 //
 
-// TODO: repace embd_norm with an enum
+// TODO: replace embd_norm with an enum
 void common_embd_normalize(const float * inp, float * out, int n, int embd_norm);
 
 float common_embd_similarity_cos(const float * embd1, const float * embd2, int n);

common/stb_image.h

@@ -2155,7 +2155,7 @@ stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
    unsigned int k;
    int sgn;
    if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
-   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing
+   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s instead of continuing
 
    sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative)
    k = stbi_lrot(j->code_buffer, n);
@@ -2170,7 +2170,7 @@ stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
 {
    unsigned int k;
    if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
-   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing
+   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s instead of continuing
    k = stbi_lrot(j->code_buffer, n);
    j->code_buffer = k & ~stbi__bmask[n];
    k &= stbi__bmask[n];
@@ -2182,7 +2182,7 @@ stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
 {
    unsigned int k;
    if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
-   if (j->code_bits < 1) return 0; // ran out of bits from stream, return 0s intead of continuing
+   if (j->code_bits < 1) return 0; // ran out of bits from stream, return 0s instead of continuing
    k = j->code_buffer;
    j->code_buffer <<= 1;
    --j->code_bits;
@@ -4961,7 +4961,7 @@ static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int
    p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
    if (p == NULL) return stbi__err("outofmem", "Out of memory");
 
-   // between here and free(out) below, exitting would leak
+   // between here and free(out) below, exiting would leak
    temp_out = p;
 
    if (pal_img_n == 3) {
@@ -5901,7 +5901,7 @@ static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req
    if (tga_height > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
    if (tga_width > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
 
-   //   do a tiny bit of precessing
+   //   do a tiny bit of processing
    if ( tga_image_type >= 8 )
    {
       tga_image_type -= 8;
@@ -6832,7 +6832,7 @@ static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, i
          // 0:  not specified.
       }
 
-      // background is what out is after the undoing of the previou frame;
+      // background is what out is after the undoing of the previous frame;
       memcpy( g->background, g->out, 4 * g->w * g->h );
    }
 
@@ -7865,7 +7865,7 @@ STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *c, void *user
       1.31  (2011-06-20)
               a few more leak fixes, bug in PNG handling (SpartanJ)
       1.30  (2011-06-11)
-              added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+              added ability to load files via callbacks to accommodate custom input streams (Ben Wenger)
               removed deprecated format-specific test/load functions
               removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
               error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)

convert_hf_to_gguf.py

@@ -2664,7 +2664,7 @@ def set_gguf_parameters(self):
             self.gguf_writer.add_vision_n_wa_pattern(n_wa_pattern)
         else:
             raise ValueError(f"Unknown QwenVL model type: {self.global_config['model_type']}")
-        # default values below are taken from HF tranformers code
+        # default values below are taken from HF transformers code
         self.gguf_writer.add_vision_attention_layernorm_eps(self.global_config.get("rms_norm_eps", 1e-6))
 
     def tensor_force_quant(self, name, new_name, bid, n_dims):
@@ -2699,7 +2699,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
                 # split Conv3D into Conv2Ds
                 c1, c2, kt, kh, kw = data_torch.shape
                 del c1, c2, kh, kw  # unused
-                assert kt == 2, "Current implmentation only support temporal_patch_size of 2"
+                assert kt == 2, "Current implementation only support temporal_patch_size of 2"
                 return [
                     (gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.V_ENC_EMBD_PATCH] + ".weight"  , data_torch[:, :, 0, ...]),
                     (gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.V_ENC_EMBD_PATCH] + ".weight.1", data_torch[:, :, 1, ...]),
@@ -3912,7 +3912,7 @@ def set_gguf_parameters(self):
         super().set_gguf_parameters()
         hparams = self.hparams
         self.gguf_writer.add_vision_projector_type(gguf.VisionProjectorType.GEMMA3)
-        # default values below are taken from HF tranformers code
+        # default values below are taken from HF transformers code
         self.gguf_writer.add_vision_attention_layernorm_eps(hparams.get("layer_norm_eps", 1e-6))
         self.gguf_writer.add_vision_use_gelu(True)
         # calculate proj_scale_factor (used by tinygemma3 test model)

ggml/src/ggml-cann/acl_tensor.cpp

@@ -130,7 +130,7 @@ int64_t ggml_cann_get_mulmat_bcast_shape(
     const size_t* input_nb, const size_t* weight_nb, const size_t* dst_nb,
     int64_t* bcast_input_ne, int64_t* bcast_weight_ne, int64_t* bcast_dst_ne,
     size_t* bcast_input_nb, size_t* bcast_weight_nb, size_t* bcast_dst_nb) {
-    // input and dst shoule in same shape, except first two dims.
+    // input and dst should in same shape, except first two dims.
     GGML_ASSERT(input_ne[2] == dst_ne[2]);
     GGML_ASSERT(input_ne[3] == dst_ne[3]);
 

ggml/src/ggml-cann/aclnn_ops.cpp

@@ -541,7 +541,7 @@ void ggml_cann_pad(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
     aclTensor* acl_dst = ggml_cann_create_tensor(dst);
 
     // padding: value in the array means how much distance will be padding.
-    // the position of elements in the array means which dirction to padding,
+    // the position of elements in the array means which direction to padding,
     // each position means: [dim0.front, dim0.behind, dim1.front, dim1.behind,
     //                       dim2.front, dim2.behind, dim3.front, dim3.behind]
     int64_t paddings[] = {
@@ -863,7 +863,7 @@ void ggml_cann_rms_norm(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
     ggml_cann_release_resources(ctx, acl_src, acl_dst, acl_gamma, acl_rstd);
 }
 
-// TODO: performace is low.
+// TODO: performance is low.
 void ggml_cann_diag_mask(ggml_backend_cann_context& ctx, ggml_tensor* dst,
                          float value) {
     ggml_tensor* src = dst->src[0];

ggml/src/ggml-cpu/llamafile/sgemm.cpp

@@ -353,7 +353,7 @@ class tinyBLAS {
         if constexpr (RN > 1) {
             return mnpack<RM, RN-1, BM>(m, n, SIZE_N, BN);
         } else {
-            GGML_LOG_ERROR("mnpack<%d, %d> bloc size not supported\n", RM, (int)SIZE_N);
+            GGML_LOG_ERROR("mnpack<%d, %d> block size not supported\n", RM, (int)SIZE_N);
             GGML_ASSERT(false); // we have miss something.
         }
     }

ggml/src/ggml-kompute/ggml-kompute.cpp

@@ -1698,7 +1698,7 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
 
                         if (ggml_is_transposed(src0) ||
                             ggml_is_transposed(src1)) {
-                            fprintf(stderr, "%s: %s: matmul on tranposed tensor not supported: %u/%u\n", __func__, ggml_op_name(dst->op), src0t, src1t);
+                            fprintf(stderr, "%s: %s: matmul on transposed tensor not supported: %u/%u\n", __func__, ggml_op_name(dst->op), src0t, src1t);
                             goto not_implemented;
                         }
 

ggml/src/ggml-sycl/common.hpp

@@ -64,10 +64,10 @@ extern int g_ggml_sycl_disable_optimize;
 
 
 #define __SYCL_ARCH__ DPCT_COMPATIBILITY_TEMP
-#define VER_4VEC 610 // todo for hardward optimize.
-#define VER_GEN9 700 // todo for hardward optimize.
-#define VER_GEN12 1000000 // todo for hardward optimize.
-#define VER_GEN13 (VER_GEN12 + 1030) // todo for hardward optimize.
+#define VER_4VEC 610 // todo for hardware optimize.
+#define VER_GEN9 700 // todo for hardware optimize.
+#define VER_GEN12 1000000 // todo for hardware optimize.
+#define VER_GEN13 (VER_GEN12 + 1030) // todo for hardware optimize.
 
 #define GGML_SYCL_MAX_NODES 8192 // TODO: adapt to hardwares
 

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -1246,7 +1246,7 @@ struct ggml_sycl_pool_host : public ggml_sycl_pool {
         size_t size = 0;
     };
 
-    // Set arbitrarly to 64
+    // Set arbitrarily to 64
     static constexpr int          MAX_POOL_SIZE{ 64 };
     std::vector<ggml_sycl_buffer> buffer_pool = std::vector<ggml_sycl_buffer>(MAX_POOL_SIZE);
     size_t                        pool_size   = 0;