vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

markdown: also show values that differ from the default

Browse source
This commit is contained in:
slaren 2023-08-17 03:06:07 +02:00
parent cac70312e3
commit 569dc6f3d0
1 changed files with 16 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

27
examples/llama-bench/llama-bench.cpp
View file

@ -357,14 +357,13 @@ struct cmd_params_instance {
int n_gen; int n_gen;
int n_batch; int n_batch;
bool f32_kv; bool f32_kv;
int n_threads;
int n_gpu_layers; int n_gpu_layers;
int main_gpu; int main_gpu;
bool mul_mat_q; bool mul_mat_q;
bool low_vram; bool low_vram;
std::array<float, LLAMA_MAX_DEVICES> tensor_split; std::array<float, LLAMA_MAX_DEVICES> tensor_split;
int n_threads;
llama_context_params to_llama_params() const { llama_context_params to_llama_params() const {
llama_context_params lparams = llama_context_default_params(); llama_context_params lparams = llama_context_default_params();
lparams.n_ctx = n_prompt + n_gen; lparams.n_ctx = n_prompt + n_gen;
@ -546,9 +545,11 @@ struct test {
max_nonzero = i; max_nonzero = i;
} }
} }
for (int i = 0; i < max_nonzero; i++) { for (int i = 0; i <= max_nonzero; i++) {
tensor_split_str += std::to_string(tensor_split[i]); char buf[32];
if (i < max_nonzero - 1) { snprintf(buf, sizeof(buf), "%.2f", tensor_split[i]);
tensor_split_str += buf;
if (i < max_nonzero) {
tensor_split_str += "/"; tensor_split_str += "/";
} }
} }
@ -593,6 +594,7 @@ struct printer {
virtual void print_footer() {}; virtual void print_footer() {};
}; };
// TODO: escape strings
struct csv_printer : public printer { struct csv_printer : public printer {
virtual void print_header(const cmd_params & params) { virtual void print_header(const cmd_params & params) {
std::vector<std::string> fields = test::get_fields(); std::vector<std::string> fields = test::get_fields();
@ -665,24 +667,27 @@ struct markdown_printer : public printer {
if (!is_cpu_backend) { if (!is_cpu_backend) {
fields.push_back("n_gpu_layers"); fields.push_back("n_gpu_layers");
} }
if (params.n_threads.size() > 1 || is_cpu_backend) { if (params.n_batch.size() > 1 || params.n_threads != cmd_params_defaults.n_threads || is_cpu_backend) {
fields.push_back("n_threads"); fields.push_back("n_threads");
} }
if (params.n_batch.size() > 1) { if (params.n_batch.size() > 1 || params.n_batch != cmd_params_defaults.n_batch) {
fields.push_back("n_batch"); fields.push_back("n_batch");
} }
if (params.f32_kv.size() > 1) { if (params.f32_kv.size() > 1 || params.f32_kv != cmd_params_defaults.f32_kv) {
fields.push_back("f16_kv"); fields.push_back("f16_kv");
} }
if (params.main_gpu.size() > 1) { if (params.main_gpu.size() > 1 || params.main_gpu != cmd_params_defaults.main_gpu) {
fields.push_back("main_gpu"); fields.push_back("main_gpu");
} }
if (params.mul_mat_q.size() > 1) { if (params.mul_mat_q.size() > 1 || params.mul_mat_q != cmd_params_defaults.mul_mat_q) {
fields.push_back("mul_mat_q"); fields.push_back("mul_mat_q");
} }
if (params.low_vram.size() > 1) { if (params.low_vram.size() > 1 || params.low_vram != cmd_params_defaults.low_vram) {
fields.push_back("low_vram"); fields.push_back("low_vram");
} }
if (params.tensor_split.size() > 1 || params.tensor_split != cmd_params_defaults.tensor_split) {
fields.push_back("tensor_split");
}
fields.push_back("test"); fields.push_back("test");
fields.push_back("t/s"); fields.push_back("t/s");