Skip to content

llama-bench : allow using a different printer for stderr with -oe #7722

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 4, 2024
Merged

llama-bench : allow using a different printer for stderr with -oe #7722

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
145 changes: 92 additions & 53 deletions examples/llama-bench/llama-bench.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -140,10 +140,11 @@ static std::string get_gpu_info() {
}

// command line params
enum output_formats {CSV, JSON, MARKDOWN, SQL};
enum output_formats {NONE, CSV, JSON, MARKDOWN, SQL};

static const char * output_format_str(output_formats format) {
switch (format) {
case NONE: return "none";
case CSV: return "csv";
case JSON: return "json";
case MARKDOWN: return "md";
Expand All @@ -152,6 +153,23 @@ static const char * output_format_str(output_formats format) {
}
}

static bool output_format_from_str(const std::string & s, output_formats & format) {
if (s == "none") {
format = NONE;
} else if (s == "csv") {
format = CSV;
} else if (s == "json") {
format = JSON;
} else if (s == "md") {
format = MARKDOWN;
} else if (s == "sql") {
format = SQL;
} else {
return false;
}
return true;
}

static const char * split_mode_str(llama_split_mode mode) {
switch (mode) {
case LLAMA_SPLIT_MODE_NONE: return "none";
Expand Down Expand Up @@ -190,31 +208,33 @@ struct cmd_params {
int reps;
bool verbose;
output_formats output_format;
output_formats output_format_stderr;
};

static const cmd_params cmd_params_defaults = {
/* model */ {"models/7B/ggml-model-q4_0.gguf"},
/* n_prompt */ {512},
/* n_gen */ {128},
/* n_pg */ {},
/* n_batch */ {2048},
/* n_ubatch */ {512},
/* type_k */ {GGML_TYPE_F16},
/* type_v */ {GGML_TYPE_F16},
/* n_threads */ {cpu_get_num_math()},
/* n_gpu_layers */ {99},
/* rpc_servers */ {""},
/* 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},
/* numa */ GGML_NUMA_STRATEGY_DISABLED,
/* reps */ 5,
/* verbose */ false,
/* output_format */ MARKDOWN
/* model */ {"models/7B/ggml-model-q4_0.gguf"},
/* n_prompt */ {512},
/* n_gen */ {128},
/* n_pg */ {},
/* n_batch */ {2048},
/* n_ubatch */ {512},
/* type_k */ {GGML_TYPE_F16},
/* type_v */ {GGML_TYPE_F16},
/* n_threads */ {cpu_get_num_math()},
/* n_gpu_layers */ {99},
/* rpc_servers */ {""},
/* 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},
/* numa */ GGML_NUMA_STRATEGY_DISABLED,
/* reps */ 5,
/* verbose */ false,
/* output_format */ MARKDOWN,
/* output_format_stderr */ NONE,
};

static void print_usage(int /* argc */, char ** argv) {
Expand Down Expand Up @@ -243,6 +263,7 @@ static void print_usage(int /* argc */, char ** argv) {
printf(" -ts, --tensor-split <ts0/ts1/..> (default: 0)\n");
printf(" -r, --repetitions <n> (default: %d)\n", cmd_params_defaults.reps);
printf(" -o, --output <csv|json|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format));
printf(" -oe, --output-err <csv|json|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("\n");
printf("Multiple values can be given for each parameter by separating them with ',' or by specifying the parameter multiple times.\n");
Expand Down Expand Up @@ -284,6 +305,7 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {

params.verbose = cmd_params_defaults.verbose;
params.output_format = cmd_params_defaults.output_format;
params.output_format_stderr = cmd_params_defaults.output_format_stderr;
params.reps = cmd_params_defaults.reps;

for (int i = 1; i < argc; i++) {
Expand Down Expand Up @@ -493,18 +515,13 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
invalid_param = true;
break;
}
if (argv[i] == std::string("csv")) {
params.output_format = CSV;
} else if (argv[i] == std::string("json")) {
params.output_format = JSON;
} else if (argv[i] == std::string("md")) {
params.output_format = MARKDOWN;
} else if (argv[i] == std::string("sql")) {
params.output_format = SQL;
} else {
invalid_param = !output_format_from_str(argv[i], params.output_format);
} else if (arg == "-oe" || arg == "--output-err") {
if (++i >= argc) {
invalid_param = true;
break;
}
invalid_param = !output_format_from_str(argv[i], params.output_format_stderr);
} else if (arg == "-v" || arg == "--verbose") {
params.verbose = true;
} else {
Expand Down Expand Up @@ -1278,6 +1295,22 @@ static void llama_null_log_callback(enum ggml_log_level level, const char * text
(void) user_data;
}

static std::unique_ptr<printer> create_printer(output_formats format) {
switch (format) {
case NONE:
return nullptr;
case CSV:
return std::unique_ptr<printer>(new csv_printer());
case JSON:
return std::unique_ptr<printer>(new json_printer());
case MARKDOWN:
return std::unique_ptr<printer>(new markdown_printer());
case SQL:
return std::unique_ptr<printer>(new sql_printer());
}
GGML_ASSERT(false);
}

int main(int argc, char ** argv) {
// try to set locale for unicode characters in markdown
setlocale(LC_CTYPE, ".UTF-8");
Expand All @@ -1304,26 +1337,18 @@ int main(int argc, char ** argv) {
llama_numa_init(params.numa);

// initialize printer
std::unique_ptr<printer> p;
switch (params.output_format) {
case CSV:
p.reset(new csv_printer());
break;
case JSON:
p.reset(new json_printer());
break;
case MARKDOWN:
p.reset(new markdown_printer());
break;
case SQL:
p.reset(new sql_printer());
break;
default:
assert(false);
exit(1);
std::unique_ptr<printer> p = create_printer(params.output_format);
std::unique_ptr<printer> p_err = create_printer(params.output_format_stderr);

if (p) {
p->fout = stdout;
p->print_header(params);
}

if (p_err) {
p_err->fout = stderr;
p_err->print_header(params);
}
p->fout = stdout;
p->print_header(params);

std::vector<cmd_params_instance> params_instances = get_cmd_params_instances(params);

Expand Down Expand Up @@ -1381,7 +1406,15 @@ int main(int argc, char ** argv) {
t.samples_ns.push_back(t_ns);
}

p->print_test(t);
if (p) {
p->print_test(t);
fflush(p->fout);
}

if (p_err) {
p_err->print_test(t);
fflush(p_err->fout);
}

llama_print_timings(ctx);

Expand All @@ -1390,7 +1423,13 @@ int main(int argc, char ** argv) {

llama_free_model(lmodel);

p->print_footer();
if (p) {
p->print_footer();
}

if (p_err) {
p_err->print_footer();
}

llama_backend_free();

Expand Down
16 changes: 9 additions & 7 deletions scripts/compare-commits.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,16 +10,18 @@ set -x

bench_args="${@:3}"

rm -f llama-bench.sqlite
rm -f llama-bench.sqlite > /dev/null

# to test a backend, call the script with the corresponding environment variable (e.g. LLAMA_CUDA=1 ./scripts/compare-commits.sh ...)

git checkout $1
make clean && make -j32 $make_opts llama-bench
./llama-bench -o sql $bench_args | tee /dev/tty | sqlite3 llama-bench.sqlite
git checkout $1 > /dev/null
make clean > /dev/null
make -j$(nproc) $make_opts llama-bench > /dev/null
./llama-bench -o sql -oe md $bench_args | sqlite3 llama-bench.sqlite

git checkout $2
make clean && make -j32 $make_opts llama-bench
./llama-bench -o sql $bench_args | tee /dev/tty | sqlite3 llama-bench.sqlite
git checkout $2 > /dev/null
make clean > /dev/null
make -j$(nproc) $make_opts llama-bench > /dev/null
./llama-bench -o sql -oe md $bench_args | sqlite3 llama-bench.sqlite

./scripts/compare-llama-bench.py -b $1 -c $2
Loading