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

log file for debug output

Browse source
This commit is contained in:
Maël Kerbiriou 2023-03-20 12:15:01 +01:00
parent aa6c2bd5d2
commit e66962ff63
3 changed files with 23 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

13
main.cpp
View file

@ -803,6 +803,7 @@ const char * llama_print_system_info(void) {
int main(int argc, char ** argv) { int main(int argc, char ** argv) {
ggml_time_init(); ggml_time_init();
const int64_t t_main_start_us = ggml_time_us(); const int64_t t_main_start_us = ggml_time_us();
log_file = fopen("out.log", "w");
gpt_params params; gpt_params params;
params.model = "models/llama-7B/ggml-model.bin"; params.model = "models/llama-7B/ggml-model.bin";
@ -960,6 +961,15 @@ int main(int argc, char ** argv) {
if (embd.size() > 0) { if (embd.size() > 0) {
const int64_t t_start_us = ggml_time_us(); const int64_t t_start_us = ggml_time_us();
if(log_file) {
std::string intokdbg = vocab.id_to_token.at(embd[0]);
for(int i = 1; i < embd.size(); i++) {
intokdbg += '|';
intokdbg += vocab.id_to_token.at(embd[i]);
}
logprintf("\nin:'%s' n_past=%d, remaining_tokens=%d, embd.size()=%zu, embd_inp.size()=%zu\n",
intokdbg.c_str(), n_past, remaining_tokens, embd.size(), embd_inp.size());
}
if (!llama_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) { if (!llama_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) {
fprintf(stderr, "Failed to predict\n"); fprintf(stderr, "Failed to predict\n");
return 1; return 1;
@ -1079,6 +1089,7 @@ int main(int argc, char ** argv) {
} }
is_interacting = false; is_interacting = false;
} }
if (log_file) fflush(log_file);
// end of text token // end of text token
if (embd.back() == EOS_TOKEN_ID) { if (embd.back() == EOS_TOKEN_ID) {
@ -1096,6 +1107,7 @@ int main(int argc, char ** argv) {
is_interacting = true; is_interacting = true;
} }
} }
logprintf("exit: remaining_tokens=%d n_past=%d goal=%lu n_predict=%d\n", remaining_tokens, n_past, embd_inp.size() + params.n_predict, params.n_predict);
#if defined (_WIN32) #if defined (_WIN32)
signal(SIGINT, SIG_DFL); signal(SIGINT, SIG_DFL);
@ -1114,6 +1126,7 @@ int main(int argc, char ** argv) {
} }
ggml_free(model.ctx); ggml_free(model.ctx);
if (log_file) fclose(log_file);
if (params.use_color) { if (params.use_color) {
printf(ANSI_COLOR_RESET); printf(ANSI_COLOR_RESET);

4
utils.cpp
View file

@ -16,6 +16,8 @@
#include <alloca.h> #include <alloca.h>
#endif #endif
FILE * log_file = NULL;
bool gpt_params_parse(int argc, char ** argv, gpt_params & params) { bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
// determine sensible default number of threads. // determine sensible default number of threads.
// std::thread::hardware_concurrency may not be equal to the number of cores, or may return 0. // std::thread::hardware_concurrency may not be equal to the number of cores, or may return 0.
@ -576,6 +578,7 @@ struct SoftMaxSampler {
break; break;
} }
} }
logprintf("%s: n: %d sum: %f\n", __func__, n, cumsum);
// discrete_distribution renormalizes the subset of probabilities to sum to 1.0 // discrete_distribution renormalizes the subset of probabilities to sum to 1.0
return std::discrete_distribution<>(probs.begin(), probs.begin() + n); return std::discrete_distribution<>(probs.begin(), probs.begin() + n);
@ -633,6 +636,7 @@ gpt_vocab::id sample_top_k_top_p(
auto dist = probs.top_p(top_p); auto dist = probs.top_p(top_p);
int sampled_tok_id = probs.sample(dist, rng); int sampled_tok_id = probs.sample(dist, rng);
probs.print(log_file, vocab, logits, 16, sampled_tok_id);
return sampled_tok_id; return sampled_tok_id;
} }

6
utils.h
View file

@ -8,6 +8,12 @@
#include <random> #include <random>
#include <thread> #include <thread>
#include <stdio.h>
extern FILE * log_file;
#define logprintf(...) { if (log_file) fprintf(log_file, __VA_ARGS__); }
// //
// CLI argument parsing // CLI argument parsing
// //