Update main.cpp to use new llama library

This commit is contained in:
Thomas Antony 2023-03-14 00:37:44 -07:00
parent b14486e1c0
commit 4b4d8a5d44

205
main.cpp
View file

@ -55,25 +55,6 @@ void sigint_handler(int signo) {
}
#endif
const char * llama_print_system_info(void) {
static std::string s;
s = "";
s += "AVX = " + std::to_string(ggml_cpu_has_avx()) + " | ";
s += "AVX2 = " + std::to_string(ggml_cpu_has_avx2()) + " | ";
s += "AVX512 = " + std::to_string(ggml_cpu_has_avx512()) + " | ";
s += "FMA = " + std::to_string(ggml_cpu_has_fma()) + " | ";
s += "NEON = " + std::to_string(ggml_cpu_has_neon()) + " | ";
s += "ARM_FMA = " + std::to_string(ggml_cpu_has_arm_fma()) + " | ";
s += "F16C = " + std::to_string(ggml_cpu_has_f16c()) + " | ";
s += "FP16_VA = " + std::to_string(ggml_cpu_has_fp16_va()) + " | ";
s += "WASM_SIMD = " + std::to_string(ggml_cpu_has_wasm_simd()) + " | ";
s += "BLAS = " + std::to_string(ggml_cpu_has_blas()) + " | ";
s += "SSE3 = " + std::to_string(ggml_cpu_has_sse3()) + " | ";
s += "VSX = " + std::to_string(ggml_cpu_has_vsx()) + " | ";
return s.c_str();
}
int main(int argc, char ** argv) {
ggml_time_init();
@ -107,41 +88,18 @@ int main(int argc, char ** argv) {
int64_t t_load_us = 0;
gpt_vocab vocab;
llama_model model;
// load the model
{
const ggml_type memory_type = params.memory_f16 ? GGML_TYPE_F16 : GGML_TYPE_F32;
const int64_t t_start_us = ggml_time_us();
if (!llama_model_load(params.model, model, vocab, params.n_ctx, memory_type)) {
fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, params.model.c_str());
return 1;
}
t_load_us = ggml_time_us() - t_start_us;
}
llama_context* ctx_ptr = llama_init_from_params(params);
llama_context & ctx = *ctx_ptr;
gpt_vocab & vocab = llama_context_get_vocab(ctx);
// print system information
{
fprintf(stderr, "\n");
fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",
params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());
}
int n_past = 0;
int64_t t_sample_us = 0;
int64_t t_predict_us = 0;
std::vector<float> logits;
llama_print_context_info(ctx);
// Add a space in front of the first character to match OG llama tokenizer behavior
params.prompt.insert(0, 1, ' ');
// tokenize the prompt
std::vector<gpt_vocab::id> embd_inp = ::llama_tokenize(vocab, params.prompt, true);
params.n_predict = std::min(params.n_predict, model.hparams.n_ctx - (int) embd_inp.size());
std::vector<gpt_vocab::id> embd_inp = llama_tokenize_text(ctx, params.prompt);
// prefix & suffix for instruct mode
const std::vector<gpt_vocab::id> inp_pfx = ::llama_tokenize(vocab, "\n\n### Instruction:\n\n", true);
@ -154,24 +112,8 @@ int main(int argc, char ** argv) {
}
// tokenize the reverse prompt
std::vector<std::vector<gpt_vocab::id>> antipromptv_inp;
for (auto antiprompt : params.antiprompt) {
antipromptv_inp.push_back(::llama_tokenize(vocab, antiprompt, false));
}
std::vector<gpt_vocab::id> antiprompt_inp = llama_tokenize_text(ctx, params.prompt);
// enable interactive mode if reverse prompt is specified
if (!antipromptv_inp.size()) {
params.interactive = true;
}
fprintf(stderr, "\n");
fprintf(stderr, "%s: prompt: '%s'\n", __func__, params.prompt.c_str());
fprintf(stderr, "%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size());
for (int i = 0; i < (int) embd_inp.size(); i++) {
fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], vocab.id_to_token.at(embd_inp[i]).c_str());
}
fprintf(stderr, "\n");
if (params.interactive) {
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
struct sigaction sigint_action;
@ -200,16 +142,6 @@ int main(int argc, char ** argv) {
fprintf(stderr, "sampling parameters: temp = %f, top_k = %d, top_p = %f, repeat_last_n = %i, repeat_penalty = %f\n", params.temp, params.top_k, params.top_p, params.repeat_last_n, params.repeat_penalty);
fprintf(stderr, "\n\n");
std::vector<gpt_vocab::id> embd;
// determine the required inference memory per token:
size_t mem_per_token = 0;
llama_eval(model, params.n_threads, 0, { 0, 1, 2, 3 }, logits, mem_per_token);
int last_n_size = params.repeat_last_n;
std::vector<gpt_vocab::id> last_n_tokens(last_n_size);
std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0);
if (params.interactive) {
fprintf(stderr, "== Running in interactive mode. ==\n"
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
@ -220,7 +152,6 @@ int main(int argc, char ** argv) {
is_interacting = true;
}
int input_consumed = 0;
bool input_noecho = false;
int remaining_tokens = params.n_predict;
@ -230,75 +161,33 @@ int main(int argc, char ** argv) {
printf(ANSI_COLOR_YELLOW);
}
while (remaining_tokens > 0 || params.interactive) {
// predict
if (embd.size() > 0) {
const int64_t t_start_us = ggml_time_us();
if(!llama_injest_input(ctx, params.prompt))
{
fprintf(stderr, "Failed to injest prompt\n");
return 1;
};
if (!llama_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) {
fprintf(stderr, "Failed to predict\n");
return 1;
}
t_predict_us += ggml_time_us() - t_start_us;
}
n_past += embd.size();
embd.clear();
if (embd_inp.size() <= input_consumed) {
// out of user input, sample next token
const float top_k = params.top_k;
const float top_p = params.top_p;
const float temp = params.temp;
const float repeat_penalty = params.repeat_penalty;
const int n_vocab = model.hparams.n_vocab;
gpt_vocab::id id = 0;
{
const int64_t t_start_sample_us = ggml_time_us();
if (params.ignore_eos) {
// set the logit of the eos token to zero to avoid sampling it
logits[logits.size() - n_vocab + EOS_TOKEN_ID] = 0;
}
id = llama_sample_top_p_top_k(vocab, logits.data() + (logits.size() - n_vocab), last_n_tokens, repeat_penalty, top_k, top_p, temp, rng);
last_n_tokens.erase(last_n_tokens.begin());
last_n_tokens.push_back(id);
t_sample_us += ggml_time_us() - t_start_sample_us;
}
// add it to the context
embd.push_back(id);
// echo this to console
input_noecho = false;
// decrement remaining sampling budget
--remaining_tokens;
} else {
// some user input remains from prompt or interaction, forward it to processing
while (embd_inp.size() > input_consumed) {
embd.push_back(embd_inp[input_consumed]);
last_n_tokens.erase(last_n_tokens.begin());
last_n_tokens.push_back(embd_inp[input_consumed]);
++input_consumed;
if ((int) embd.size() >= params.n_batch) {
break;
}
}
}
// display text
if (!input_noecho) {
// display text
input_noecho = false;
const std::vector<gpt_vocab::id>& embd = llama_context_get_embd(ctx);
if (!input_noecho) {
for (auto id : embd) {
printf("%s", vocab.id_to_token[id].c_str());
}
printf("%s", vocab.id_to_token[id].c_str());
}
fflush(stdout);
}
if (!input_noecho && params.use_color) {
printf(ANSI_COLOR_RESET);
}
const std::vector<gpt_vocab::id>& last_n_tokens = llama_context_get_last_n_tokens(ctx);
while (llama_context_not_finished(ctx) > 0) {
gpt_vocab::id model_output = 0;
bool response = llama_inference(ctx, model_output);
if (response) {
printf("%s", vocab.id_to_token[model_output].c_str());
fflush(stdout);
}
// reset color to default if we there is no pending user input
@ -306,9 +195,10 @@ int main(int argc, char ** argv) {
printf(ANSI_COLOR_RESET);
}
// in interactive mode, and not currently processing queued inputs;
// check if we should prompt the user for more
if (params.interactive && embd_inp.size() <= input_consumed) {
if (params.interactive) {
// check for reverse prompt
for (auto antiprompt_inp : antipromptv_inp) {
if (antiprompt_inp.size() && std::equal(antiprompt_inp.rbegin(), antiprompt_inp.rend(), last_n_tokens.rbegin())) {
@ -337,15 +227,8 @@ int main(int argc, char ** argv) {
} else {
line.pop_back(); // Remove the continue character
}
buffer += line + '\n'; // Append the line to the result
} while (another_line);
if (params.use_color) printf(ANSI_COLOR_RESET);
std::vector<gpt_vocab::id> line_inp = ::llama_tokenize(vocab, buffer, false);
embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end());
if (params.instruct) {
embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
// Do not clear existing context in interactive mode
llama_init_context_with_prompt(ctx, buf, false);
}
remaining_tokens -= line_inp.size();
@ -371,24 +254,14 @@ int main(int argc, char ** argv) {
is_interacting = true;
}
}
#if defined (_WIN32)
signal(SIGINT, SIG_DFL);
#endif
// report timing
// report timing from context
{
const int64_t t_main_end_us = ggml_time_us();
fprintf(stderr, "\n\n");
fprintf(stderr, "%s: mem per token = %8zu bytes\n", __func__, mem_per_token);
fprintf(stderr, "%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f);
fprintf(stderr, "%s: sample time = %8.2f ms\n", __func__, t_sample_us/1000.0f);
fprintf(stderr, "%s: predict time = %8.2f ms / %.2f ms per token\n", __func__, t_predict_us/1000.0f, t_predict_us/1000.0f/n_past);
llama_print_end_stats(ctx);
fprintf(stderr, "%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f);
}
ggml_free(model.ctx);
llama_free_context(ctx_ptr);
if (params.use_color) {
printf(ANSI_COLOR_RESET);