move llama_client_slot back to server.cpp
This commit is contained in:
ngxson
parent
85c0334084
commit
72a8d59d48
2 changed files with 291 additions and 316 deletions
|
|
@ -144,6 +144,297 @@ static json probs_vector_to_json(const llama_context *ctx, const std::vector<com
|
|||
return out;
|
||||
}
|
||||
|
||||
struct llama_client_slot
|
||||
{
|
||||
int id;
|
||||
int task_id = -1;
|
||||
|
||||
struct slot_params params;
|
||||
|
||||
slot_state state = IDLE;
|
||||
slot_command command = NONE;
|
||||
|
||||
// used to determine the slot that has been used the longest
|
||||
int64_t t_last_used = -1;
|
||||
|
||||
// generation props
|
||||
int32_t n_ctx = 0; // context size per slot
|
||||
int32_t n_past = 0;
|
||||
int32_t n_decoded = 0;
|
||||
int32_t n_remaining = -1;
|
||||
int32_t i_batch = -1;
|
||||
int32_t n_predict = -1;
|
||||
|
||||
int32_t num_prompt_tokens = 0;
|
||||
int32_t num_prompt_tokens_processed = 0;
|
||||
|
||||
json prompt;
|
||||
std::string generated_text;
|
||||
llama_token sampled;
|
||||
std::vector<llama_token> cache_tokens;
|
||||
std::vector<completion_token_output> generated_token_probs;
|
||||
|
||||
bool infill = false;
|
||||
bool embedding = false;
|
||||
bool has_next_token = true;
|
||||
bool truncated = false;
|
||||
bool stopped_eos = false;
|
||||
bool stopped_word = false;
|
||||
bool stopped_limit = false;
|
||||
|
||||
bool oaicompat = false;
|
||||
std::string oaicompat_model;
|
||||
|
||||
std::string stopping_word;
|
||||
|
||||
// sampling
|
||||
struct llama_sampling_params sparams;
|
||||
llama_sampling_context *ctx_sampling = nullptr;
|
||||
|
||||
int32_t ga_i = 0; // group-attention state
|
||||
int32_t ga_n = 1; // group-attention factor
|
||||
int32_t ga_w = 512; // group-attention width
|
||||
|
||||
int32_t n_past_se = 0; // self-extend
|
||||
|
||||
// multimodal
|
||||
std::vector<slot_image> images;
|
||||
|
||||
// stats
|
||||
size_t sent_count = 0;
|
||||
size_t sent_token_probs_index = 0;
|
||||
|
||||
int64_t t_start_process_prompt;
|
||||
int64_t t_start_genereration;
|
||||
|
||||
double t_prompt_processing; // ms
|
||||
double t_token_generation; // ms
|
||||
|
||||
// multitasks
|
||||
int multitask_id = -1;
|
||||
|
||||
void reset() {
|
||||
num_prompt_tokens = 0;
|
||||
generated_text = "";
|
||||
truncated = false;
|
||||
stopped_eos = false;
|
||||
stopped_word = false;
|
||||
stopped_limit = false;
|
||||
stopping_word = "";
|
||||
n_past = 0;
|
||||
sent_count = 0;
|
||||
sent_token_probs_index = 0;
|
||||
infill = false;
|
||||
ga_i = 0;
|
||||
n_past_se = 0;
|
||||
|
||||
generated_token_probs.clear();
|
||||
|
||||
for (slot_image & img : images)
|
||||
{
|
||||
free(img.image_embedding);
|
||||
if (img.img_data) {
|
||||
clip_image_u8_free(img.img_data);
|
||||
}
|
||||
img.prefix_prompt = "";
|
||||
}
|
||||
|
||||
images.clear();
|
||||
}
|
||||
|
||||
bool has_budget(gpt_params &global_params) {
|
||||
if (params.n_predict == -1 && global_params.n_predict == -1)
|
||||
{
|
||||
return true; // limitless
|
||||
}
|
||||
|
||||
n_remaining = -1;
|
||||
|
||||
if (params.n_predict != -1)
|
||||
{
|
||||
n_remaining = params.n_predict - n_decoded;
|
||||
}
|
||||
else if (global_params.n_predict != -1)
|
||||
{
|
||||
n_remaining = global_params.n_predict - n_decoded;
|
||||
}
|
||||
|
||||
return n_remaining > 0; // no budget
|
||||
}
|
||||
|
||||
bool available() const {
|
||||
return state == IDLE && command == NONE;
|
||||
}
|
||||
|
||||
bool is_processing() const {
|
||||
return (state == IDLE && command == LOAD_PROMPT) || state == PROCESSING;
|
||||
}
|
||||
|
||||
void add_token_string(const completion_token_output &token) {
|
||||
if (command == RELEASE)
|
||||
{
|
||||
return;
|
||||
}
|
||||
cache_tokens.push_back(token.tok);
|
||||
generated_token_probs.push_back(token);
|
||||
}
|
||||
|
||||
void release() {
|
||||
if (state == PROCESSING)
|
||||
{
|
||||
t_token_generation = (ggml_time_us() - t_start_genereration) / 1e3;
|
||||
command = RELEASE;
|
||||
}
|
||||
}
|
||||
|
||||
json get_formated_timings() {
|
||||
return json
|
||||
{
|
||||
{"prompt_n", num_prompt_tokens_processed},
|
||||
{"prompt_ms", t_prompt_processing},
|
||||
{"prompt_per_token_ms", t_prompt_processing / num_prompt_tokens_processed},
|
||||
{"prompt_per_second", 1e3 / t_prompt_processing * num_prompt_tokens_processed},
|
||||
|
||||
{"predicted_n", n_decoded},
|
||||
{"predicted_ms", t_token_generation},
|
||||
{"predicted_per_token_ms", t_token_generation / n_decoded},
|
||||
{"predicted_per_second", 1e3 / t_token_generation * n_decoded},
|
||||
};
|
||||
}
|
||||
|
||||
void print_timings() const {
|
||||
char buffer[512];
|
||||
double t_token = t_prompt_processing / num_prompt_tokens_processed;
|
||||
double n_tokens_second = 1e3 / t_prompt_processing * num_prompt_tokens_processed;
|
||||
sprintf(buffer, "prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)",
|
||||
t_prompt_processing, num_prompt_tokens_processed,
|
||||
t_token, n_tokens_second);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_prompt_processing", t_prompt_processing},
|
||||
{"num_prompt_tokens_processed", num_prompt_tokens_processed},
|
||||
{"t_token", t_token},
|
||||
{"n_tokens_second", n_tokens_second},
|
||||
});
|
||||
|
||||
t_token = t_token_generation / n_decoded;
|
||||
n_tokens_second = 1e3 / t_token_generation * n_decoded;
|
||||
sprintf(buffer, "generation eval time = %10.2f ms / %5d runs (%8.2f ms per token, %8.2f tokens per second)",
|
||||
t_token_generation, n_decoded,
|
||||
t_token, n_tokens_second);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_token_generation", t_token_generation},
|
||||
{"n_decoded", n_decoded},
|
||||
{"t_token", t_token},
|
||||
{"n_tokens_second", n_tokens_second},
|
||||
});
|
||||
|
||||
sprintf(buffer, " total time = %10.2f ms", t_prompt_processing + t_token_generation);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_prompt_processing", t_prompt_processing},
|
||||
{"t_token_generation", t_token_generation},
|
||||
{"t_total", t_prompt_processing + t_token_generation},
|
||||
});
|
||||
}
|
||||
|
||||
// context extension via Self-Extend
|
||||
void grp_attn_update_params() {
|
||||
int grpa_i = 0;
|
||||
// copy to local variables
|
||||
int32_t grpa_n = ga_n;
|
||||
int32_t grpa_w = ga_w;
|
||||
int32_t slot_npast = 0;
|
||||
for (int k = 0; k < n_past; ++k)
|
||||
{
|
||||
while (slot_npast >= grpa_i + grpa_w) {
|
||||
const int bd = (grpa_w/grpa_n)*(grpa_n - 1);
|
||||
slot_npast -= bd;
|
||||
grpa_i += grpa_w/grpa_n;
|
||||
}
|
||||
slot_npast++;
|
||||
}
|
||||
n_past_se = slot_npast;
|
||||
ga_i = grpa_i;
|
||||
}
|
||||
|
||||
int32_t grp_attn_calc_npast() {
|
||||
int32_t slot_npast = n_past_se > 0 ? n_past_se : n_past;
|
||||
// copy to local variables
|
||||
int32_t grpa_i = ga_i;
|
||||
int32_t grpa_n = ga_n;
|
||||
int32_t grpa_w = ga_w;
|
||||
while (slot_npast >= grpa_i + grpa_w) {
|
||||
const int bd = (grpa_w/grpa_n)*(grpa_n - 1);
|
||||
slot_npast -= bd;
|
||||
grpa_i += grpa_w/grpa_n;
|
||||
}
|
||||
return slot_npast;
|
||||
}
|
||||
|
||||
void grp_attn_shift(llama_context * ctx, const int32_t n_tokens) {
|
||||
while (n_past_se >= ga_i + ga_w)
|
||||
{
|
||||
const int ib = (ga_n * ga_i) / ga_w;
|
||||
const int bd = (ga_w / ga_n) * (ga_n - 1);
|
||||
const int dd = (ga_w / ga_n) - ib * bd - ga_w;
|
||||
|
||||
LOG_TEE("\n");
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i, n_past_se, ib * bd, ga_i + ib * bd, n_past_se + ib * bd);
|
||||
LOG_TEE("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib * bd, ga_i + ib * bd + ga_w, ga_n, (ga_i + ib * bd) / ga_n, (ga_i + ib * bd + ga_w) / ga_n);
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib * bd + ga_w, n_past_se + ib * bd, dd, ga_i + ib * bd + ga_w + dd, n_past_se + ib * bd + dd);
|
||||
|
||||
llama_kv_cache_seq_shift(ctx, id, ga_i, n_past_se, ib * bd);
|
||||
llama_kv_cache_seq_div(ctx, id, ga_i + ib * bd, ga_i + ib * bd + ga_w,ga_n);
|
||||
llama_kv_cache_seq_shift(ctx, id, ga_i + ib * bd + ga_w,n_past_se + ib * bd, dd);
|
||||
|
||||
n_past_se -= bd;
|
||||
|
||||
ga_i += ga_w / ga_n;
|
||||
|
||||
LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", n_past_se + bd, n_past_se, ga_i);
|
||||
}
|
||||
n_past_se += n_tokens;
|
||||
}
|
||||
};
|
||||
|
||||
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;
|
||||
|
|
@ -1683,34 +1974,9 @@ struct llama_server_context
|
|||
if (slot.ga_n != 1)
|
||||
{
|
||||
// context extension via Self-Extend
|
||||
<<<<<<< HEAD
|
||||
// TODO @ngxson: What happen if we're retrying with smaller n_batch?
|
||||
// By the second time we retry, "grp_attn_shift" has already been called
|
||||
slot.grp_attn_shift(ctx, n_tokens);
|
||||
=======
|
||||
while (slot.n_past_se >= slot.ga_i + slot.ga_w)
|
||||
{
|
||||
const int ib = (slot.ga_n * slot.ga_i) / slot.ga_w;
|
||||
const int bd = (slot.ga_w / slot.ga_n) * (slot.ga_n - 1);
|
||||
const int dd = (slot.ga_w / slot.ga_n) - ib * bd - slot.ga_w;
|
||||
|
||||
LOG_TEE("\n");
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
|
||||
LOG_TEE("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
|
||||
|
||||
llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
|
||||
llama_kv_cache_seq_div(ctx, slot.id, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w,slot.ga_n);
|
||||
llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i + ib * bd + slot.ga_w,slot.n_past_se + ib * bd, dd);
|
||||
|
||||
slot.n_past_se -= bd;
|
||||
|
||||
slot.ga_i += slot.ga_w / slot.ga_n;
|
||||
|
||||
LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
|
||||
}
|
||||
slot.n_past_se += n_tokens;
|
||||
>>>>>>> master
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -174,297 +174,6 @@ struct completion_token_output
|
|||
std::string text_to_send;
|
||||
};
|
||||
|
||||
struct llama_client_slot
|
||||
{
|
||||
int id;
|
||||
int task_id = -1;
|
||||
|
||||
struct slot_params params;
|
||||
|
||||
slot_state state = IDLE;
|
||||
slot_command command = NONE;
|
||||
|
||||
// used to determine the slot that has been used the longest
|
||||
int64_t t_last_used = -1;
|
||||
|
||||
// generation props
|
||||
int32_t n_ctx = 0; // context size per slot
|
||||
int32_t n_past = 0;
|
||||
int32_t n_decoded = 0;
|
||||
int32_t n_remaining = -1;
|
||||
int32_t i_batch = -1;
|
||||
int32_t n_predict = -1;
|
||||
|
||||
int32_t num_prompt_tokens = 0;
|
||||
int32_t num_prompt_tokens_processed = 0;
|
||||
|
||||
json prompt;
|
||||
std::string generated_text;
|
||||
llama_token sampled;
|
||||
std::vector<llama_token> cache_tokens;
|
||||
std::vector<completion_token_output> generated_token_probs;
|
||||
|
||||
bool infill = false;
|
||||
bool embedding = false;
|
||||
bool has_next_token = true;
|
||||
bool truncated = false;
|
||||
bool stopped_eos = false;
|
||||
bool stopped_word = false;
|
||||
bool stopped_limit = false;
|
||||
|
||||
bool oaicompat = false;
|
||||
std::string oaicompat_model;
|
||||
|
||||
std::string stopping_word;
|
||||
|
||||
// sampling
|
||||
struct llama_sampling_params sparams;
|
||||
llama_sampling_context *ctx_sampling = nullptr;
|
||||
|
||||
int32_t ga_i = 0; // group-attention state
|
||||
int32_t ga_n = 1; // group-attention factor
|
||||
int32_t ga_w = 512; // group-attention width
|
||||
|
||||
int32_t n_past_se = 0; // self-extend
|
||||
|
||||
// multimodal
|
||||
std::vector<slot_image> images;
|
||||
|
||||
// stats
|
||||
size_t sent_count = 0;
|
||||
size_t sent_token_probs_index = 0;
|
||||
|
||||
int64_t t_start_process_prompt;
|
||||
int64_t t_start_genereration;
|
||||
|
||||
double t_prompt_processing; // ms
|
||||
double t_token_generation; // ms
|
||||
|
||||
// multitasks
|
||||
int multitask_id = -1;
|
||||
|
||||
void reset() {
|
||||
num_prompt_tokens = 0;
|
||||
generated_text = "";
|
||||
truncated = false;
|
||||
stopped_eos = false;
|
||||
stopped_word = false;
|
||||
stopped_limit = false;
|
||||
stopping_word = "";
|
||||
n_past = 0;
|
||||
sent_count = 0;
|
||||
sent_token_probs_index = 0;
|
||||
infill = false;
|
||||
ga_i = 0;
|
||||
n_past_se = 0;
|
||||
|
||||
generated_token_probs.clear();
|
||||
|
||||
for (slot_image & img : images)
|
||||
{
|
||||
free(img.image_embedding);
|
||||
if (img.img_data) {
|
||||
clip_image_u8_free(img.img_data);
|
||||
}
|
||||
img.prefix_prompt = "";
|
||||
}
|
||||
|
||||
images.clear();
|
||||
}
|
||||
|
||||
bool has_budget(gpt_params &global_params) {
|
||||
if (params.n_predict == -1 && global_params.n_predict == -1)
|
||||
{
|
||||
return true; // limitless
|
||||
}
|
||||
|
||||
n_remaining = -1;
|
||||
|
||||
if (params.n_predict != -1)
|
||||
{
|
||||
n_remaining = params.n_predict - n_decoded;
|
||||
}
|
||||
else if (global_params.n_predict != -1)
|
||||
{
|
||||
n_remaining = global_params.n_predict - n_decoded;
|
||||
}
|
||||
|
||||
return n_remaining > 0; // no budget
|
||||
}
|
||||
|
||||
bool available() const {
|
||||
return state == IDLE && command == NONE;
|
||||
}
|
||||
|
||||
bool is_processing() const {
|
||||
return (state == IDLE && command == LOAD_PROMPT) || state == PROCESSING;
|
||||
}
|
||||
|
||||
void add_token_string(const completion_token_output &token) {
|
||||
if (command == RELEASE)
|
||||
{
|
||||
return;
|
||||
}
|
||||
cache_tokens.push_back(token.tok);
|
||||
generated_token_probs.push_back(token);
|
||||
}
|
||||
|
||||
void release() {
|
||||
if (state == PROCESSING)
|
||||
{
|
||||
t_token_generation = (ggml_time_us() - t_start_genereration) / 1e3;
|
||||
command = RELEASE;
|
||||
}
|
||||
}
|
||||
|
||||
json get_formated_timings() {
|
||||
return json
|
||||
{
|
||||
{"prompt_n", num_prompt_tokens_processed},
|
||||
{"prompt_ms", t_prompt_processing},
|
||||
{"prompt_per_token_ms", t_prompt_processing / num_prompt_tokens_processed},
|
||||
{"prompt_per_second", 1e3 / t_prompt_processing * num_prompt_tokens_processed},
|
||||
|
||||
{"predicted_n", n_decoded},
|
||||
{"predicted_ms", t_token_generation},
|
||||
{"predicted_per_token_ms", t_token_generation / n_decoded},
|
||||
{"predicted_per_second", 1e3 / t_token_generation * n_decoded},
|
||||
};
|
||||
}
|
||||
|
||||
void print_timings() const {
|
||||
char buffer[512];
|
||||
double t_token = t_prompt_processing / num_prompt_tokens_processed;
|
||||
double n_tokens_second = 1e3 / t_prompt_processing * num_prompt_tokens_processed;
|
||||
sprintf(buffer, "prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)",
|
||||
t_prompt_processing, num_prompt_tokens_processed,
|
||||
t_token, n_tokens_second);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_prompt_processing", t_prompt_processing},
|
||||
{"num_prompt_tokens_processed", num_prompt_tokens_processed},
|
||||
{"t_token", t_token},
|
||||
{"n_tokens_second", n_tokens_second},
|
||||
});
|
||||
|
||||
t_token = t_token_generation / n_decoded;
|
||||
n_tokens_second = 1e3 / t_token_generation * n_decoded;
|
||||
sprintf(buffer, "generation eval time = %10.2f ms / %5d runs (%8.2f ms per token, %8.2f tokens per second)",
|
||||
t_token_generation, n_decoded,
|
||||
t_token, n_tokens_second);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_token_generation", t_token_generation},
|
||||
{"n_decoded", n_decoded},
|
||||
{"t_token", t_token},
|
||||
{"n_tokens_second", n_tokens_second},
|
||||
});
|
||||
|
||||
sprintf(buffer, " total time = %10.2f ms", t_prompt_processing + t_token_generation);
|
||||
LOG_INFO(buffer, {
|
||||
{"slot_id", id},
|
||||
{"task_id", task_id},
|
||||
{"t_prompt_processing", t_prompt_processing},
|
||||
{"t_token_generation", t_token_generation},
|
||||
{"t_total", t_prompt_processing + t_token_generation},
|
||||
});
|
||||
}
|
||||
|
||||
// context extension via Self-Extend
|
||||
void grp_attn_update_params() {
|
||||
int grpa_i = 0;
|
||||
// copy to local variables
|
||||
int32_t grpa_n = ga_n;
|
||||
int32_t grpa_w = ga_w;
|
||||
int32_t slot_npast = 0;
|
||||
for (int k = 0; k < n_past; ++k)
|
||||
{
|
||||
while (slot_npast >= grpa_i + grpa_w) {
|
||||
const int bd = (grpa_w/grpa_n)*(grpa_n - 1);
|
||||
slot_npast -= bd;
|
||||
grpa_i += grpa_w/grpa_n;
|
||||
}
|
||||
slot_npast++;
|
||||
}
|
||||
n_past_se = slot_npast;
|
||||
ga_i = grpa_i;
|
||||
}
|
||||
|
||||
int32_t grp_attn_calc_npast() {
|
||||
int32_t slot_npast = n_past_se > 0 ? n_past_se : n_past;
|
||||
// copy to local variables
|
||||
int32_t grpa_i = ga_i;
|
||||
int32_t grpa_n = ga_n;
|
||||
int32_t grpa_w = ga_w;
|
||||
while (slot_npast >= grpa_i + grpa_w) {
|
||||
const int bd = (grpa_w/grpa_n)*(grpa_n - 1);
|
||||
slot_npast -= bd;
|
||||
grpa_i += grpa_w/grpa_n;
|
||||
}
|
||||
return slot_npast;
|
||||
}
|
||||
|
||||
void grp_attn_shift(llama_context * ctx, const int32_t n_tokens) {
|
||||
while (n_past_se >= ga_i + ga_w)
|
||||
{
|
||||
const int ib = (ga_n * ga_i) / ga_w;
|
||||
const int bd = (ga_w / ga_n) * (ga_n - 1);
|
||||
const int dd = (ga_w / ga_n) - ib * bd - ga_w;
|
||||
|
||||
LOG_TEE("\n");
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i, n_past_se, ib * bd, ga_i + ib * bd, n_past_se + ib * bd);
|
||||
LOG_TEE("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib * bd, ga_i + ib * bd + ga_w, ga_n, (ga_i + ib * bd) / ga_n, (ga_i + ib * bd + ga_w) / ga_n);
|
||||
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib * bd + ga_w, n_past_se + ib * bd, dd, ga_i + ib * bd + ga_w + dd, n_past_se + ib * bd + dd);
|
||||
|
||||
llama_kv_cache_seq_shift(ctx, id, ga_i, n_past_se, ib * bd);
|
||||
llama_kv_cache_seq_div(ctx, id, ga_i + ib * bd, ga_i + ib * bd + ga_w,ga_n);
|
||||
llama_kv_cache_seq_shift(ctx, id, ga_i + ib * bd + ga_w,n_past_se + ib * bd, dd);
|
||||
|
||||
n_past_se -= bd;
|
||||
|
||||
ga_i += ga_w / ga_n;
|
||||
|
||||
LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", n_past_se + bd, n_past_se, ga_i);
|
||||
}
|
||||
n_past_se += n_tokens;
|
||||
}
|
||||
};
|
||||
|
||||
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;
|
||||
}
|
||||
};
|
||||
|
||||
//
|
||||
// server utils
|
||||
//
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue