Skip to content

Commit a39ab21

Browse files
ngxsonngxson
authored
llama : reduce compile time and binary size (#9712)
* llama : speed up compile time * fix build * fix build (2)
1 parent f536f4c commit a39ab21

File tree

4 files changed

+35
-26
lines changed

4 files changed

+35
-26
lines changed

‎src/llama.cpp

+11-11
Original file line numberDiff line numberDiff line change
@@ -610,7 +610,7 @@ enum llm_tensor {
610610
LLM_TENSOR_CLS_OUT,
611611
};
612612

613-
static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NAMES = {
613+
static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_NAMES = {
614614
{
615615
LLM_ARCH_LLAMA,
616616
{
@@ -1566,32 +1566,32 @@ struct LLM_TN {
15661566
return LLM_TENSOR_NAMES.at(arch).at(tensor);
15671567
}
15681568

1569-
std::string operator()(llm_tensor tensor, const std::string & suffix) const {
1569+
std::string operator()(llm_tensor tensor, const char * suffix) const {
15701570
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
15711571
return "__missing__";
15721572
}
1573-
return LLM_TENSOR_NAMES.at(arch).at(tensor) + "." + suffix;
1573+
return std::string(LLM_TENSOR_NAMES.at(arch).at(tensor)) + "." + suffix;
15741574
}
15751575

15761576
std::string operator()(llm_tensor tensor, int bid) const {
15771577
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
15781578
return "__missing__";
15791579
}
1580-
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid);
1580+
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid);
15811581
}
15821582

1583-
std::string operator()(llm_tensor tensor, const std::string & suffix, int bid) const {
1583+
std::string operator()(llm_tensor tensor, const char * suffix, int bid) const {
15841584
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
15851585
return "__missing__";
15861586
}
1587-
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid) + "." + suffix;
1587+
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid) + "." + suffix;
15881588
}
15891589

1590-
std::string operator()(llm_tensor tensor, const std::string & suffix, int bid, int xid) const {
1590+
std::string operator()(llm_tensor tensor, const char * suffix, int bid, int xid) const {
15911591
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
15921592
return "__missing__";
15931593
}
1594-
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid, xid) + "." + suffix;
1594+
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid, xid) + "." + suffix;
15951595
}
15961596
};
15971597

@@ -4916,7 +4916,7 @@ struct llama_model_loader {
49164916
static const int TENSOR_NOT_REQUIRED = 1;
49174917
static const int TENSOR_DUPLICATED = 2;
49184918

4919-
struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::vector<int64_t> & ne, int flags = 0) {
4919+
struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::initializer_list<int64_t> & ne, int flags = 0) {
49204920
const struct ggml_tensor * cur = check_tensor_dims(name, ne, !(flags & TENSOR_NOT_REQUIRED));
49214921

49224922
if (cur == NULL) {
@@ -4926,7 +4926,7 @@ struct llama_model_loader {
49264926
return create_tensor_for(ctx, cur, flags & TENSOR_DUPLICATED);
49274927
}
49284928

4929-
struct ggml_tensor * create_tensor_as_view(struct ggml_context * ctx, struct ggml_tensor * base, const std::string & name, const std::vector<int64_t> & ne, size_t offset, bool required = true) {
4929+
struct ggml_tensor * create_tensor_as_view(struct ggml_context * ctx, struct ggml_tensor * base, const std::string & name, const std::initializer_list<int64_t> & ne, size_t offset, bool required = true) {
49304930
const struct ggml_tensor * cur = check_tensor_dims(name, ne, required);
49314931

49324932
if (cur == NULL) {
@@ -4939,7 +4939,7 @@ struct llama_model_loader {
49394939

49404940
std::array<int64_t, GGML_MAX_DIMS> dims;
49414941
for (size_t i = 0; i < GGML_MAX_DIMS; ++i) {
4942-
dims[i] = i < ne.size() ? ne[i] : 1;
4942+
dims[i] = i < ne.size() ? ne.begin()[i] : 1;
49434943
}
49444944

49454945
struct ggml_tensor * tensor = ggml_view_4d(ctx, base,

‎src/unicode-data.cpp

+6-4
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
#include <unordered_map>
88
#include <unordered_set>
99

10-
const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = { // start, flags // last=next_start-1
10+
const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = { // start, flags // last=next_start-1
1111
{0x000000, 0x0080},
1212
{0x000020, 0x0008},
1313
{0x000021, 0x0020},
@@ -2311,7 +2311,8 @@ const std::unordered_set<uint32_t> unicode_set_whitespace = {
23112311
0x003000,
23122312
};
23132313

2314-
const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = {
2314+
// list is always in ascending order, to enable binary searh
2315+
const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase = {
23152316
{0x000041, 0x000061},
23162317
{0x000042, 0x000062},
23172318
{0x000043, 0x000063},
@@ -3747,7 +3748,8 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = {
37473748
{0x01E921, 0x01E943},
37483749
};
37493750

3750-
const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = {
3751+
// list is always in ascending order, to enable binary searh
3752+
const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase = {
37513753
{0x000061, 0x000041},
37523754
{0x000062, 0x000042},
37533755
{0x000063, 0x000043},
@@ -5200,7 +5202,7 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = {
52005202
{0x01E943, 0x01E921},
52015203
};
52025204

5203-
const std::vector<range_nfd> unicode_ranges_nfd = { // start, last, nfd
5205+
const std::initializer_list<range_nfd> unicode_ranges_nfd = { // start, last, nfd
52045206
{0x000000, 0x000000, 0x000000},
52055207
{0x0000C0, 0x0000C5, 0x000041},
52065208
{0x0000C7, 0x0000C7, 0x000043},

‎src/unicode-data.h

+4-4
Original file line numberDiff line numberDiff line change
@@ -13,8 +13,8 @@ struct range_nfd {
1313

1414
static const uint32_t MAX_CODEPOINTS = 0x110000;
1515

16-
extern const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
16+
extern const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
1717
extern const std::unordered_set<uint32_t> unicode_set_whitespace;
18-
extern const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase;
19-
extern const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase;
20-
extern const std::vector<range_nfd> unicode_ranges_nfd;
18+
extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase;
19+
extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase;
20+
extern const std::initializer_list<range_nfd> unicode_ranges_nfd;

‎src/unicode.cpp

+14-7
Original file line numberDiff line numberDiff line change
@@ -123,11 +123,11 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
123123
static std::vector<codepoint_flags> unicode_cpt_flags_array() {
124124
std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED);
125125

126-
assert (unicode_ranges_flags.front().first == 0);
127-
assert (unicode_ranges_flags.back().first == MAX_CODEPOINTS);
126+
assert (unicode_ranges_flags.begin()[0].first == 0);
127+
assert (unicode_ranges_flags.begin()[unicode_ranges_flags.size()-1].first == MAX_CODEPOINTS);
128128
for (size_t i = 1; i < unicode_ranges_flags.size(); ++i) {
129-
const auto range_ini = unicode_ranges_flags[i-1]; // codepoint_ini, flags
130-
const auto range_end = unicode_ranges_flags[i]; // codepoint_end, flags
129+
const auto range_ini = unicode_ranges_flags.begin()[i-1]; // codepoint_ini, flags
130+
const auto range_end = unicode_ranges_flags.begin()[i]; // codepoint_end, flags
131131
for (uint32_t cpt = range_ini.first; cpt < range_end.first; ++cpt) {
132132
cpt_flags[cpt] = range_ini.second;
133133
}
@@ -597,7 +597,7 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
597597
std::vector<uint32_t> result(cpts.size());
598598
for (size_t i = 0; i < cpts.size(); ++i) {
599599
const uint32_t cpt = cpts[i];
600-
auto it = std::upper_bound(unicode_ranges_nfd.cbegin(), unicode_ranges_nfd.cend(), cpt, comp) - 1;
600+
auto it = std::upper_bound(unicode_ranges_nfd.begin(), unicode_ranges_nfd.end(), cpt, comp) - 1;
601601
result[i] = (it->first <= cpt && cpt <= it->last) ? it->nfd : cpt;
602602
}
603603
return result;
@@ -639,8 +639,15 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
639639
}
640640

641641
uint32_t unicode_tolower(uint32_t cp) {
642-
auto it = unicode_map_lowercase.find(cp);
643-
return it == unicode_map_lowercase.end() ? cp : it->second;
642+
// binary search
643+
auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cp,
644+
[](const std::pair<uint32_t, uint32_t> & pair, uint32_t value) {
645+
return pair.first < value;
646+
});
647+
if (it != unicode_map_lowercase.end() && it->first == cp) {
648+
return it->second;
649+
}
650+
return cp; // Return the original code point if no lowercase mapping is found
644651
}
645652

646653
std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {

0 commit comments

Comments
 (0)