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

server: metrics - move to a dedicated struct

Browse source
This commit is contained in:
Pierrick HYMBERT 2024-02-25 12:44:39 +01:00
parent 7b29648da5
commit 542f42a604
1 changed files with 43 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

79
examples/server/server.cpp
View file

@ -311,6 +311,39 @@ struct llama_client_slot
}
};
struct llama_metrics {
uint64_t n_prompt_tokens_processed_total = 0;
uint64_t n_tokens_predicted_total = 0;
uint64_t n_prompt_tokens_processed = 0;
uint64_t t_prompt_processing = 0;
uint64_t n_tokens_predicted = 0;
uint64_t t_tokens_generation = 0;
void on_prompt_eval(const llama_client_slot &slot) {
n_prompt_tokens_processed_total += slot.num_prompt_tokens_processed;
n_prompt_tokens_processed += slot.num_prompt_tokens_processed;
t_prompt_processing += slot.t_prompt_processing;
}
void on_prediction(const llama_client_slot &slot) {
n_tokens_predicted_total += slot.n_decoded;
n_tokens_predicted += slot.n_decoded;
t_tokens_generation += slot.t_token_generation;
}
void reset_bucket() {
n_prompt_tokens_processed = 0;
t_prompt_processing = 0;
n_tokens_predicted = 0;
t_tokens_generation = 0;
}
};
struct llama_server_context
{
llama_model *model = nullptr;
@ -345,15 +378,7 @@ struct llama_server_context
llama_server_queue queue_tasks;
llama_server_response queue_results;
// metrics
uint64_t n_prompt_tokens_processed_total = 0;
uint64_t n_tokens_predicted_total = 0;
uint64_t n_prompt_tokens_processed = 0;
uint64_t t_prompt_processing = 0;
uint64_t n_tokens_predicted = 0;
uint64_t t_tokens_generation = 0;
llama_metrics metrics;
~llama_server_context()
{
@ -1453,24 +1478,20 @@ struct llama_server_context
{ "processing", n_processing_slots },
{ "deferred", queue_tasks.queue_tasks_deferred.size() },
{ "n_prompt_tokens_processed_total", n_prompt_tokens_processed_total},
{ "n_tokens_predicted_total", n_tokens_predicted_total},
{ "n_prompt_tokens_processed_total", metrics.n_prompt_tokens_processed_total},
{ "n_tokens_predicted_total", metrics.n_tokens_predicted_total},
{ "n_prompt_tokens_processed", n_prompt_tokens_processed},
{ "t_prompt_processing", t_prompt_processing},
{ "n_tokens_predicted", n_tokens_predicted},
{ "t_tokens_generation", t_tokens_generation},
{ "n_prompt_tokens_processed", metrics.n_prompt_tokens_processed},
{ "t_prompt_processing", metrics.t_prompt_processing},
{ "n_tokens_predicted", metrics.n_tokens_predicted},
{ "t_tokens_generation", metrics.t_tokens_generation},
{ "kv_cache_tokens_count", llama_get_kv_cache_token_count(ctx)},
{ "kv_cache_used_cells", llama_get_kv_cache_used_cells(ctx)},
{ "slots", slots_data },
};
// reset metrics for the next bucket
n_prompt_tokens_processed = 0;
t_prompt_processing = 0;
n_tokens_predicted = 0;
t_tokens_generation = 0;
metrics.reset_bucket();
queue_results.send(res);
} break;
}
@ -1878,7 +1899,7 @@ struct llama_server_context
{
slot.t_start_genereration = ggml_time_us();
slot.t_prompt_processing = (slot.t_start_genereration - slot.t_start_process_prompt) / 1e3;
update_metrics_prompt_eval(slot);
metrics.on_prompt_eval(slot);
}
llama_token_data_array cur_p = { slot.ctx_sampling->cur.data(), slot.ctx_sampling->cur.size(), false };
@ -1901,7 +1922,7 @@ struct llama_server_context
slot.release();
slot.print_timings();
send_final_response(slot);
update_metrics_prediction(slot);
metrics.on_prediction(slot);
}
slot.i_batch = -1;
@ -1913,20 +1934,6 @@ struct llama_server_context
void run_on_all_tasks_finished() {
update_slots();
}
void update_metrics_prompt_eval(const llama_client_slot &slot) {
n_prompt_tokens_processed_total += slot.num_prompt_tokens_processed;
n_prompt_tokens_processed += slot.num_prompt_tokens_processed;
t_prompt_processing += slot.t_prompt_processing;
}
void update_metrics_prediction(const llama_client_slot &slot) {
n_tokens_predicted_total += slot.n_decoded;
n_tokens_predicted += slot.n_decoded;
t_tokens_generation += slot.t_token_generation;
}
};
static void server_print_usage(const char *argv0, const gpt_params &params,