CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama-bench : log benchmark progress (#9287)

Browse Source 
* llama-bench : add optional progress messages
This commit is contained in:
Aarni Koskela
2024-09-07 00:03:01 +03:00
committed by GitHub
parent 815b1fb20a
commit 134bc38ecf
1 changed files with 26 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
examples/llama-bench/llama-bench.cpp
View File

@@ -249,6 +249,7 @@ struct cmd_params {
ggml_sched_priority prio; ggml_sched_priority prio;
int delay; int delay;
bool verbose; bool verbose;
bool progress;
output_formats output_format; output_formats output_format;
output_formats output_format_stderr; output_formats output_format_stderr;
}; };
@@ -280,6 +281,7 @@ static const cmd_params cmd_params_defaults = {
/* prio */ GGML_SCHED_PRIO_NORMAL, /* prio */ GGML_SCHED_PRIO_NORMAL,
/* delay */ 0, /* delay */ 0,
/* verbose */ false, /* verbose */ false,
/* progress */ false,
/* output_format */ MARKDOWN, /* output_format */ MARKDOWN,
/* output_format_stderr */ NONE, /* output_format_stderr */ NONE,
}; };
@@ -319,6 +321,7 @@ static void print_usage(int /* argc */, char ** argv) {
printf(" -o, --output <csv|json|jsonl|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format)); printf(" -o, --output <csv|json|jsonl|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format));
printf(" -oe, --output-err <csv|json|jsonl|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format_stderr)); printf(" -oe, --output-err <csv|json|jsonl|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format_stderr));
printf(" -v, --verbose (default: %s)\n", cmd_params_defaults.verbose ? "1" : "0"); printf(" -v, --verbose (default: %s)\n", cmd_params_defaults.verbose ? "1" : "0");
printf(" --progress (default: %s)\n", cmd_params_defaults.progress ? "1" : "0");
printf("\n"); printf("\n");
printf("Multiple values can be given for each parameter by separating them with ',' or by specifying the parameter multiple times.\n"); printf("Multiple values can be given for each parameter by separating them with ',' or by specifying the parameter multiple times.\n");
} }
@@ -364,6 +367,7 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
params.numa = cmd_params_defaults.numa; params.numa = cmd_params_defaults.numa;
params.prio = cmd_params_defaults.prio; params.prio = cmd_params_defaults.prio;
params.delay = cmd_params_defaults.delay; params.delay = cmd_params_defaults.delay;
params.progress = cmd_params_defaults.progress;
for (int i = 1; i < argc; i++) { for (int i = 1; i < argc; i++) {
arg = argv[i]; arg = argv[i];
@@ -616,6 +620,8 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
invalid_param = !output_format_from_str(argv[i], params.output_format_stderr); invalid_param = !output_format_from_str(argv[i], params.output_format_stderr);
} else if (arg == "-v" || arg == "--verbose") { } else if (arg == "-v" || arg == "--verbose") {
params.verbose = true; params.verbose = true;
} else if (arg == "--progress") {
params.progress = true;
} else { } else {
invalid_param = true; invalid_param = true;
break; break;
@@ -1523,7 +1529,13 @@ int main(int argc, char ** argv) {
llama_model * lmodel = nullptr; llama_model * lmodel = nullptr;
const cmd_params_instance * prev_inst = nullptr; const cmd_params_instance * prev_inst = nullptr;
int params_idx = 0;
auto params_count = params_instances.size();
for (const auto & inst : params_instances) { for (const auto & inst : params_instances) {
params_idx ++;
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%ld: starting\n", params_idx, params_count);
}
// keep the same model between tests when possible // keep the same model between tests when possible
if (!lmodel || !prev_inst || !inst.equal_mparams(*prev_inst)) { if (!lmodel || !prev_inst || !inst.equal_mparams(*prev_inst)) {
if (lmodel) { if (lmodel) {
@@ -1556,7 +1568,7 @@ int main(int argc, char ** argv) {
struct ggml_threadpool_params tpp = ggml_threadpool_params_default(t.n_threads); struct ggml_threadpool_params tpp = ggml_threadpool_params_default(t.n_threads);
if (!parse_cpu_mask(t.cpu_mask, tpp.cpumask)) { if (!parse_cpu_mask(t.cpu_mask, tpp.cpumask)) {
LOG_TEE("%s: failed to parse cpu-mask: %s\n", __func__, t.cpu_mask.c_str()); fprintf(stderr, "%s: failed to parse cpu-mask: %s\n", __func__, t.cpu_mask.c_str());
exit(1); exit(1);
} }
tpp.strict_cpu = t.cpu_strict; tpp.strict_cpu = t.cpu_strict;
@@ -1565,7 +1577,7 @@ int main(int argc, char ** argv) {
struct ggml_threadpool* threadpool = ggml_threadpool_new(&tpp); struct ggml_threadpool* threadpool = ggml_threadpool_new(&tpp);
if (!threadpool) { if (!threadpool) {
LOG_TEE("%s: threadpool create failed : n_threads %d\n", __func__, tpp.n_threads); fprintf(stderr, "%s: threadpool create failed : n_threads %d\n", __func__, tpp.n_threads);
exit(1); exit(1);
} }
@@ -1573,10 +1585,16 @@ int main(int argc, char ** argv) {
// warmup run // warmup run
if (t.n_prompt > 0) { if (t.n_prompt > 0) {
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%ld: warmup prompt run\n", params_idx, params_count);
}
//test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads); //test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads);
test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads); test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads);
} }
if (t.n_gen > 0) { if (t.n_gen > 0) {
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%ld: warmup generation run\n", params_idx, params_count);
}
test_gen(ctx, 1, 0, t.n_threads); test_gen(ctx, 1, 0, t.n_threads);
} }
@@ -1586,9 +1604,15 @@ int main(int argc, char ** argv) {
uint64_t t_start = get_time_ns(); uint64_t t_start = get_time_ns();
if (t.n_prompt > 0) { if (t.n_prompt > 0) {
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%ld: prompt run %d/%d\n", params_idx, params_count, i + 1, params.reps);
}
test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads); test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads);
} }
if (t.n_gen > 0) { if (t.n_gen > 0) {
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%ld: generation run %d/%d\n", params_idx, params_count, i + 1, params.reps);
}
test_gen(ctx, t.n_gen, t.n_prompt, t.n_threads); test_gen(ctx, t.n_gen, t.n_prompt, t.n_threads);
} }