root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-14 23:09:53 +00:00
Code Issues Actions 14 Packages Projects Releases Wiki Activity

llama : reduce compile time and binary size (#9712)
Some checks are pending
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/full-cuda.Dockerfile platforms:linux/amd64 tag:full-cuda]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/full.Dockerfile platforms:linux/amd64,linux/arm64 tag:full]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-cli-cuda.Dockerfile platforms:linux/amd64 tag:light-cuda]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-cli-intel.Dockerfile platforms:linux/amd64 tag:light-intel]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-cli.Dockerfile platforms:linux/amd64,linux/arm64 tag:light]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-server-cuda.Dockerfile platforms:linux/amd64 tag:server-cuda]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-server-intel.Dockerfile platforms:linux/amd64 tag:server-intel]) (push) Waiting to run
Details
Publish Docker image / Push Docker image to Docker Hub (map[dockerfile:.devops/llama-server.Dockerfile platforms:linux/amd64,linux/arm64 tag:server]) (push) Waiting to run
Details
Nix CI / nix-eval (macos-latest) (push) Waiting to run
Details
Nix CI / nix-eval (ubuntu-latest) (push) Waiting to run
Details
Nix CI / nix-build (macos-latest) (push) Waiting to run
Details
Nix CI / nix-build (ubuntu-latest) (push) Waiting to run
Details
flake8 Lint / Lint (push) Waiting to run
Details

Browse Source 
* llama : speed up compile time

* fix build

* fix build (2)
This commit is contained in:
Xuan Son Nguyen 2024-10-02 15:49:55 +02:00 committed by GitHub
parent f536f4c439
commit a39ab216aa
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 35 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/llama.cpp
View File

@ -610,7 +610,7 @@ enum llm_tensor {
LLM_TENSOR_CLS_OUT, LLM_TENSOR_CLS_OUT,
}; };
static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NAMES = { static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_NAMES = {
{ {
LLM_ARCH_LLAMA, LLM_ARCH_LLAMA,
{ {
@ -1566,32 +1566,32 @@ struct LLM_TN {
return LLM_TENSOR_NAMES.at(arch).at(tensor); return LLM_TENSOR_NAMES.at(arch).at(tensor);
} }
std::string operator()(llm_tensor tensor, const std::string & suffix) const { std::string operator()(llm_tensor tensor, const char * suffix) const {
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) { if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
return "__missing__"; return "__missing__";
} }
return LLM_TENSOR_NAMES.at(arch).at(tensor) + "." + suffix; return std::string(LLM_TENSOR_NAMES.at(arch).at(tensor)) + "." + suffix;
} }
std::string operator()(llm_tensor tensor, int bid) const { std::string operator()(llm_tensor tensor, int bid) const {
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) { if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
return "__missing__"; return "__missing__";
} }
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid); return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid);
} }
std::string operator()(llm_tensor tensor, const std::string & suffix, int bid) const { std::string operator()(llm_tensor tensor, const char * suffix, int bid) const {
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) { if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
return "__missing__"; return "__missing__";
} }
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid) + "." + suffix; return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid) + "." + suffix;
} }
std::string operator()(llm_tensor tensor, const std::string & suffix, int bid, int xid) const { std::string operator()(llm_tensor tensor, const char * suffix, int bid, int xid) const {
if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) { if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
return "__missing__"; return "__missing__";
} }
return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor).c_str(), bid, xid) + "." + suffix; return ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid, xid) + "." + suffix;
} }
}; };
@ -4916,7 +4916,7 @@ struct llama_model_loader {
static const int TENSOR_NOT_REQUIRED = 1; static const int TENSOR_NOT_REQUIRED = 1;
static const int TENSOR_DUPLICATED = 2; static const int TENSOR_DUPLICATED = 2;
struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::vector<int64_t> & ne, int flags = 0) { struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::initializer_list<int64_t> & ne, int flags = 0) {
const struct ggml_tensor * cur = check_tensor_dims(name, ne, !(flags & TENSOR_NOT_REQUIRED)); const struct ggml_tensor * cur = check_tensor_dims(name, ne, !(flags & TENSOR_NOT_REQUIRED));
if (cur == NULL) { if (cur == NULL) {
@ -4926,7 +4926,7 @@ struct llama_model_loader {
return create_tensor_for(ctx, cur, flags & TENSOR_DUPLICATED); return create_tensor_for(ctx, cur, flags & TENSOR_DUPLICATED);
} }
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) { 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) {
const struct ggml_tensor * cur = check_tensor_dims(name, ne, required); const struct ggml_tensor * cur = check_tensor_dims(name, ne, required);
if (cur == NULL) { if (cur == NULL) {
@ -4939,7 +4939,7 @@ struct llama_model_loader {
std::array<int64_t, GGML_MAX_DIMS> dims; std::array<int64_t, GGML_MAX_DIMS> dims;
for (size_t i = 0; i < GGML_MAX_DIMS; ++i) { for (size_t i = 0; i < GGML_MAX_DIMS; ++i) {
dims[i] = i < ne.size() ? ne[i] : 1; dims[i] = i < ne.size() ? ne.begin()[i] : 1;
} }
struct ggml_tensor * tensor = ggml_view_4d(ctx, base, struct ggml_tensor * tensor = ggml_view_4d(ctx, base,

10
src/unicode-data.cpp
View File

@ -7,7 +7,7 @@
#include <unordered_map> #include <unordered_map>
#include <unordered_set> #include <unordered_set>
const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = { // start, flags // last=next_start-1 const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = { // start, flags // last=next_start-1
{0x000000, 0x0080}, {0x000000, 0x0080},
{0x000020, 0x0008}, {0x000020, 0x0008},
{0x000021, 0x0020}, {0x000021, 0x0020},
@ -2311,7 +2311,8 @@ const std::unordered_set<uint32_t> unicode_set_whitespace = {
0x003000, 0x003000,
}; };
const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = { // list is always in ascending order, to enable binary searh
const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase = {
{0x000041, 0x000061}, {0x000041, 0x000061},
{0x000042, 0x000062}, {0x000042, 0x000062},
{0x000043, 0x000063}, {0x000043, 0x000063},
@ -3747,7 +3748,8 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = {
{0x01E921, 0x01E943}, {0x01E921, 0x01E943},
}; };
const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = { // list is always in ascending order, to enable binary searh
const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase = {
{0x000061, 0x000041}, {0x000061, 0x000041},
{0x000062, 0x000042}, {0x000062, 0x000042},
{0x000063, 0x000043}, {0x000063, 0x000043},
@ -5200,7 +5202,7 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = {
{0x01E943, 0x01E921}, {0x01E943, 0x01E921},
}; };
const std::vector<range_nfd> unicode_ranges_nfd = { // start, last, nfd const std::initializer_list<range_nfd> unicode_ranges_nfd = { // start, last, nfd
{0x000000, 0x000000, 0x000000}, {0x000000, 0x000000, 0x000000},
{0x0000C0, 0x0000C5, 0x000041}, {0x0000C0, 0x0000C5, 0x000041},
{0x0000C7, 0x0000C7, 0x000043}, {0x0000C7, 0x0000C7, 0x000043},

8
src/unicode-data.h
View File

@ -13,8 +13,8 @@ struct range_nfd {
static const uint32_t MAX_CODEPOINTS = 0x110000; static const uint32_t MAX_CODEPOINTS = 0x110000;
extern const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags; extern const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
extern const std::unordered_set<uint32_t> unicode_set_whitespace; extern const std::unordered_set<uint32_t> unicode_set_whitespace;
extern const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase; extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase;
extern const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase; extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase;
extern const std::vector<range_nfd> unicode_ranges_nfd; extern const std::initializer_list<range_nfd> unicode_ranges_nfd;

21
src/unicode.cpp
View File

@ -123,11 +123,11 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
static std::vector<codepoint_flags> unicode_cpt_flags_array() { static std::vector<codepoint_flags> unicode_cpt_flags_array() {
std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED); std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED);
assert (unicode_ranges_flags.front().first == 0); assert (unicode_ranges_flags.begin()[0].first == 0);
assert (unicode_ranges_flags.back().first == MAX_CODEPOINTS); assert (unicode_ranges_flags.begin()[unicode_ranges_flags.size()-1].first == MAX_CODEPOINTS);
for (size_t i = 1; i < unicode_ranges_flags.size(); ++i) { for (size_t i = 1; i < unicode_ranges_flags.size(); ++i) {
const auto range_ini = unicode_ranges_flags[i-1]; // codepoint_ini, flags const auto range_ini = unicode_ranges_flags.begin()[i-1]; // codepoint_ini, flags
const auto range_end = unicode_ranges_flags[i]; // codepoint_end, flags const auto range_end = unicode_ranges_flags.begin()[i]; // codepoint_end, flags
for (uint32_t cpt = range_ini.first; cpt < range_end.first; ++cpt) { for (uint32_t cpt = range_ini.first; cpt < range_end.first; ++cpt) {
cpt_flags[cpt] = range_ini.second; cpt_flags[cpt] = range_ini.second;
} }
@ -597,7 +597,7 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
std::vector<uint32_t> result(cpts.size()); std::vector<uint32_t> result(cpts.size());
for (size_t i = 0; i < cpts.size(); ++i) { for (size_t i = 0; i < cpts.size(); ++i) {
const uint32_t cpt = cpts[i]; const uint32_t cpt = cpts[i];
auto it = std::upper_bound(unicode_ranges_nfd.cbegin(), unicode_ranges_nfd.cend(), cpt, comp) - 1; auto it = std::upper_bound(unicode_ranges_nfd.begin(), unicode_ranges_nfd.end(), cpt, comp) - 1;
result[i] = (it->first <= cpt && cpt <= it->last) ? it->nfd : cpt; result[i] = (it->first <= cpt && cpt <= it->last) ? it->nfd : cpt;
} }
return result; return result;
@ -639,8 +639,15 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
} }
uint32_t unicode_tolower(uint32_t cp) { uint32_t unicode_tolower(uint32_t cp) {
auto it = unicode_map_lowercase.find(cp); // binary search
return it == unicode_map_lowercase.end() ? cp : it->second; auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cp,
[](const std::pair<uint32_t, uint32_t> & pair, uint32_t value) {
return pair.first < value;
});
if (it != unicode_map_lowercase.end() && it->first == cp) {
return it->second;
}
return cp; // Return the original code point if no lowercase mapping is found
} }
std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) { std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {