nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix clang warnings on macOS (#1354) 

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2026-03-03 16:27:16 +01:00 committed by GitHub
parent ea3e8e30e1
commit fd16a418de
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
10 changed files with 38 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
common/common.cpp
View File

@ -2513,7 +2513,7 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
"number of tokens to draft for speculative decoding (default: %d)", params.speculative.n_max });
options.push_back({ "*", "--draft-min, --draft-n-min N", "minimum number of draft tokens to use for speculative decoding" });
options.push_back({ "*", "--draft-p-min P", "minimum speculative decoding probability (greedy) (default: %.1f)", (double)params.speculative.p_min });
options.push_back({ "*", "--spec-type Name", "[none | ngram - cache | ngram - simple | ngram - map - k | ngram - map - k4v | ngram - mod]", "type of speculative decoding to use when no draft model is provided (default: %s)\n", (double)params.speculative.type});
options.push_back({ "*", "--spec-type Name [none | ngram - cache | ngram - simple | ngram - map - k | ngram - map - k4v | ngram - mod]", "type of speculative decoding to use when no draft model is provided (default: %d)\n", (int)params.speculative.type});
options.push_back({ "*", "--spec-ngram-size-n N", "ngram size N for ngram-simple/ngram-map speculative decoding, length of lookup n-gram (default: %d)\n",params.speculative.ngram_size_n });
options.push_back({ "*", "--spec-ngram-size-m N", "ngram size M for ngram-simple/ngram-map speculative decoding, length of draft m-gram (default: %d)\n", params.speculative.ngram_size_m });

24
examples/server/CMakeLists.txt
View File

@ -36,15 +36,15 @@ foreach(asset ${PUBLIC_ASSETS})
OUTPUT "${output}"
COMMAND "${CMAKE_COMMAND}" "-DINPUT=${input}" "-DOUTPUT=${output}" -P "${PROJECT_SOURCE_DIR}/scripts/xxd.cmake"
)
message("TARGET_SRCS contains: ${input}")
set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
message("TARGET_SRCS contains: ${input}")
set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
endforeach()
# include new llamacpp webui
set(ALT_PUBLIC_ASSETS
index_llamacpp.html.gz
loading.html
loading.html
)
foreach(asset ${ALT_PUBLIC_ASSETS})
@ -56,9 +56,9 @@ foreach(asset ${ALT_PUBLIC_ASSETS})
OUTPUT "${output}"
COMMAND "${CMAKE_COMMAND}" "-DINPUT=${input}" "-DOUTPUT=${output}" -P "${PROJECT_SOURCE_DIR}/scripts/xxd.cmake"
)
message("TARGET_SRCS contains: ${input}")
set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
message("TARGET_SRCS contains: ${input}")
set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
endforeach()
@ -68,14 +68,14 @@ target_compile_definitions(${TARGET} PRIVATE
SERVER_VERBOSE=$<BOOL:${LLAMA_SERVER_VERBOSE}>
)
if (MSVC)
target_link_options(${TARGET} PRIVATE
$<$<CONFIG:DEBUG>:/STACK:20971520,1048576 >
$<$<CONFIG:RELEASE>:/STACK:20971520,1048576>
)
target_link_options(${TARGET} PRIVATE
$<$<CONFIG:DEBUG>:/STACK:20971520,1048576 >
$<$<CONFIG:RELEASE>:/STACK:20971520,1048576>
)
endif()
# target_link_libraries(${TARGET} PRIVATE "/STACK:104857600")
target_include_directories(${TARGET} PRIVATE ${CMAKE_SOURCE_DIR})
target_link_libraries(${TARGET} PRIVATE common ${CMAKE_THREAD_LIBS_INIT})
#target_link_libraries(${TARGET} PRIVATE common ${CMAKE_THREAD_LIBS_INIT})
target_include_directories(${TARGET} PRIVATE ../mtmd)
target_link_libraries(${TARGET} PRIVATE common mtmd cpp-httplib ${CMAKE_THREAD_LIBS_INIT})

4
examples/server/server-context.cpp
View File

@ -1534,7 +1534,7 @@ void server_context::populate_token_probs(const server_slot& slot, completion_to
for (size_t i = 0; i < std::min(max_probs, n_probs); i++) {
result.probs.push_back({
cur_p->data[i].id,
common_token_to_piece(ctx, {cur_p->data[i].id}, special),
common_token_to_piece(ctx, cur_p->data[i].id, special),
cur_p->data[i].p
});
}
@ -1550,7 +1550,7 @@ void server_context::populate_token_probs(const server_slot& slot, completion_to
for (size_t i = 0; i < std::min(n_vocab, n_probs); i++) {
result.probs.push_back({
cur[i].id,
common_token_to_piece(ctx, {cur[i].id}, special),
common_token_to_piece(ctx, cur[i].id, special),
cur[i].p
});
}

2
ggml/include/ggml.h
View File

@ -773,7 +773,7 @@ extern "C" {
struct ggml_tensor {
enum ggml_type type;
GGML_DEPRECATED(enum ggml_backend_type backend, "use the buffer type to find the storage location of the tensor");
enum ggml_backend_type __backend;
struct ggml_backend_buffer * buffer;

2
ggml/src/ggml-metal.m
View File

@ -4158,7 +4158,7 @@ GGML_CALL static void * ggml_backend_metal_buffer_get_base(ggml_backend_buffer_t
return ctx->all_data;
}
GGML_CALL void ggml_backend_metal_buffer_memset_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
GGML_CALL static void ggml_backend_metal_buffer_memset_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
memset((char *)tensor->data + offset, value, size);
GGML_UNUSED(buffer);

2
ggml/src/ggml.c
View File

@ -24763,7 +24763,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
case GGML_OP_ADD_ID:
{
GGML_ABORT("fatal error"); // TODO: implement
} break;
}
case GGML_OP_ADD1:
{
if (src0->grad) {

19
ggml/src/iqk/iqk_quantize.cpp
View File

@ -8261,13 +8261,14 @@ float QuantizerIQKT<block_size, group_size, num_bits, is_abs, is_int>::find_best
}
template <int block_size, int group_size, int num_bits, bool is_abs, bool is_int>
void QuantizerIQKT<block_size, group_size, num_bits, is_abs, is_int>::find_best_match(float d, const float * xb, const float * weight, int * best_idx) const {
void QuantizerIQKT<block_size, group_size, num_bits, is_abs, is_int>::find_best_match(float d,
[[maybe_unused]] const float * xb, [[maybe_unused]] const float * weight, int * best_idx) const {
if (!d) {
std::memset(best_idx, 0, kNg*sizeof(int));
return;
}
int ncluster = m_clusters.size()/kGroupSize;
float id = 1/d;
[[maybe_unused]] int ncluster = m_clusters.size()/kGroupSize;
[[maybe_unused]] float id = 1/d;
#ifdef __AVX2__
if constexpr (kGroupSize == 8) {
__m256 sqx[8];
@ -9802,8 +9803,8 @@ bool check_tensor_for_blocks_256_fp16(const ggml_tensor * tensor) {
nbad += check_row_for_blocks_256_fp16(nblock, x);
}
if (nbad > 0) {
fprintf(stderr, "%s: found %d NaN block scales out of %ld blocks in tensor %s\n", __func__,
nbad, ggml_nrows(tensor)*nblock, tensor->name);
fprintf(stderr, "%s: found %d NaN block scales out of %g blocks in tensor %s\n", __func__,
nbad, 1.*ggml_nrows(tensor)*nblock, tensor->name);
if (tensor->ne[2] > 1) {
int nb = tensor->ne[0]/QK_K;
for (int64_t i02 = 0; i02 < tensor->ne[2]; ++i02) {
@ -9813,7 +9814,7 @@ bool check_tensor_for_blocks_256_fp16(const ggml_tensor * tensor) {
auto xr = (const Block *)(xex + i01*tensor->nb[1]);
nbad_expert += check_row_for_blocks_256_fp16(nb, xr);
}
if (nbad_expert > 0) fprintf(stderr," there are %d NaN block scales for expert %ld\n", nbad_expert, i02);
if (nbad_expert > 0) fprintf(stderr," there are %d NaN block scales for expert %g\n", nbad_expert, 1.*i02);
}
}
return false;
@ -9839,8 +9840,8 @@ bool check_tensor_for_blocks_256_fp16_repacked(const ggml_tensor * tensor) {
nbad += check_row_for_blocks_256_fp16(nblock, x, nr);
}
if (nbad > 0) {
fprintf(stderr, "%s: found %d NaN block scales out of %ld blocks in tensor %s\n", __func__,
nbad, ggml_nrows(tensor)*nblock, tensor->name);
fprintf(stderr, "%s: found %d NaN block scales out of %g blocks in tensor %s\n", __func__,
nbad, 1.*ggml_nrows(tensor)*nblock, tensor->name);
if (tensor->ne[2] > 1) {
int nb = tensor->ne[0]/QK_K;
for (int64_t i02 = 0; i02 < tensor->ne[2]; ++i02) {
@ -9850,7 +9851,7 @@ bool check_tensor_for_blocks_256_fp16_repacked(const ggml_tensor * tensor) {
auto xr = (const Block *)(xex + i01*tensor->nb[1]);
nbad_expert += check_row_for_blocks_256_fp16(nb, xr, nr);
}
if (nbad_expert > 0) fprintf(stderr," there are %d NaN block scales for expert %ld\n", nbad_expert, i02);
if (nbad_expert > 0) fprintf(stderr," there are %d NaN block scales for expert %g\n", nbad_expert, 1.*i02);
}
}
return false;

3
include/llama.h
View File

@ -575,6 +575,7 @@ extern "C" {
LLAMA_API enum llama_pooling_type llama_pooling_type(const struct llama_context * ctx);
struct llama_vocab;
LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_vocab * vocab);
LLAMA_API enum llama_rope_type llama_rope_type (const struct llama_model * model);
@ -586,7 +587,7 @@ extern "C" {
LLAMA_API int32_t llama_n_ctx_train(const struct llama_model * model);
LLAMA_API int32_t llama_model_n_embd (const struct llama_model * model);
LLAMA_API int32_t llama_model_n_embd_inp(const struct llama_model* model);
LLAMA_API int32_t llama_n_layer (const struct llama_model * model);
// Compat

2
src/llama-mmap.cpp
View File

@ -275,7 +275,7 @@ struct llama_mmap::impl {
#ifdef _POSIX_MAPPED_FILES
std::vector<std::pair<size_t, size_t>> mapped_fragments;
impl(struct llama_file * file, size_t prefetch, bool numa, bool use_thp) {
impl(struct llama_file * file, size_t prefetch, bool numa, [[maybe_unused]] bool use_thp) {
size = file->size();
int fd = file->file_id();
int flags = MAP_SHARED;

14
src/llama.cpp
View File

@ -1708,8 +1708,8 @@ static void llm_prepare_mla(llama_model & model, int mla) {
l.wk_b = l.computed_wk_b.get();
model.tensors_by_name.push_back(std::make_pair(name, l.wk_b));
printf("Computed %s as %ld x %ld x %ld and stored in buffer %s\n", name.c_str(), wk_b->ne[0], wk_b->ne[1], wk_b->ne[2],
ggml_backend_buffer_name(l.computed_wk_b->buffer));
printf("Computed %s as %d x %d x %d of type %s and stored in buffer %s\n", name.c_str(), (int)wk_b->ne[0], (int)wk_b->ne[1], (int)wk_b->ne[2],
ggml_type_name(wk_b->type), ggml_backend_buffer_name(l.computed_wk_b->buffer));
ggml_graph_clear(graph);
auto wv_b = ggml_cont(ctx, ggml_view_3d(ctx, &wkv_b, kv_lora_rank, n_embd_head_v, n_head,
@ -1740,8 +1740,8 @@ static void llm_prepare_mla(llama_model & model, int mla) {
l.wv_b = l.computed_wv_b.get();
model.tensors_by_name.push_back(std::make_pair(name, l.wv_b));
printf("Computed %s as %ld x %ld x %ld and stored in buffer %s\n", name.c_str(), wv_b->ne[0], wv_b->ne[1], wv_b->ne[2],
ggml_backend_buffer_name(l.computed_wv_b->buffer));
printf("Computed %s as %d x %d x %d of type %s and stored in buffer %s\n", name.c_str(), (int)wv_b->ne[0], (int)wv_b->ne[1], (int)wv_b->ne[2],
ggml_type_name(wv_b->type), ggml_backend_buffer_name(l.computed_wv_b->buffer));
ggml_graph_clear(graph);
}
@ -1875,8 +1875,8 @@ static void llm_prepare_mla(llama_model & model, int mla) {
l.wkv_b = l.computed_wkv_b.get();
model.tensors_by_name.push_back(std::make_pair(name, l.wkv_b));
printf("Computed %s as %ld x %ld and stored in buffer %s\n", name.c_str(), wkv_b->ne[0], wkv_b->ne[1],
ggml_backend_buffer_name(l.computed_wkv_b->buffer));
printf("Computed %s as %d x %d of type %s and stored in buffer %s\n", name.c_str(), (int)wkv_b->ne[0], (int)wkv_b->ne[1],
ggml_type_name(wkv_b->type), ggml_backend_buffer_name(l.computed_wkv_b->buffer));
ggml_graph_clear(graph);
}
@ -6349,7 +6349,7 @@ struct llama_data_read {
}
continue;
}
const uint64_t k_size_row = (ctx->cparams.mla_attn == 0) ? ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa) : ggml_row_size(kv_self.k_l[il]->type, kv_lora_rank + n_embd_head_qk_rope);
const size_t k_size_row = (ctx->cparams.mla_attn == 0) ? ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa) : ggml_row_size(kv_self.k_l[il]->type, kv_lora_rank + n_embd_head_qk_rope);
if (k_size_row != k_size_row_ref) {
LLAMA_LOG_ERROR("%s: mismatched key row size (%zu != %zu, layer %d)\n", __func__, k_size_row, (size_t) k_size_row_ref, il);
return false;