Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-20 01:31:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.devops/main-intel.Dockerfile
#	.devops/main-vulkan.Dockerfile
#	.devops/server-intel.Dockerfile
#	.devops/server-vulkan.Dockerfile
#	.github/workflows/bench.yml
#	.github/workflows/build.yml
#	.github/workflows/python-lint.yml
#	.github/workflows/server.yml
#	.gitignore
#	Makefile
#	README-sycl.md
#	README.md
#	ci/run.sh
#	flake.lock
#	llama.cpp
#	models/ggml-vocab-falcon.gguf
#	models/ggml-vocab-llama-spm.gguf
#	models/ggml-vocab-mpt.gguf
#	models/ggml-vocab-stablelm.gguf
#	models/ggml-vocab-starcoder.gguf
#	requirements.txt
#	scripts/check-requirements.sh
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-grammar-integration.cpp
#	tests/test-tokenizer-0-bpe.py
#	tests/test-tokenizer-0-spm.py
#	tests/test-tokenizer-1-spm.cpp
This commit is contained in:
Concedo
2024-04-30 21:04:17 +08:00
parent 4025efb497 a68a1e7ed0
commit 17a24d753c
52 changed files with 4978 additions and 1249 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/batched-bench/batched-bench.cpp
+17 -11
View File
@@ -32,7 +32,7 @@ int main(int argc, char ** argv) {
gpt_params params;
if (argc == 1 || argv[1][0] == '-') {
printf("usage: %s MODEL_PATH [N_KV_MAX] [N_BATCH] [N_UBATCH] [IS_PP_SHARED] [NGL] <PP> <TG> <PL>\n" , argv[0]);
printf("usage: %s MODEL_PATH [N_KV_MAX] [N_BATCH] [N_UBATCH] [FATTN] [IS_PP_SHARED] [NGL] <PP> <TG> <PL>\n" , argv[0]);
printf(" <PP>, <TG> and PL are comma-separated lists of numbers without spaces\n\n");
printf(" example: %s ggml-model-f16.gguf 2048 2048 512 0 999 128,256,512 128,256 1,2,4,8,16,32\n\n", argv[0]);
return 1 ;
@@ -41,6 +41,7 @@ int main(int argc, char ** argv) {
int n_kv_max = 2048;
int n_batch = 2048;
int n_ubatch = 512;
bool flash_attn = false;
int is_pp_shared = 0;
int n_gpu_layers = 0;
@@ -66,23 +67,27 @@ int main(int argc, char ** argv) {
}
if (argc >= 6) {
is_pp_shared = std::atoi(argv[5]);
flash_attn = std::atoi(argv[5]);
}
if (argc >= 7) {
n_gpu_layers = std::atoi(argv[6]);
is_pp_shared = std::atoi(argv[6]);
}
if (argc >= 8) {
n_pp = parse_list(argv[7]);
n_gpu_layers = std::atoi(argv[7]);
}
if (argc >= 9) {
n_tg = parse_list(argv[8]);
n_pp = parse_list(argv[8]);
}
if (argc >= 10) {
n_pl = parse_list(argv[9]);
n_tg = parse_list(argv[9]);
}
if (argc >= 11) {
n_pl = parse_list(argv[10]);
}
// init LLM
@@ -108,10 +113,11 @@ int main(int argc, char ** argv) {
llama_context_params ctx_params = llama_context_default_params();
ctx_params.seed = 1234;
ctx_params.n_ctx = n_kv_max;
ctx_params.n_batch = n_batch;
ctx_params.n_ubatch = n_ubatch;
ctx_params.seed = 1234;
ctx_params.n_ctx = n_kv_max;
ctx_params.n_batch = n_batch;
ctx_params.n_ubatch = n_ubatch;
ctx_params.flash_attn = flash_attn;
ctx_params.n_threads = params.n_threads;
ctx_params.n_threads_batch = params.n_threads_batch == -1 ? params.n_threads : params.n_threads_batch;
@@ -169,7 +175,7 @@ int main(int argc, char ** argv) {
}
LOG_TEE("\n");
LOG_TEE("%s: n_kv_max = %d, n_batch = %d, n_ubatch = %d, is_pp_shared = %d, n_gpu_layers = %d, n_threads = %u, n_threads_batch = %u\n", __func__, n_kv_max, n_batch, n_ubatch, is_pp_shared, n_gpu_layers, ctx_params.n_threads, ctx_params.n_threads_batch);
LOG_TEE("%s: n_kv_max = %d, n_batch = %d, n_ubatch = %d, flash_attn = %d, is_pp_shared = %d, n_gpu_layers = %d, n_threads = %u, n_threads_batch = %u\n", __func__, n_kv_max, n_batch, n_ubatch, flash_attn, is_pp_shared, n_gpu_layers, ctx_params.n_threads, ctx_params.n_threads_batch);
LOG_TEE("\n");
LOG_TEE("|%6s | %6s | %4s | %6s | %8s | %8s | %8s | %8s | %8s | %8s |\n", "PP", "TG", "B", "N_KV", "T_PP s", "S_PP t/s", "T_TG s", "S_TG t/s", "T s", "S t/s");
examples/imatrix/imatrix.cpp
+44 -33
View File
@@ -24,6 +24,7 @@ struct Stats {
};
struct StatParams {
std::string dataset;
std::string ofile = "imatrix.dat";
int n_output_frequency = 10;
int verbosity = 1;
@@ -47,7 +48,7 @@ private:
std::vector<float> m_src1_data;
std::vector<char> m_ids; // the expert ids from ggml_mul_mat_id
//
void save_imatrix(const char * file_name) const;
void save_imatrix(const char * file_name, const char * dataset) const;
void keep_imatrix(int ncall) const;
};
@@ -200,7 +201,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
}
void IMatrixCollector::save_imatrix() const {
save_imatrix(m_params.ofile.empty() ? "imatrix.dat" : m_params.ofile.c_str());
save_imatrix(m_params.ofile.empty() ? "imatrix.dat" : m_params.ofile.c_str(), m_params.dataset.c_str());
}
void IMatrixCollector::keep_imatrix(int ncall) const {
@@ -208,24 +209,33 @@ void IMatrixCollector::keep_imatrix(int ncall) const {
if (file_name.empty()) file_name = "imatrix.dat";
file_name += ".at_";
file_name += std::to_string(ncall);
save_imatrix(file_name.c_str());
save_imatrix(file_name.c_str(), m_params.dataset.c_str());
}
void IMatrixCollector::save_imatrix(const char * fname) const {
void IMatrixCollector::save_imatrix(const char * fname, const char * dataset) const {
std::ofstream out(fname, std::ios::binary);
int n_entries = m_stats.size();
out.write((const char*)&n_entries, sizeof(n_entries));
for (auto& p : m_stats) {
out.write((const char *) &n_entries, sizeof(n_entries));
for (const auto & p : m_stats) {
int len = p.first.size();
out.write((const char*)&len, sizeof(len));
out.write((const char *) &len, sizeof(len));
out.write(p.first.c_str(), len);
out.write((const char*)&p.second.ncall, sizeof(p.second.ncall));
out.write((const char *) &p.second.ncall, sizeof(p.second.ncall));
int nval = p.second.values.size();
out.write((const char*)&nval, sizeof(nval));
if (nval > 0) out.write((const char*)p.second.values.data(), nval*sizeof(float));
out.write((const char *) &nval, sizeof(nval));
if (nval > 0) out.write((const char *) p.second.values.data(), nval * sizeof(float));
}
// Write the number of call the matrix was computed with
out.write((const char *) &m_last_call, sizeof(m_last_call));
// Write the dataset name at the end of the file to later on specify it in quantize
int n_dataset = strlen(dataset);
out.write((const char *) &n_dataset, sizeof(n_dataset));
out.write(dataset, n_dataset);
if (m_params.verbosity > 0) {
fprintf(stderr, "\n%s: stored collected data after %d chunks in %s\n",__func__,m_last_call,fname);
fprintf(stderr, "\n%s: stored collected data after %d chunks in %s\n", __func__, m_last_call, fname);
}
}
@@ -548,6 +558,29 @@ int main(int argc, char ** argv) {
}
}
gpt_params params;
params.n_batch = 512;
if (!gpt_params_parse(args.size(), args.data(), params)) {
return 1;
}
params.logits_all = true;
params.n_batch = std::min(params.n_batch, params.n_ctx);
print_build_info();
if (params.seed == LLAMA_DEFAULT_SEED) {
params.seed = time(NULL);
}
fprintf(stderr, "%s: seed = %u\n", __func__, params.seed);
std::mt19937 rng(params.seed);
if (params.random_prompt) {
params.prompt = gpt_random_prompt(rng);
}
sparams.dataset = params.prompt_file;
g_collector.set_parameters(std::move(sparams));
if (!combine_files.empty()) {
@@ -586,28 +619,6 @@ int main(int argc, char ** argv) {
}
}
gpt_params params;
params.n_batch = 512;
if (!gpt_params_parse(args.size(), args.data(), params)) {
return 1;
}
params.logits_all = true;
params.n_batch = std::min(params.n_batch, params.n_ctx);
print_build_info();
if (params.seed == LLAMA_DEFAULT_SEED) {
params.seed = time(NULL);
}
fprintf(stderr, "%s: seed = %u\n", __func__, params.seed);
std::mt19937 rng(params.seed);
if (params.random_prompt) {
params.prompt = gpt_random_prompt(rng);
}
llama_backend_init();
llama_numa_init(params.numa);
examples/llama-bench/llama-bench.cpp
+27 -3
View File
@@ -175,6 +175,7 @@ struct cmd_params {
std::vector<llama_split_mode> split_mode;
std::vector<int> main_gpu;
std::vector<bool> no_kv_offload;
std::vector<bool> flash_attn;
std::vector<std::vector<float>> tensor_split;
std::vector<bool> use_mmap;
std::vector<bool> embeddings;
@@ -196,6 +197,7 @@ static const cmd_params cmd_params_defaults = {
/* split_mode */ {LLAMA_SPLIT_MODE_LAYER},
/* main_gpu */ {0},
/* no_kv_offload */ {false},
/* flash_attn */ {false},
/* tensor_split */ {std::vector<float>(llama_max_devices(), 0.0f)},
/* use_mmap */ {true},
/* embeddings */ {false},
@@ -221,6 +223,7 @@ static void print_usage(int /* argc */, char ** argv) {
printf(" -sm, --split-mode <none|layer|row> (default: %s)\n", join(transform_to_str(cmd_params_defaults.split_mode, split_mode_str), ",").c_str());
printf(" -mg, --main-gpu <i> (default: %s)\n", join(cmd_params_defaults.main_gpu, ",").c_str());
printf(" -nkvo, --no-kv-offload <0|1> (default: %s)\n", join(cmd_params_defaults.no_kv_offload, ",").c_str());
printf(" -fa, --flash-attn <0|1> (default: %s)\n", join(cmd_params_defaults.flash_attn, ",").c_str());
printf(" -mmp, --mmap <0|1> (default: %s)\n", join(cmd_params_defaults.use_mmap, ",").c_str());
printf(" -embd, --embeddings <0|1> (default: %s)\n", join(cmd_params_defaults.embeddings, ",").c_str());
printf(" -ts, --tensor-split <ts0/ts1/..> (default: 0)\n");
@@ -394,6 +397,13 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
}
auto p = split<bool>(argv[i], split_delim);
params.no_kv_offload.insert(params.no_kv_offload.end(), p.begin(), p.end());
} else if (arg == "-fa" || arg == "--flash-attn") {
if (++i >= argc) {
invalid_param = true;
break;
}
auto p = split<bool>(argv[i], split_delim);
params.flash_attn.insert(params.flash_attn.end(), p.begin(), p.end());
} else if (arg == "-mmp" || arg == "--mmap") {
if (++i >= argc) {
invalid_param = true;
@@ -478,6 +488,7 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
if (params.split_mode.empty()) { params.split_mode = cmd_params_defaults.split_mode; }
if (params.main_gpu.empty()) { params.main_gpu = cmd_params_defaults.main_gpu; }
if (params.no_kv_offload.empty()){ params.no_kv_offload = cmd_params_defaults.no_kv_offload; }
if (params.flash_attn.empty()) { params.flash_attn = cmd_params_defaults.flash_attn; }
if (params.tensor_split.empty()) { params.tensor_split = cmd_params_defaults.tensor_split; }
if (params.use_mmap.empty()) { params.use_mmap = cmd_params_defaults.use_mmap; }
if (params.embeddings.empty()) { params.embeddings = cmd_params_defaults.embeddings; }
@@ -499,6 +510,7 @@ struct cmd_params_instance {
llama_split_mode split_mode;
int main_gpu;
bool no_kv_offload;
bool flash_attn;
std::vector<float> tensor_split;
bool use_mmap;
bool embeddings;
@@ -533,6 +545,7 @@ struct cmd_params_instance {
cparams.type_k = type_k;
cparams.type_v = type_v;
cparams.offload_kqv = !no_kv_offload;
cparams.flash_attn = flash_attn;
cparams.embeddings = embeddings;
return cparams;
@@ -555,6 +568,7 @@ static std::vector<cmd_params_instance> get_cmd_params_instances(const cmd_param
for (const auto & tk : params.type_k)
for (const auto & tv : params.type_v)
for (const auto & nkvo : params.no_kv_offload)
for (const auto & fa : params.flash_attn)
for (const auto & nt : params.n_threads) {
for (const auto & n_prompt : params.n_prompt) {
if (n_prompt == 0) {
@@ -573,6 +587,7 @@ static std::vector<cmd_params_instance> get_cmd_params_instances(const cmd_param
/* .split_mode = */ sm,
/* .main_gpu = */ mg,
/* .no_kv_offload= */ nkvo,
/* .flash_attn = */ fa,
/* .tensor_split = */ ts,
/* .use_mmap = */ mmp,
/* .embeddings = */ embd,
@@ -597,6 +612,7 @@ static std::vector<cmd_params_instance> get_cmd_params_instances(const cmd_param
/* .split_mode = */ sm,
/* .main_gpu = */ mg,
/* .no_kv_offload= */ nkvo,
/* .flash_attn = */ fa,
/* .tensor_split = */ ts,
/* .use_mmap = */ mmp,
/* .embeddings = */ embd,
@@ -634,6 +650,7 @@ struct test {
llama_split_mode split_mode;
int main_gpu;
bool no_kv_offload;
bool flash_attn;
std::vector<float> tensor_split;
bool use_mmap;
bool embeddings;
@@ -658,6 +675,7 @@ struct test {
split_mode = inst.split_mode;
main_gpu = inst.main_gpu;
no_kv_offload = inst.no_kv_offload;
flash_attn = inst.flash_attn;
tensor_split = inst.tensor_split;
use_mmap = inst.use_mmap;
embeddings = inst.embeddings;
@@ -732,7 +750,7 @@ struct test {
"n_batch", "n_ubatch",
"n_threads", "type_k", "type_v",
"n_gpu_layers", "split_mode",
"main_gpu", "no_kv_offload",
"main_gpu", "no_kv_offload", "flash_attn",
"tensor_split", "use_mmap", "embeddings",
"n_prompt", "n_gen", "test_time",
"avg_ns", "stddev_ns",
@@ -754,7 +772,7 @@ struct test {
}
if (field == "cuda" || field == "opencl" || field == "vulkan" || field == "kompute" || field == "metal" ||
field == "gpu_blas" || field == "blas" || field == "sycl" ||field == "f16_kv" || field == "no_kv_offload" ||
field == "use_mmap" || field == "embeddings") {
field == "flash_attn" || field == "use_mmap" || field == "embeddings") {
return BOOL;
}
if (field == "avg_ts" || field == "stddev_ts") {
@@ -788,7 +806,7 @@ struct test {
std::to_string(n_batch), std::to_string(n_ubatch),
std::to_string(n_threads), ggml_type_name(type_k), ggml_type_name(type_v),
std::to_string(n_gpu_layers), split_mode_str(split_mode),
std::to_string(main_gpu), std::to_string(no_kv_offload),
std::to_string(main_gpu), std::to_string(no_kv_offload), std::to_string(flash_attn),
tensor_split_str, std::to_string(use_mmap), std::to_string(embeddings),
std::to_string(n_prompt), std::to_string(n_gen), test_time,
std::to_string(avg_ns()), std::to_string(stdev_ns()),
@@ -956,6 +974,9 @@ struct markdown_printer : public printer {
if (field == "no_kv_offload") {
return "nkvo";
}
if (field == "flash_attn") {
return "fa";
}
if (field == "use_mmap") {
return "mmap";
}
@@ -1002,6 +1023,9 @@ struct markdown_printer : public printer {
if (params.no_kv_offload.size() > 1 || params.no_kv_offload != cmd_params_defaults.no_kv_offload) {
fields.emplace_back("no_kv_offload");
}
if (params.flash_attn.size() > 1 || params.flash_attn != cmd_params_defaults.flash_attn) {
fields.emplace_back("flash_attn");
}
if (params.tensor_split.size() > 1 || params.tensor_split != cmd_params_defaults.tensor_split) {
fields.emplace_back("tensor_split");
}
examples/llava/llava-cli.cpp
+38 -29
View File
@@ -113,11 +113,11 @@ struct llava_context {
};
static void show_additional_info(int /*argc*/, char ** argv) {
LOG_TEE("\n example usage: %s -m <llava-v1.5-7b/ggml-model-q5_k.gguf> --mmproj <llava-v1.5-7b/mmproj-model-f16.gguf> --image <path/to/an/image.jpg> [--temp 0.1] [-p \"describe the image in detail.\"]\n", argv[0]);
LOG_TEE("\n example usage: %s -m <llava-v1.5-7b/ggml-model-q5_k.gguf> --mmproj <llava-v1.5-7b/mmproj-model-f16.gguf> --image <path/to/an/image.jpg> --image <path/to/another/image.jpg> [--temp 0.1] [-p \"describe the image in detail.\"]\n", argv[0]);
LOG_TEE(" note: a lower temperature value like 0.1 is recommended for better quality.\n");
}
static struct llava_image_embed * load_image(llava_context * ctx_llava, gpt_params * params) {
static struct llava_image_embed * load_image(llava_context * ctx_llava, gpt_params * params, const std::string & fname) {
// load and preprocess the image
llava_image_embed * embed = NULL;
@@ -133,9 +133,9 @@ static struct llava_image_embed * load_image(llava_context * ctx_llava, gpt_para
}
params->prompt = remove_image_from_prompt(prompt);
} else {
embed = llava_image_embed_make_with_filename(ctx_llava->ctx_clip, params->n_threads, params->image.c_str());
embed = llava_image_embed_make_with_filename(ctx_llava->ctx_clip, params->n_threads, fname.c_str());
if (!embed) {
LOG_TEE("%s: is %s really an image file?\n", __func__, params->image.c_str());
fprintf(stderr, "%s: is %s really an image file?\n", __func__, fname.c_str());
return NULL;
}
}
@@ -207,17 +207,7 @@ static void process_prompt(struct llava_context * ctx_llava, struct llava_image_
printf("\n");
}
static struct llava_context * llava_init(gpt_params * params) {
const char * clip_path = params->mmproj.c_str();
auto prompt = params->prompt;
if (prompt.empty()) {
prompt = "describe the image in detail.";
}
auto ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1);
static struct llama_model * llava_init(gpt_params * params) {
llama_backend_init();
llama_numa_init(params->numa);
@@ -228,6 +218,19 @@ static struct llava_context * llava_init(gpt_params * params) {
LOG_TEE("%s: error: unable to load model\n" , __func__);
return NULL;
}
return model;
}
static struct llava_context * llava_init_context(gpt_params * params, llama_model * model) {
const char * clip_path = params->mmproj.c_str();
auto prompt = params->prompt;
if (prompt.empty()) {
prompt = "describe the image in detail.";
}
auto ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1);
llama_context_params ctx_params = llama_context_params_from_gpt_params(*params);
ctx_params.n_ctx = params->n_ctx < 2048 ? 2048 : params->n_ctx; // we need a longer context size to process image embeddings
@@ -286,24 +289,30 @@ int main(int argc, char ** argv) {
show_additional_info(argc, argv);
return 1;
}
auto ctx_llava = llava_init(&params);
if (ctx_llava == NULL) {
LOG_TEE("%s: error: failed to init llava\n", __func__);
auto model = llava_init(&params);
if (model == NULL) {
fprintf(stderr, "%s: error: failed to init llava model\n", __func__);
return 1;
}
auto image_embed = load_image(ctx_llava, &params);
if (!image_embed) {
return 1;
for (auto & image : params.image) {
auto ctx_llava = llava_init_context(&params, model);
auto image_embed = load_image(ctx_llava, &params, image);
if (!image_embed) {
std::cerr << "error: failed to load image " << image << ". Terminating\n\n";
return 1;
}
// process the prompt
process_prompt(ctx_llava, image_embed, &params, params.prompt);
llama_print_timings(ctx_llava->ctx_llama);
llava_image_embed_free(image_embed);
ctx_llava->model = NULL;
llava_free(ctx_llava);
}
llama_free_model(model);
// process the prompt
process_prompt(ctx_llava, image_embed, &params, params.prompt);
llama_print_timings(ctx_llava->ctx_llama);
llava_image_embed_free(image_embed);
llava_free(ctx_llava);
return 0;
}
examples/main-cmake-pkg/README.md
+6 -10
View File
@@ -17,11 +17,9 @@ In this case, CLBlast was already installed so the CMake package is referenced i
```cmd
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
mkdir build
cd build
cmake .. -DBUILD_SHARED_LIBS=OFF -DLLAMA_CLBLAST=ON -DCMAKE_PREFIX_PATH=C:/CLBlast/lib/cmake/CLBlast -G "Visual Studio 17 2022" -A x64
cmake --build . --config Release
cmake --install . --prefix C:/LlamaCPP
cmake -B build -DBUILD_SHARED_LIBS=OFF -DLLAMA_CLBLAST=ON -DCMAKE_PREFIX_PATH=C:/CLBlast/lib/cmake/CLBlast -G "Visual Studio 17 2022" -A x64
cmake --build build --config Release
cmake --install build --prefix C:/LlamaCPP
```
### Build main-cmake-pkg
@@ -29,9 +27,7 @@ cmake --install . --prefix C:/LlamaCPP
```cmd
cd ..\examples\main-cmake-pkg
mkdir build
cd build
cmake .. -DBUILD_SHARED_LIBS=OFF -DCMAKE_PREFIX_PATH="C:/CLBlast/lib/cmake/CLBlast;C:/LlamaCPP/lib/cmake/Llama" -G "Visual Studio 17 2022" -A x64
cmake --build . --config Release
cmake --install . --prefix C:/MyLlamaApp
cmake -B build -DBUILD_SHARED_LIBS=OFF -DCMAKE_PREFIX_PATH="C:/CLBlast/lib/cmake/CLBlast;C:/LlamaCPP/lib/cmake/Llama" -G "Visual Studio 17 2022" -A x64
cmake --build build --config Release
cmake --install build --prefix C:/MyLlamaApp
```
examples/main/README.md
+1 -1
View File
@@ -66,7 +66,7 @@ main.exe -m models\7B\ggml-model.bin --ignore-eos -n -1 --random-prompt
In this section, we cover the most commonly used options for running the `main` program with the LLaMA models:
- `-m FNAME, --model FNAME`: Specify the path to the LLaMA model file (e.g., `models/7B/ggml-model.bin`).
- `-m FNAME, --model FNAME`: Specify the path to the LLaMA model file (e.g., `models/7B/ggml-model.gguf`; inferred from `--model-url` if set).
- `-mu MODEL_URL --model-url MODEL_URL`: Specify a remote http url to download the file (e.g https://huggingface.co/ggml-org/models/resolve/main/phi-2/ggml-model-q4_0.gguf).
- `-i, --interactive`: Run the program in interactive mode, allowing you to provide input directly and receive real-time responses.
- `-ins, --instruct`: Run the program in instruction mode, which is particularly useful when working with Alpaca models.
examples/main/main.cpp
+1 -1
View File
@@ -325,7 +325,7 @@ int main(int argc, char ** argv) {
log_tostr(embd_inp.empty()), n_matching_session_tokens, embd_inp.size(), session_tokens.size(), embd_inp.size());
// if we will use the cache for the full prompt without reaching the end of the cache, force
// reevaluation of the last token token to recalculate the cached logits
// reevaluation of the last token to recalculate the cached logits
if (!embd_inp.empty() && n_matching_session_tokens == embd_inp.size() && session_tokens.size() > embd_inp.size()) {
LOGLN("recalculate the cached logits (do): session_tokens.resize( %zu )", embd_inp.size() - 1);
examples/quantize-stats/quantize-stats.cpp
+1 -1
View File
@@ -24,7 +24,7 @@
#endif
struct quantize_stats_params {
std::string model = "models/7B/ggml-model-f16.gguf";
std::string model = DEFAULT_MODEL_PATH;
bool verbose = false;
bool per_layer_stats = false;
bool print_histogram = false;
examples/quantize/CMakeLists.txt
+1 -1
View File
@@ -1,6 +1,6 @@
set(TARGET quantize)
add_executable(${TARGET} quantize.cpp)
install(TARGETS ${TARGET} RUNTIME)
target_link_libraries(${TARGET} PRIVATE llama build_info ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE llama common ${CMAKE_THREAD_LIBS_INIT})
target_include_directories(${TARGET} PRIVATE ../../common)
target_compile_features(${TARGET} PRIVATE cxx_std_11)
examples/quantize/quantize.cpp
+59 -44
View File
@@ -9,7 +9,6 @@
#include <unordered_map>
#include <fstream>
#include <cmath>
#include <algorithm>
struct quant_option {
std::string name;
@@ -54,6 +53,10 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "COPY", LLAMA_FTYPE_ALL_F32, "only copy tensors, no quantizing", },
};
static const char * const LLM_KV_QUANTIZE_IMATRIX_FILE = "quantize.imatrix.file";
static const char * const LLM_KV_QUANTIZE_IMATRIX_DATASET = "quantize.imatrix.dataset";
static const char * const LLM_KV_QUANTIZE_IMATRIX_N_ENTRIES = "quantize.imatrix.entries_count";
static const char * const LLM_KV_QUANTIZE_IMATRIX_N_CHUNKS = "quantize.imatrix.chunks_count";
static bool try_parse_ftype(const std::string & ftype_str_in, llama_ftype & ftype, std::string & ftype_str_out) {
std::string ftype_str;
@@ -114,7 +117,7 @@ static void usage(const char * executable) {
exit(1);
}
static void load_imatrix(const std::string & imatrix_file, std::unordered_map<std::string, std::vector<float>> & imatrix_data) {
static int load_imatrix(const std::string & imatrix_file, std::string & imatrix_dataset, std::unordered_map<std::string, std::vector<float>> & imatrix_data) {
std::ifstream in(imatrix_file.c_str(), std::ios::binary);
if (!in) {
printf("%s: failed to open %s\n",__func__, imatrix_file.c_str());
@@ -161,18 +164,33 @@ static void load_imatrix(const std::string & imatrix_file, std::unordered_map<st
printf("%s: loaded data (size = %6d, ncall = %6d) for '%s'\n", __func__, int(e.size()), ncall, name.c_str());
}
}
printf("%s: loaded %d importance matrix entries from %s\n", __func__, int(imatrix_data.size()), imatrix_file.c_str());
// latest imatrix version contains the dataset filename at the end of the file
int m_last_call = 0;
if (in.peek() != EOF) {
in.read((char *)&m_last_call, sizeof(m_last_call));
int dataset_len;
in.read((char *)&dataset_len, sizeof(dataset_len));
std::vector<char> dataset_as_vec(dataset_len);
in.read(dataset_as_vec.data(), dataset_len);
imatrix_dataset.assign(dataset_as_vec.begin(), dataset_as_vec.end());
printf("%s: imatrix dataset='%s'\n", __func__, imatrix_dataset.c_str());
}
printf("%s: loaded %d importance matrix entries from %s computed on %d chunks\n", __func__, int(imatrix_data.size()), imatrix_file.c_str(), m_last_call);
return m_last_call;
}
static void prepare_imatrix(const std::string & imatrix_file,
static int prepare_imatrix(const std::string & imatrix_file,
std::string & imatrix_dataset,
const std::vector<std::string> & included_weights,
const std::vector<std::string> & excluded_weights,
std::unordered_map<std::string, std::vector<float>> & imatrix_data) {
int m_last_call = -1;
if (!imatrix_file.empty()) {
load_imatrix(imatrix_file, imatrix_data);
m_last_call = load_imatrix(imatrix_file, imatrix_dataset, imatrix_data);
}
if (imatrix_data.empty()) {
return;
return m_last_call;
}
if (!excluded_weights.empty()) {
for (auto& name : excluded_weights) {
@@ -198,6 +216,7 @@ static void prepare_imatrix(const std::string & imatrix_file,
if (!imatrix_data.empty()) {
printf("%s: have %d importance matrix entries\n", __func__, int(imatrix_data.size()));
}
return m_last_call;
}
static ggml_type parse_ggml_type(const char * arg) {
@@ -212,43 +231,6 @@ static ggml_type parse_ggml_type(const char * arg) {
return result;
}
static bool parse_kv_override(const char * data, std::vector<llama_model_kv_override> & overrides) {
const char* sep = strchr(data, '=');
if (sep == nullptr || sep - data >= 128) {
fprintf(stderr, "%s: malformed KV override '%s'\n", __func__, data);
return false;
}
llama_model_kv_override kvo;
std::strncpy(kvo.key, data, sep - data);
kvo.key[sep - data] = 0;
sep++;
if (strncmp(sep, "int:", 4) == 0) {
sep += 4;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_INT;
kvo.int_value = std::atol(sep);
} else if (strncmp(sep, "float:", 6) == 0) {
sep += 6;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_FLOAT;
kvo.float_value = std::atof(sep);
} else if (strncmp(sep, "bool:", 5) == 0) {
sep += 5;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_BOOL;
if (std::strcmp(sep, "true") == 0) {
kvo.bool_value = true;
} else if (std::strcmp(sep, "false") == 0) {
kvo.bool_value = false;
} else {
fprintf(stderr, "%s: invalid boolean value for KV override '%s'\n", __func__, data);
return false;
}
} else {
fprintf(stderr, "%s: invalid type for KV override '%s'\n", __func__, data);
return false;
}
overrides.emplace_back(std::move(kvo));
return true;
}
int main(int argc, char ** argv) {
if (argc < 3) {
usage(argv[0]);
@@ -317,10 +299,43 @@ int main(int argc, char ** argv) {
usage(argv[0]);
}
std::string imatrix_dataset;
std::unordered_map<std::string, std::vector<float>> imatrix_data;
prepare_imatrix(imatrix_file, included_weights, excluded_weights, imatrix_data);
int m_last_call = prepare_imatrix(imatrix_file, imatrix_dataset, included_weights, excluded_weights, imatrix_data);
if (!imatrix_data.empty()) {
params.imatrix = &imatrix_data;
{
llama_model_kv_override kvo;
std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_FILE);
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
strncpy(kvo.val_str, imatrix_file.c_str(), 127);
kvo.val_str[127] = '\0';
kv_overrides.emplace_back(std::move(kvo));
}
if (!imatrix_dataset.empty()) {
llama_model_kv_override kvo;
std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_DATASET);
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
strncpy(kvo.val_str, imatrix_dataset.c_str(), 127);
kvo.val_str[127] = '\0';
kv_overrides.emplace_back(std::move(kvo));
}
{
llama_model_kv_override kvo;
std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_N_ENTRIES);
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_INT;
kvo.val_i64 = imatrix_data.size();
kv_overrides.emplace_back(std::move(kvo));
}
if (m_last_call > 0) {
llama_model_kv_override kvo;
std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_N_CHUNKS);
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_INT;
kvo.val_i64 = m_last_call;
kv_overrides.emplace_back(std::move(kvo));
}
}
if (!kv_overrides.empty()) {
kv_overrides.emplace_back();
examples/server/README.md
+7 -6
View File
@@ -74,15 +74,18 @@ page cache before using this. See https://github.com/ggerganov/llama.cpp/issues/
- Using `make`:
```bash
make
make server
```
- Using `CMake`:
```bash
cmake --build . --config Release
cmake -B build
cmake --build build --config Release -t server
```
Binary is at `./build/bin/server`
## Build with SSL
`server` can also be built with SSL support using OpenSSL 3
@@ -99,10 +102,8 @@ page cache before using this. See https://github.com/ggerganov/llama.cpp/issues/
- Using `CMake`:
```bash
mkdir build
cd build
cmake .. -DLLAMA_SERVER_SSL=ON
make server
cmake -B build -DLLAMA_SERVER_SSL=ON
cmake --build build --config Release -t server
```
## Quick Start
examples/server/bench/bench.py
+1
View File
@@ -268,6 +268,7 @@ def start_server_background(args):
server_args.extend(['--defrag-thold', "0.1"])
server_args.append('--cont-batching')
server_args.append('--metrics')
server_args.append('--flash-attn')
server_args.extend(['--log-format', "text"])
args = [str(arg) for arg in [server_path, *server_args]]
print(f"bench: starting server with: {' '.join(args)}")
examples/server/bench/script.js
+2 -1
View File
@@ -90,7 +90,8 @@ export default function () {
"model": model,
"stream": true,
"seed": 42,
"max_tokens": max_tokens
"max_tokens": max_tokens,
"stop": ["<|im_end|>"] // This is temporary for phi-2 base (i.e. not instructed) since the server expects that the model always to emit BOS
}
const params = {method: 'POST', body: JSON.stringify(payload)};
examples/server/server.cpp
+30 -36
View File
@@ -1208,6 +1208,27 @@ struct server_context {
LOG_VERBOSE("eos token found", {});
}
auto n_ctx_train = llama_n_ctx_train(model);
if (slot.params.n_predict < 1 && slot.n_predict < 1 && slot.ga_n == 1
&& slot.n_prompt_tokens + slot.n_decoded >= n_ctx_train) {
LOG_WARNING("n_predict is not set and self-context extend is disabled."
" Limiting generated tokens to n_ctx_train to avoid EOS-less generation infinite loop", {
{ "id_slot", slot.id },
{ "params.n_predict", slot.params.n_predict },
{ "slot.n_prompt_tokens", slot.n_prompt_tokens },
{ "slot.n_decoded", slot.n_decoded },
{ "slot.n_predict", slot.n_predict },
{ "n_slots", params.n_parallel },
{ "slot.n_ctx", slot.n_ctx },
{ "n_ctx", n_ctx },
{ "n_ctx_train", n_ctx_train },
{ "ga_n", slot.ga_n },
});
slot.truncated = true;
slot.stopped_limit = true;
slot.has_next_token = false; // stop prediction
}
LOG_VERBOSE("next token", {
{"id_slot", slot.id},
{"id_task", slot.id_task},
@@ -2142,7 +2163,7 @@ struct server_context {
});
// process the created batch of tokens
for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += n_batch) {
for (int32_t i = 0; i < batch.n_tokens; i += n_batch) {
const int32_t n_tokens = std::min(n_batch, batch.n_tokens - i);
for (auto & slot : slots) {
@@ -2333,7 +2354,7 @@ static void server_print_usage(const char * argv0, const gpt_params & params, co
printf(" disable KV offload\n");
}
printf(" -m FNAME, --model FNAME\n");
printf(" model path (default: %s)\n", params.model.c_str());
printf(" model path (default: models/$filename with filename from --hf-file or --model-url if set, otherwise %s)\n", DEFAULT_MODEL_PATH);
printf(" -mu MODEL_URL, --model-url MODEL_URL\n");
printf(" model download url (default: unused)\n");
printf(" -hfr REPO, --hf-repo REPO\n");
@@ -2357,6 +2378,7 @@ static void server_print_usage(const char * argv0, const gpt_params & params, co
printf(" --embeddings enable embedding vector output (default: %s)\n", params.embedding ? "enabled" : "disabled");
printf(" -np N, --parallel N number of slots for process requests (default: %d)\n", params.n_parallel);
printf(" -cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)\n");
printf(" -fa, --flash-attn enable Flash Attention (default: %s)\n", params.flash_attn ? "enabled" : "disabled");
printf(" -spf FNAME, --system-prompt-file FNAME\n");
printf(" set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications.\n");
printf(" -ctk TYPE, --cache-type-k TYPE\n");
@@ -2372,7 +2394,7 @@ static void server_print_usage(const char * argv0, const gpt_params & params, co
printf(" -n, --n-predict maximum tokens to predict (default: %d)\n", params.n_predict);
printf(" --override-kv KEY=TYPE:VALUE\n");
printf(" advanced option to override model metadata by key. may be specified multiple times.\n");
printf(" types: int, float, bool. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");
printf(" types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");
printf(" -gan N, --grp-attn-n N set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`\n");
printf(" -gaw N, --grp-attn-w N set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`\n");
printf(" --chat-template JINJA_TEMPLATE\n");
@@ -2722,6 +2744,8 @@ static void server_params_parse(int argc, char ** argv, server_params & sparams,
params.embedding = true;
} else if (arg == "-cb" || arg == "--cont-batching") {
params.cont_batching = true;
} else if (arg == "-fa" || arg == "--flash-attn") {
params.flash_attn = true;
} else if (arg == "-np" || arg == "--parallel") {
if (++i >= argc) {
invalid_param = true;
@@ -2803,43 +2827,11 @@ static void server_params_parse(int argc, char ** argv, server_params & sparams,
invalid_param = true;
break;
}
char * sep = strchr(argv[i], '=');
if (sep == nullptr || sep - argv[i] >= 128) {
fprintf(stderr, "error: Malformed KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
struct llama_model_kv_override kvo;
std::strncpy(kvo.key, argv[i], sep - argv[i]);
kvo.key[sep - argv[i]] = 0;
sep++;
if (strncmp(sep, "int:", 4) == 0) {
sep += 4;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_INT;
kvo.int_value = std::atol(sep);
} else if (strncmp(sep, "float:", 6) == 0) {
sep += 6;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_FLOAT;
kvo.float_value = std::atof(sep);
} else if (strncmp(sep, "bool:", 5) == 0) {
sep += 5;
kvo.tag = LLAMA_KV_OVERRIDE_TYPE_BOOL;
if (std::strcmp(sep, "true") == 0) {
kvo.bool_value = true;
} else if (std::strcmp(sep, "false") == 0) {
kvo.bool_value = false;
} else {
fprintf(stderr, "error: Invalid boolean value for KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
} else {
if (!parse_kv_override(argv[i], params.kv_overrides)) {
fprintf(stderr, "error: Invalid type for KV override: %s\n", argv[i]);
invalid_param = true;
break;
}
params.kv_overrides.push_back(kvo);
} else {
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
server_print_usage(argv[0], default_params, default_sparams);
@@ -2847,6 +2839,8 @@ static void server_params_parse(int argc, char ** argv, server_params & sparams,
}
}
gpt_params_handle_model_default(params);
if (!params.kv_overrides.empty()) {
params.kv_overrides.emplace_back();
params.kv_overrides.back().key[0] = 0;
examples/server/tests/features/embeddings.feature
+1 -1
View File
@@ -5,7 +5,7 @@ Feature: llama.cpp server
Background: Server startup
Given a server listening on localhost:8080
And a model url https://huggingface.co/ggml-org/models/resolve/main/bert-bge-small/ggml-model-f16.gguf
And a model file ggml-model-f16.gguf
And a model file bert-bge-small.gguf
And a model alias bert-bge-small
And 42 as server seed
And 2 slots