llama.cpp/examples/main/main.cpp
Georgi Gerganov ecb217db4f
llama : Metal inference (#1642)
* mtl : export the LLaMA computation graph

* ci : disable temporary

* mtl : adapt the MNIST example as starter

* mtl : no need for mtl-export tool, add cli arg for main instead

* mtl : export just a small part of the graph for now to make it easier

* mtl : move MSL code into separate file for easy editing

* mtl : initial get_rows_q4_0 kernel

* mtl : confirmed get_rows_q4_0 is working correctly

* mtl : add rms_norm kernel + confirm working

* mtl : add mul kernel + confirm working

* mtl : initial mul_mat Q4 kernel (wrong results)

* mtl : mul_mat fixes (still wrong)

* mtl : another mul_mat Q4 (still does not work)

* mtl : working mul_mat q4

* ggml : fix handling of "view" ops in ggml_graph_import()

* mtl : add rope kernel

* mtl : add reshape and transpose handling

* ggml : store offset as opt arg for ggml_view_xd() operators

* mtl : add cpy kernel + handle view ops

* mtl : confirm f16 x f32 attention mul mat

* mtl : add scale kernel

* mtl : add diag_mask_inf kernel

* mtl : fix soft_max kernel

* ggml : update ggml_nbytes() to handle non-contiguous tensors

* mtl : verify V tensor contents

* mtl : add f32 -> f32 cpy kernel

* mtl : add silu kernel

* mtl : add non-broadcast mul kernel

* mtl : full GPU inference of the computation graph

* mtl : optimize rms_norm and soft_max kernels

* mtl : add f16 mat x f32 vec multiplication kernel

* mtl : fix bug in f16 x f32 mul mat + speed-up computation

* mtl : faster mul_mat_q4_0_f32 kernel

* mtl : fix kernel signature + roll inner loop

* mtl : more threads for rms_norm + better timing

* mtl : remove printfs from inner loop

* mtl : simplify implementation

* mtl : add save/load vocab to ggml file

* mtl : plug Metal inference into llama.cpp (very quick-n-dirty)

* mtl : make it work with main example

Lots of hacks but at least now it generates text

* mtl : preparing for merge

* mtl : clean-up ggml mtl interface + suport scratch / inplace

* mtl : remove temp / debug code

* metal : final refactoring and simplification

* Revert "ci : disable temporary"

This reverts commit 98c267fc77.

* metal : add comments

* metal : clean-up stuff, fix typos

* readme : add Metal instructions

* readme : add example for main
2023-06-04 23:34:30 +03:00

643 lines
25 KiB
C++

// Defines sigaction on msys:
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#include "common.h"
#include "llama.h"
#include "build-info.h"
#include <cassert>
#include <cinttypes>
#include <cmath>
#include <cstdio>
#include <cstring>
#include <ctime>
#include <fstream>
#include <iostream>
#include <string>
#include <vector>
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
#include <signal.h>
#include <unistd.h>
#elif defined (_WIN32)
#define WIN32_LEAN_AND_MEAN
#define NOMINMAX
#include <windows.h>
#include <signal.h>
#endif
static console_state con_st;
static llama_context ** g_ctx;
static bool is_interacting = false;
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
void sigint_handler(int signo) {
if (signo == SIGINT) {
if (!is_interacting) {
is_interacting=true;
} else {
console_cleanup(con_st);
printf("\n");
llama_print_timings(*g_ctx);
_exit(130);
}
}
}
#endif
int main(int argc, char ** argv) {
gpt_params params;
if (gpt_params_parse(argc, argv, params) == false) {
return 1;
}
// save choice to use color for later
// (note for later: this is a slightly awkward choice)
con_st.use_color = params.use_color;
con_st.multiline_input = params.multiline_input;
console_init(con_st);
atexit([]() { console_cleanup(con_st); });
if (params.perplexity) {
printf("\n************\n");
printf("%s: please use the 'perplexity' tool for perplexity calculations\n", __func__);
printf("************\n\n");
return 0;
}
if (params.embedding) {
printf("\n************\n");
printf("%s: please use the 'embedding' tool for embedding calculations\n", __func__);
printf("************\n\n");
return 0;
}
if (params.n_ctx > 2048) {
fprintf(stderr, "%s: warning: model does not support context sizes greater than 2048 tokens (%d specified);"
"expect poor results\n", __func__, params.n_ctx);
}
fprintf(stderr, "%s: build = %d (%s)\n", __func__, BUILD_NUMBER, BUILD_COMMIT);
if (params.seed < 0) {
params.seed = time(NULL);
}
fprintf(stderr, "%s: seed = %d\n", __func__, params.seed);
std::mt19937 rng(params.seed);
if (params.random_prompt) {
params.prompt = gpt_random_prompt(rng);
}
llama_init_backend();
llama_context * ctx;
g_ctx = &ctx;
// load the model and apply lora adapter, if any
ctx = llama_init_from_gpt_params(params);
if (ctx == NULL) {
fprintf(stderr, "%s: error: unable to load model\n", __func__);
return 1;
}
// 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());
}
// determine the maximum memory usage needed to do inference for the given n_batch and n_predict parameters
// uncomment the "used_mem" line in llama.cpp to see the results
if (params.mem_test) {
{
const std::vector<llama_token> tmp(params.n_batch, llama_token_bos());
llama_eval(ctx, tmp.data(), tmp.size(), 0, params.n_threads);
}
{
const std::vector<llama_token> tmp = { 0, };
llama_eval(ctx, tmp.data(), tmp.size(), params.n_predict - 1, params.n_threads);
}
llama_print_timings(ctx);
llama_free(ctx);
return 0;
}
// export the cgraph and exit
if (params.export_cgraph) {
llama_eval_export(ctx, "llama.ggml");
llama_free(ctx);
return 0;
}
std::string path_session = params.path_prompt_cache;
std::vector<llama_token> session_tokens;
if (!path_session.empty()) {
fprintf(stderr, "%s: attempting to load saved session from '%s'\n", __func__, path_session.c_str());
// fopen to check for existing session
FILE * fp = std::fopen(path_session.c_str(), "rb");
if (fp != NULL) {
std::fclose(fp);
session_tokens.resize(params.n_ctx);
size_t n_token_count_out = 0;
if (!llama_load_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.capacity(), &n_token_count_out)) {
fprintf(stderr, "%s: error: failed to load session file '%s'\n", __func__, path_session.c_str());
return 1;
}
session_tokens.resize(n_token_count_out);
llama_set_rng_seed(ctx, params.seed);
fprintf(stderr, "%s: loaded a session with prompt size of %d tokens\n", __func__, (int) session_tokens.size());
} else {
fprintf(stderr, "%s: session file does not exist, will create\n", __func__);
}
}
// tokenize the prompt
std::vector<llama_token> embd_inp;
if (params.interactive_first || params.instruct || !params.prompt.empty() || session_tokens.empty()) {
// Add a space in front of the first character to match OG llama tokenizer behavior
params.prompt.insert(0, 1, ' ');
embd_inp = ::llama_tokenize(ctx, params.prompt, true);
} else {
embd_inp = session_tokens;
}
const int n_ctx = llama_n_ctx(ctx);
if ((int) embd_inp.size() > n_ctx - 4) {
fprintf(stderr, "%s: error: prompt is too long (%d tokens, max %d)\n", __func__, (int) embd_inp.size(), n_ctx - 4);
return 1;
}
// debug message about similarity of saved session, if applicable
size_t n_matching_session_tokens = 0;
if (session_tokens.size()) {
for (llama_token id : session_tokens) {
if (n_matching_session_tokens >= embd_inp.size() || id != embd_inp[n_matching_session_tokens]) {
break;
}
n_matching_session_tokens++;
}
if (params.prompt.empty() && n_matching_session_tokens == embd_inp.size()) {
fprintf(stderr, "%s: using full prompt from session file\n", __func__);
} else if (n_matching_session_tokens >= embd_inp.size()) {
fprintf(stderr, "%s: session file has exact match for prompt!\n", __func__);
} else if (n_matching_session_tokens < (embd_inp.size() / 2)) {
fprintf(stderr, "%s: warning: session file has low similarity to prompt (%zu / %zu tokens); will mostly be reevaluated\n",
__func__, n_matching_session_tokens, embd_inp.size());
} else {
fprintf(stderr, "%s: session file matches %zu / %zu tokens of prompt\n",
__func__, n_matching_session_tokens, embd_inp.size());
}
}
// if we will use the cache for the full prompt without reaching the end of the cache, force
// reevaluation of the last token token to recalculate the cached logits
if (!embd_inp.empty() && n_matching_session_tokens == embd_inp.size() &&
session_tokens.size() > embd_inp.size()) {
session_tokens.resize(embd_inp.size() - 1);
}
// number of tokens to keep when resetting context
if (params.n_keep < 0 || params.n_keep > (int) embd_inp.size() || params.instruct) {
params.n_keep = (int)embd_inp.size();
}
// prefix & suffix for instruct mode
const auto inp_pfx = ::llama_tokenize(ctx, "\n\n### Instruction:\n\n", true);
const auto inp_sfx = ::llama_tokenize(ctx, "\n\n### Response:\n\n", false);
// in instruct mode, we inject a prefix and a suffix to each input by the user
if (params.instruct) {
params.interactive_first = true;
params.antiprompt.push_back("### Instruction:\n\n");
}
// enable interactive mode if interactive start is specified
if (params.interactive_first) {
params.interactive = true;
}
// determine newline token
auto llama_token_newline = ::llama_tokenize(ctx, "\n", false);
if (params.verbose_prompt) {
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], llama_token_to_str(ctx, embd_inp[i]));
}
if (params.n_keep > 0) {
fprintf(stderr, "%s: static prompt based on n_keep: '", __func__);
for (int i = 0; i < params.n_keep; i++) {
fprintf(stderr, "%s", llama_token_to_str(ctx, embd_inp[i]));
}
fprintf(stderr, "'\n");
}
fprintf(stderr, "\n");
}
if (params.interactive) {
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
struct sigaction sigint_action;
sigint_action.sa_handler = sigint_handler;
sigemptyset (&sigint_action.sa_mask);
sigint_action.sa_flags = 0;
sigaction(SIGINT, &sigint_action, NULL);
#elif defined (_WIN32)
auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false;
};
SetConsoleCtrlHandler(static_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
#endif
fprintf(stderr, "%s: interactive mode on.\n", __func__);
if (params.antiprompt.size()) {
for (auto antiprompt : params.antiprompt) {
fprintf(stderr, "Reverse prompt: '%s'\n", antiprompt.c_str());
}
}
if (!params.input_prefix.empty()) {
fprintf(stderr, "Input prefix: '%s'\n", params.input_prefix.c_str());
}
if (!params.input_suffix.empty()) {
fprintf(stderr, "Input suffix: '%s'\n", params.input_suffix.c_str());
}
}
fprintf(stderr, "sampling: repeat_last_n = %d, repeat_penalty = %f, presence_penalty = %f, frequency_penalty = %f, top_k = %d, tfs_z = %f, top_p = %f, typical_p = %f, temp = %f, mirostat = %d, mirostat_lr = %f, mirostat_ent = %f\n",
params.repeat_last_n, params.repeat_penalty, params.presence_penalty, params.frequency_penalty, params.top_k, params.tfs_z, params.top_p, params.typical_p, params.temp, params.mirostat, params.mirostat_eta, params.mirostat_tau);
fprintf(stderr, "generate: n_ctx = %d, n_batch = %d, n_predict = %d, n_keep = %d\n", n_ctx, params.n_batch, params.n_predict, params.n_keep);
fprintf(stderr, "\n\n");
// TODO: replace with ring-buffer
std::vector<llama_token> last_n_tokens(n_ctx);
std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0);
if (params.interactive) {
const char *control_message;
if (con_st.multiline_input) {
control_message = " - To return control to LLaMa, end your input with '\\'.\n"
" - To return control without starting a new line, end your input with '/'.\n";
} else {
control_message = " - Press Return to return control to LLaMa.\n"
" - To return control without starting a new line, end your input with '/'.\n"
" - If you want to submit another line, end your input with '\\'.\n";
}
fprintf(stderr, "== Running in interactive mode. ==\n"
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
" - Press Ctrl+C to interject at any time.\n"
#endif
"%s\n", control_message);
is_interacting = params.interactive_first;
}
bool is_antiprompt = false;
bool input_echo = true;
bool need_to_save_session = !path_session.empty() && n_matching_session_tokens < embd_inp.size();
int n_past = 0;
int n_remain = params.n_predict;
int n_consumed = 0;
int n_session_consumed = 0;
// the first thing we will do is to output the prompt, so set color accordingly
console_set_color(con_st, CONSOLE_COLOR_PROMPT);
std::vector<llama_token> embd;
while ((n_remain != 0 && !is_antiprompt) || params.interactive) {
// predict
if (embd.size() > 0) {
// infinite text generation via context swapping
// if we run out of context:
// - take the n_keep first tokens from the original prompt (via n_past)
// - take half of the last (n_ctx - n_keep) tokens and recompute the logits in batches
if (n_past + (int) embd.size() > n_ctx) {
const int n_left = n_past - params.n_keep;
// always keep the first token - BOS
n_past = std::max(1, params.n_keep);
// insert n_left/2 tokens at the start of embd from last_n_tokens
embd.insert(embd.begin(), last_n_tokens.begin() + n_ctx - n_left/2 - embd.size(), last_n_tokens.end() - embd.size());
// stop saving session if we run out of context
path_session.clear();
//printf("\n---\n");
//printf("resetting: '");
//for (int i = 0; i < (int) embd.size(); i++) {
// printf("%s", llama_token_to_str(ctx, embd[i]));
//}
//printf("'\n");
//printf("\n---\n");
}
// try to reuse a matching prefix from the loaded session instead of re-eval (via n_past)
if (n_session_consumed < (int) session_tokens.size()) {
size_t i = 0;
for ( ; i < embd.size(); i++) {
if (embd[i] != session_tokens[n_session_consumed]) {
session_tokens.resize(n_session_consumed);
break;
}
n_past++;
n_session_consumed++;
if (n_session_consumed >= (int) session_tokens.size()) {
++i;
break;
}
}
if (i > 0) {
embd.erase(embd.begin(), embd.begin() + i);
}
}
// evaluate tokens in batches
// embd is typically prepared beforehand to fit within a batch, but not always
for (int i = 0; i < (int) embd.size(); i += params.n_batch) {
int n_eval = (int) embd.size() - i;
if (n_eval > params.n_batch) {
n_eval = params.n_batch;
}
if (llama_eval(ctx, &embd[i], n_eval, n_past, params.n_threads)) {
fprintf(stderr, "%s : failed to eval\n", __func__);
return 1;
}
n_past += n_eval;
}
if (embd.size() > 0 && !path_session.empty()) {
session_tokens.insert(session_tokens.end(), embd.begin(), embd.end());
n_session_consumed = session_tokens.size();
}
}
embd.clear();
if ((int) embd_inp.size() <= n_consumed && !is_interacting) {
// out of user input, sample next token
const float temp = params.temp;
const int32_t top_k = params.top_k <= 0 ? llama_n_vocab(ctx) : params.top_k;
const float top_p = params.top_p;
const float tfs_z = params.tfs_z;
const float typical_p = params.typical_p;
const int32_t repeat_last_n = params.repeat_last_n < 0 ? n_ctx : params.repeat_last_n;
const float repeat_penalty = params.repeat_penalty;
const float alpha_presence = params.presence_penalty;
const float alpha_frequency = params.frequency_penalty;
const int mirostat = params.mirostat;
const float mirostat_tau = params.mirostat_tau;
const float mirostat_eta = params.mirostat_eta;
const bool penalize_nl = params.penalize_nl;
// optionally save the session on first sample (for faster prompt loading next time)
if (!path_session.empty() && need_to_save_session) {
need_to_save_session = false;
llama_save_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.size());
}
llama_token id = 0;
{
auto logits = llama_get_logits(ctx);
auto n_vocab = llama_n_vocab(ctx);
// Apply params.logit_bias map
for (auto it = params.logit_bias.begin(); it != params.logit_bias.end(); it++) {
logits[it->first] += it->second;
}
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
for (llama_token token_id = 0; token_id < n_vocab; token_id++) {
candidates.emplace_back(llama_token_data{token_id, logits[token_id], 0.0f});
}
llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
// Apply penalties
float nl_logit = logits[llama_token_nl()];
auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), n_ctx);
llama_sample_repetition_penalty(ctx, &candidates_p,
last_n_tokens.data() + last_n_tokens.size() - last_n_repeat,
last_n_repeat, repeat_penalty);
llama_sample_frequency_and_presence_penalties(ctx, &candidates_p,
last_n_tokens.data() + last_n_tokens.size() - last_n_repeat,
last_n_repeat, alpha_frequency, alpha_presence);
if (!penalize_nl) {
logits[llama_token_nl()] = nl_logit;
}
if (temp <= 0) {
// Greedy sampling
id = llama_sample_token_greedy(ctx, &candidates_p);
} else {
if (mirostat == 1) {
static float mirostat_mu = 2.0f * mirostat_tau;
const int mirostat_m = 100;
llama_sample_temperature(ctx, &candidates_p, temp);
id = llama_sample_token_mirostat(ctx, &candidates_p, mirostat_tau, mirostat_eta, mirostat_m, &mirostat_mu);
} else if (mirostat == 2) {
static float mirostat_mu = 2.0f * mirostat_tau;
llama_sample_temperature(ctx, &candidates_p, temp);
id = llama_sample_token_mirostat_v2(ctx, &candidates_p, mirostat_tau, mirostat_eta, &mirostat_mu);
} else {
// Temperature sampling
llama_sample_top_k(ctx, &candidates_p, top_k, 1);
llama_sample_tail_free(ctx, &candidates_p, tfs_z, 1);
llama_sample_typical(ctx, &candidates_p, typical_p, 1);
llama_sample_top_p(ctx, &candidates_p, top_p, 1);
llama_sample_temperature(ctx, &candidates_p, temp);
id = llama_sample_token(ctx, &candidates_p);
}
}
// printf("`%d`", candidates_p.size);
last_n_tokens.erase(last_n_tokens.begin());
last_n_tokens.push_back(id);
}
// replace end of text token with newline token when in interactive mode
if (id == llama_token_eos() && params.interactive && !params.instruct) {
id = llama_token_newline.front();
if (params.antiprompt.size() != 0) {
// tokenize and inject first reverse prompt
const auto first_antiprompt = ::llama_tokenize(ctx, params.antiprompt.front(), false);
embd_inp.insert(embd_inp.end(), first_antiprompt.begin(), first_antiprompt.end());
}
}
// add it to the context
embd.push_back(id);
// echo this to console
input_echo = true;
// decrement remaining sampling budget
--n_remain;
} else {
// some user input remains from prompt or interaction, forward it to processing
while ((int) embd_inp.size() > n_consumed) {
embd.push_back(embd_inp[n_consumed]);
last_n_tokens.erase(last_n_tokens.begin());
last_n_tokens.push_back(embd_inp[n_consumed]);
++n_consumed;
if ((int) embd.size() >= params.n_batch) {
break;
}
}
}
// display text
if (input_echo) {
for (auto id : embd) {
printf("%s", llama_token_to_str(ctx, id));
}
fflush(stdout);
}
// reset color to default if we there is no pending user input
if (input_echo && (int)embd_inp.size() == n_consumed) {
console_set_color(con_st, CONSOLE_COLOR_DEFAULT);
}
// if not currently processing queued inputs;
if ((int) embd_inp.size() <= n_consumed) {
// check for reverse prompt
if (params.antiprompt.size()) {
std::string last_output;
for (auto id : last_n_tokens) {
last_output += llama_token_to_str(ctx, id);
}
is_antiprompt = false;
// Check if each of the reverse prompts appears at the end of the output.
// If we're not running interactively, the reverse prompt might be tokenized with some following characters
// so we'll compensate for that by widening the search window a bit.
for (std::string & antiprompt : params.antiprompt) {
size_t extra_padding = params.interactive ? 0 : 2;
size_t search_start_pos = last_output.length() > static_cast<size_t>(antiprompt.length() + extra_padding)
? last_output.length() - static_cast<size_t>(antiprompt.length() + extra_padding)
: 0;
if (last_output.find(antiprompt.c_str(), search_start_pos) != std::string::npos) {
if (params.interactive) {
is_interacting = true;
console_set_color(con_st, CONSOLE_COLOR_USER_INPUT);
}
is_antiprompt = true;
fflush(stdout);
break;
}
}
}
if (n_past > 0 && is_interacting) {
if (params.instruct) {
printf("\n> ");
}
std::string buffer;
if (!params.input_prefix.empty()) {
buffer += params.input_prefix;
printf("%s", buffer.c_str());
}
std::string line;
bool another_line = true;
do {
another_line = console_readline(con_st, line);
buffer += line;
} while (another_line);
// done taking input, reset color
console_set_color(con_st, CONSOLE_COLOR_DEFAULT);
// Add tokens to embd only if the input buffer is non-empty
// Entering a empty line lets the user pass control back
if (buffer.length() > 1) {
// append input suffix if any
if (!params.input_suffix.empty()) {
buffer += params.input_suffix;
printf("%s", params.input_suffix.c_str());
}
// instruct mode: insert instruction prefix
if (params.instruct && !is_antiprompt) {
n_consumed = embd_inp.size();
embd_inp.insert(embd_inp.end(), inp_pfx.begin(), inp_pfx.end());
}
auto line_inp = ::llama_tokenize(ctx, buffer, false);
embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end());
// instruct mode: insert response suffix
if (params.instruct) {
embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
}
n_remain -= line_inp.size();
}
input_echo = false; // do not echo this again
}
if (n_past > 0) {
is_interacting = false;
}
}
// end of text token
if (!embd.empty() && embd.back() == llama_token_eos()) {
if (params.instruct) {
is_interacting = true;
} else {
fprintf(stderr, " [end of text]\n");
break;
}
}
// In interactive mode, respect the maximum number of tokens and drop back to user input when reached.
if (params.interactive && n_remain <= 0 && params.n_predict != -1) {
n_remain = params.n_predict;
is_interacting = true;
}
}
if (!path_session.empty() && params.prompt_cache_all) {
fprintf(stderr, "\n%s: saving final output to session file '%s'\n", __func__, path_session.c_str());
llama_save_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.size());
}
llama_print_timings(ctx);
llama_free(ctx);
return 0;
}