vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-09 23:24:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add llama_model_n_flops

Browse source
This commit is contained in:
Lizonghang 2024-11-20 19:40:27 +04:00
parent 10f6f92c7e
commit 477ecf2084
4 changed files with 445 additions and 107 deletions
Download patch file Download diff file Expand all files Collapse all files

4
common/common.cpp
View file

@ -841,6 +841,8 @@ static void llama_assign_n_layer_window(
return;
}
(void)my_rank;
std::fill_n(n_layer_window, n_world, DEFAULT_N_LAYER_WINDOW);
}
@ -894,7 +896,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
device_info dev_info;
dev_info.rank = params.rank;
llama_profile_device(&dev_info, model, params.model.c_str(), params.cpuparams.n_threads);
llama_profile_device(&dev_info, model, ml, params.model.c_str(), params.cpuparams.n_threads);
// create llama context
struct llama_context_params cparams = llama_context_params_from_gpt_params(params);

15
common/profiler.h
View file

@ -67,6 +67,21 @@ struct device_info {
: rank(0), device_name(""), disk_read_bandwidth(0.0f), cpu_props(), memory(), gpu_support(), gpu_props() {}
};
struct flops_info {
// model flops
int64_t input_flops;
int64_t output_flops;
int64_t layer_flops;
// model params
int64_t input_params;
int64_t output_params;
int64_t layer_params;
flops_info()
: input_flops(0), output_flops(0), layer_flops(0), input_params(0), output_params(0), layer_params(0) {}
};
enum profiler_backend_type {
PROFILER_BACKEND_TYPE_CPU = 0,
PROFILER_BACKEND_TYPE_METAL = 1,