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

fix llama_reset_model_time

Browse source
This commit is contained in:
Jia Liu 2024-09-19 11:30:47 +08:00
parent 24bea1549b
commit 216e7d9648
3 changed files with 7 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/llama-bench/llama-bench.cpp
View file

@ -1558,7 +1558,8 @@ int main(int argc, char ** argv) {
} }
prev_inst = &inst; prev_inst = &inst;
} else { } else {
llama_model_reset_time(lmodel); // ensure load_time dost not accumulate in llama_bench when not loading the same model
llama_reset_model_time(lmodel);
} }
llama_context * ctx = llama_new_context_with_model(lmodel, inst.to_llama_cparams()); llama_context * ctx = llama_new_context_with_model(lmodel, inst.to_llama_cparams());

2
include/llama.h
View file

@ -414,7 +414,7 @@ extern "C" {
const char * path_model, const char * path_model,
struct llama_model_params params); struct llama_model_params params);
LLAMA_API void llama_model_reset_time(struct llama_model * model); LLAMA_API void llama_reset_model_time(struct llama_model * model);
LLAMA_API void llama_free_model(struct llama_model * model); LLAMA_API void llama_free_model(struct llama_model * model);

9
src/llama.cpp
View file

@ -8809,11 +8809,6 @@ static bool llm_load_tensors(
return true; return true;
} }
void llama_model_reset_time(llama_model * model) {
model->t_start_us = ggml_time_us();
model->t_load_us = ggml_time_us() - model->t_start_us;
}
// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback // Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) { static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) {
model.t_start_us = ggml_time_us(); model.t_start_us = ggml_time_us();
@ -18695,6 +18690,10 @@ struct llama_model * llama_load_model_from_file(
return model; return model;
} }
void llama_reset_model_time(llama_model * model) {
model->t_start_us = ggml_time_us() - model->t_load_us;
}
void llama_free_model(struct llama_model * model) { void llama_free_model(struct llama_model * model) {
delete model; delete model;
} }