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

ggml: unify backend logging mechanism (#9709)

Browse source 
* Add scaffolding for ggml logging macros

* Metal backend now uses GGML logging

* Cuda backend now uses GGML logging

* Cann backend now uses GGML logging

* Add enum tag to parameters

* Use C memory allocation funcs

* Fix compile error

* Use GGML_LOG instead of GGML_PRINT

* Rename llama_state to llama_logger_state

* Prevent null format string

* Fix whitespace

* Remove log callbacks from ggml backends

* Remove cuda log statement
This commit is contained in:
bandoti 2024-10-03 12:39:03 -03:00 • committed by GitHub
parent e3c355ba65
commit d6fe7abf04
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 197 additions and 340 deletions
Download patch file Download diff file Expand all files Collapse all files

26
src/llama.cpp
View file

@ -2266,17 +2266,12 @@ static std::string llama_token_to_piece(const struct llama_model * model, llama_
// globals
//
struct llama_state {
llama_state() {
llama_log_set(log_callback, log_callback_user_data);
}
// We save the log callback globally
struct llama_logger_state {
ggml_log_callback log_callback = llama_log_callback_default;
void * log_callback_user_data = nullptr;
};
static llama_state g_state;
static llama_logger_state g_logger_state;
// available llama models
enum e_model {
@ -21850,16 +21845,9 @@ const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal
}
void llama_log_set(ggml_log_callback log_callback, void * user_data) {
g_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
g_state.log_callback_user_data = user_data;
ggml_backend_set_log_callback(log_callback, user_data);
#ifdef GGML_USE_METAL
ggml_backend_metal_log_set_callback(g_state.log_callback, g_state.log_callback_user_data);
#elif defined(GGML_USE_CANN)
ggml_backend_cann_log_set_callback(g_state.log_callback, g_state.log_callback_user_data);
#endif
ggml_log_set(log_callback, user_data);
g_logger_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
g_logger_state.log_callback_user_data = user_data;
}
static void llama_log_internal_v(ggml_log_level level, const char * format, va_list args) {
@ -21868,12 +21856,12 @@ static void llama_log_internal_v(ggml_log_level level, const char * format, va_l
char buffer[128];
int len = vsnprintf(buffer, 128, format, args);
if (len < 128) {
g_state.log_callback(level, buffer, g_state.log_callback_user_data);
g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
} else {
char * buffer2 = new char[len + 1];
vsnprintf(buffer2, len + 1, format, args_copy);
buffer2[len] = 0;
g_state.log_callback(level, buffer2, g_state.log_callback_user_data);
g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
delete[] buffer2;
}
va_end(args_copy);