2023-04-11 15:19:54 +02:00
|
|
|
// Defines sigaction on msys:
|
|
|
|
#ifndef _GNU_SOURCE
|
|
|
|
#define _GNU_SOURCE
|
|
|
|
#endif
|
|
|
|
|
2023-03-25 19:26:40 +01:00
|
|
|
#include "common.h"
|
2023-08-04 17:20:12 +02:00
|
|
|
#include "console.h"
|
2023-03-22 06:32:36 +01:00
|
|
|
#include "llama.h"
|
2023-05-01 18:23:47 +02:00
|
|
|
#include "build-info.h"
|
2023-07-24 05:58:10 +02:00
|
|
|
#include "grammar-parser.h"
|
2023-03-10 19:40:58 +01:00
|
|
|
|
|
|
|
#include <cassert>
|
2023-03-20 11:17:23 +01:00
|
|
|
#include <cinttypes>
|
2023-03-10 19:40:58 +01:00
|
|
|
#include <cmath>
|
|
|
|
#include <cstdio>
|
|
|
|
#include <cstring>
|
2023-04-16 12:13:00 +02:00
|
|
|
#include <ctime>
|
2023-03-10 19:40:58 +01:00
|
|
|
#include <fstream>
|
2023-03-19 20:44:30 +01:00
|
|
|
#include <iostream>
|
2023-03-10 19:40:58 +01:00
|
|
|
#include <string>
|
|
|
|
#include <vector>
|
|
|
|
|
2023-03-13 04:08:01 +01:00
|
|
|
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
|
2023-03-12 22:13:28 +01:00
|
|
|
#include <signal.h>
|
|
|
|
#include <unistd.h>
|
2023-03-15 20:56:24 +01:00
|
|
|
#elif defined (_WIN32)
|
2023-05-03 03:01:57 +02:00
|
|
|
#define WIN32_LEAN_AND_MEAN
|
2023-06-16 20:25:01 +02:00
|
|
|
#ifndef NOMINMAX
|
2023-05-03 03:01:57 +02:00
|
|
|
#define NOMINMAX
|
2023-06-16 20:25:01 +02:00
|
|
|
#endif
|
2023-05-03 03:01:57 +02:00
|
|
|
#include <windows.h>
|
2023-03-15 20:56:24 +01:00
|
|
|
#include <signal.h>
|
2023-03-13 04:08:01 +01:00
|
|
|
#endif
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-06-16 20:23:53 +02:00
|
|
|
#if defined(_MSC_VER)
|
|
|
|
#pragma warning(disable: 4244 4267) // possible loss of data
|
|
|
|
#endif
|
|
|
|
|
2023-04-22 10:56:35 +02:00
|
|
|
static llama_context ** g_ctx;
|
2023-03-12 22:13:28 +01:00
|
|
|
static bool is_interacting = false;
|
|
|
|
|
2023-03-15 20:56:24 +01:00
|
|
|
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
|
2023-03-12 22:13:28 +01:00
|
|
|
void sigint_handler(int signo) {
|
|
|
|
if (signo == SIGINT) {
|
|
|
|
if (!is_interacting) {
|
2023-08-23 22:08:04 +02:00
|
|
|
is_interacting = true;
|
2023-03-12 22:13:28 +01:00
|
|
|
} else {
|
2023-08-04 17:20:12 +02:00
|
|
|
console::cleanup();
|
2023-05-09 04:45:48 +02:00
|
|
|
printf("\n");
|
2023-04-22 10:56:35 +02:00
|
|
|
llama_print_timings(*g_ctx);
|
2023-03-12 22:13:28 +01:00
|
|
|
_exit(130);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2023-03-13 04:08:01 +01:00
|
|
|
#endif
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-10 19:40:58 +01:00
|
|
|
int main(int argc, char ** argv) {
|
|
|
|
gpt_params params;
|
|
|
|
|
|
|
|
if (gpt_params_parse(argc, argv, params) == false) {
|
|
|
|
return 1;
|
|
|
|
}
|
2023-03-19 17:37:02 +01:00
|
|
|
|
2023-03-25 21:29:22 +01:00
|
|
|
// save choice to use color for later
|
|
|
|
// (note for later: this is a slightly awkward choice)
|
2023-08-04 17:20:12 +02:00
|
|
|
console::init(params.simple_io, params.use_color);
|
|
|
|
atexit([]() { console::cleanup(); });
|
2023-03-25 21:29:22 +01:00
|
|
|
|
2023-03-25 19:26:40 +01:00
|
|
|
if (params.perplexity) {
|
|
|
|
printf("\n************\n");
|
|
|
|
printf("%s: please use the 'perplexity' tool for perplexity calculations\n", __func__);
|
|
|
|
printf("************\n\n");
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
if (params.embedding) {
|
|
|
|
printf("\n************\n");
|
|
|
|
printf("%s: please use the 'embedding' tool for embedding calculations\n", __func__);
|
|
|
|
printf("************\n\n");
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
llama : add custom RoPE (#2054)
* Implement customizable RoPE
The original RoPE has pre-defined parameters
theta_i = 10000^(−2(i−1)/d), for i in [1, 2, ..., d/2]
Our customizable RoPE, ggml_rope_custom_inplace, uses
theta_i = scale * base^(−2(i−1)/d), for i in [1, 2, ..., d/2]
with the default matches the original
scale = 1.0
base = 10000
The new command line arguments
--rope-freq-base
--rope-freq-scale
set the two new RoPE parameter.
Recent researches show changing these two parameters extends the context limit with minimal loss.
1. Extending Context to 8K
kaiokendev
https://kaiokendev.github.io/til#extending-context-to-8k
2. Extending Context Window of Large Language Models via Positional Interpolation
Shouyuan Chen, Sherman Wong, Liangjian Chen, Yuandong Tian
https://arxiv.org/abs/2306.15595
3. NTK-Aware Scaled RoPE allows LLaMA models to have extended (8k+) context size without any fine-tuning and minimal perplexity degradation.
https://www.reddit.com/user/bloc97
https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/
For the bold, try adding the following command line parameters to your favorite model:
-c 16384 --rope-freq-base 80000 --rope-freq-scale 0.5
* ggml-metal: fix custom rope
* common: fix argument names in help
* llama: increase MEM_REQ_EVAL for MODEL_3B
It avoids crashing for quantized weights on CPU.
Better ways to calculate the required buffer size would be better.
* llama: make MEM_REQ_EVAL depend on n_ctx
* server: use proper Content-Type in curl examples
Without the header Content-Type: application/json, curl will POST with
Content-Type: application/x-www-form-urlencoded
Though our simple server doesn't care, the httplib.h used has a limit
with CPPHTTPLIB_FORM_URL_ENCODED_PAYLOAD_MAX_LENGTH 8192
With Content-Type: application/json, we can send large json data.
* style : minor fixes, mostly indentations
* ggml : fix asserts
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-07-15 12:34:16 +02:00
|
|
|
if (params.rope_freq_base != 10000.0) {
|
|
|
|
fprintf(stderr, "%s: warning: changing RoPE frequency base to %g (default 10000.0)\n", __func__, params.rope_freq_base);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (params.rope_freq_scale != 1.0) {
|
|
|
|
fprintf(stderr, "%s: warning: scaling RoPE frequency by %g (default 1.0)\n", __func__, params.rope_freq_scale);
|
|
|
|
}
|
|
|
|
|
2023-03-19 01:10:47 +01:00
|
|
|
if (params.n_ctx > 2048) {
|
2023-07-23 14:09:47 +02:00
|
|
|
// TODO: determine the actual max context of the model (e.g. 4096 for LLaMA v2) and use that instead of 2048
|
|
|
|
fprintf(stderr, "%s: warning: base model only supports context sizes no greater than 2048 tokens (%d specified)\n", __func__, params.n_ctx);
|
2023-06-11 16:19:17 +02:00
|
|
|
} else if (params.n_ctx < 8) {
|
|
|
|
fprintf(stderr, "%s: warning: minimum context size is 8, using minimum size.\n", __func__);
|
|
|
|
params.n_ctx = 8;
|
2023-03-19 01:10:47 +01:00
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-05-01 18:23:47 +02:00
|
|
|
fprintf(stderr, "%s: build = %d (%s)\n", __func__, BUILD_NUMBER, BUILD_COMMIT);
|
|
|
|
|
2023-06-29 15:15:15 +02:00
|
|
|
if (params.seed == LLAMA_DEFAULT_SEED) {
|
2023-03-10 19:40:58 +01:00
|
|
|
params.seed = time(NULL);
|
|
|
|
}
|
|
|
|
|
2023-06-29 15:15:15 +02:00
|
|
|
fprintf(stderr, "%s: seed = %u\n", __func__, params.seed);
|
2023-03-10 19:40:58 +01:00
|
|
|
|
|
|
|
std::mt19937 rng(params.seed);
|
2023-03-19 19:36:19 +01:00
|
|
|
if (params.random_prompt) {
|
2023-03-10 19:40:58 +01:00
|
|
|
params.prompt = gpt_random_prompt(rng);
|
|
|
|
}
|
|
|
|
|
2023-07-10 17:49:56 +02:00
|
|
|
llama_backend_init(params.numa);
|
2023-04-22 12:54:13 +02:00
|
|
|
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_model * model;
|
2023-03-22 06:32:36 +01:00
|
|
|
llama_context * ctx;
|
2023-07-11 18:18:43 +02:00
|
|
|
llama_context * ctx_guidance = NULL;
|
2023-04-22 10:56:35 +02:00
|
|
|
g_ctx = &ctx;
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-05-02 22:39:51 +02:00
|
|
|
// load the model and apply lora adapter, if any
|
2023-06-24 10:47:58 +02:00
|
|
|
std::tie(model, ctx) = llama_init_from_gpt_params(params);
|
2023-07-11 18:18:43 +02:00
|
|
|
if (params.cfg_scale > 1.f) {
|
|
|
|
struct llama_context_params lparams = llama_context_params_from_gpt_params(params);
|
|
|
|
ctx_guidance = llama_new_context_with_model(model, lparams);
|
|
|
|
}
|
|
|
|
|
2023-06-24 10:47:58 +02:00
|
|
|
if (model == NULL) {
|
2023-05-02 22:39:51 +02:00
|
|
|
fprintf(stderr, "%s: error: unable to load model\n", __func__);
|
|
|
|
return 1;
|
2023-04-17 17:28:55 +02:00
|
|
|
}
|
|
|
|
|
2023-03-13 18:15:08 +01:00
|
|
|
// 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());
|
|
|
|
}
|
|
|
|
|
2023-07-22 20:17:57 +02:00
|
|
|
// determine the maximum memory usage needed to do inference for the given n_batch and n_ctx parameters
|
2023-03-24 22:17:37 +01:00
|
|
|
// uncomment the "used_mem" line in llama.cpp to see the results
|
|
|
|
if (params.mem_test) {
|
|
|
|
{
|
2023-07-22 20:17:57 +02:00
|
|
|
fprintf(stderr, "%s: testing memory usage for n_batch = %d, n_ctx = %d\n", __func__, params.n_batch, params.n_ctx);
|
2023-03-24 22:17:37 +01:00
|
|
|
|
2023-08-21 22:07:43 +02:00
|
|
|
const std::vector<llama_token> tmp(params.n_batch, llama_token_bos(ctx));
|
2023-07-22 20:17:57 +02:00
|
|
|
llama_eval(ctx, tmp.data(), tmp.size(), params.n_ctx, params.n_threads);
|
2023-03-24 22:17:37 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
llama_print_timings(ctx);
|
|
|
|
llama_free(ctx);
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_free_model(model);
|
2023-03-24 22:17:37 +01:00
|
|
|
|
|
|
|
return 0;
|
2023-03-22 06:32:36 +01:00
|
|
|
}
|
2023-03-21 17:27:42 +01:00
|
|
|
|
2023-06-04 22:34:30 +02:00
|
|
|
// export the cgraph and exit
|
|
|
|
if (params.export_cgraph) {
|
|
|
|
llama_eval_export(ctx, "llama.ggml");
|
|
|
|
llama_free(ctx);
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_free_model(model);
|
2023-06-04 22:34:30 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2023-03-22 06:32:36 +01:00
|
|
|
|
2023-05-10 17:37:14 +02:00
|
|
|
std::string path_session = params.path_prompt_cache;
|
2023-04-28 17:59:37 +02:00
|
|
|
std::vector<llama_token> session_tokens;
|
|
|
|
|
|
|
|
if (!path_session.empty()) {
|
2023-05-01 13:54:59 +02:00
|
|
|
fprintf(stderr, "%s: attempting to load saved session from '%s'\n", __func__, path_session.c_str());
|
2023-04-28 17:59:37 +02:00
|
|
|
|
2023-05-01 13:54:59 +02:00
|
|
|
// fopen to check for existing session
|
2023-04-28 17:59:37 +02:00
|
|
|
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;
|
2023-05-01 13:54:59 +02:00
|
|
|
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;
|
|
|
|
}
|
2023-04-28 17:59:37 +02:00
|
|
|
session_tokens.resize(n_token_count_out);
|
2023-05-26 04:18:01 +02:00
|
|
|
llama_set_rng_seed(ctx, params.seed);
|
2023-04-28 17:59:37 +02:00
|
|
|
|
2023-05-01 13:54:59 +02:00
|
|
|
fprintf(stderr, "%s: loaded a session with prompt size of %d tokens\n", __func__, (int) session_tokens.size());
|
2023-04-28 17:59:37 +02:00
|
|
|
} else {
|
|
|
|
fprintf(stderr, "%s: session file does not exist, will create\n", __func__);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-08-26 13:45:53 +02:00
|
|
|
// Add BOS if SPM tokenizer
|
|
|
|
const bool add_bos = llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM;
|
2023-08-23 22:08:04 +02:00
|
|
|
|
2023-03-10 19:40:58 +01:00
|
|
|
// tokenize the prompt
|
2023-05-26 04:18:01 +02:00
|
|
|
std::vector<llama_token> embd_inp;
|
2023-08-26 13:45:53 +02:00
|
|
|
|
|
|
|
if (llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM) {
|
|
|
|
// Add a space in front of the first character to match OG llama tokenizer behavior
|
|
|
|
params.prompt.insert(0, 1, ' ');
|
|
|
|
}
|
|
|
|
|
2023-07-11 18:18:43 +02:00
|
|
|
if (params.interactive_first || params.instruct || !params.prompt.empty() || session_tokens.empty()) {
|
2023-08-26 13:45:53 +02:00
|
|
|
embd_inp = ::llama_tokenize(ctx, params.prompt, add_bos);
|
2023-05-26 04:18:01 +02:00
|
|
|
} else {
|
|
|
|
embd_inp = session_tokens;
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-08-23 16:46:03 +02:00
|
|
|
// Should not run without any tokens
|
|
|
|
if (embd_inp.empty()) {
|
|
|
|
embd_inp.push_back(llama_token_bos(ctx));
|
|
|
|
}
|
|
|
|
|
2023-07-11 18:18:43 +02:00
|
|
|
// Tokenize negative prompt
|
|
|
|
std::vector<llama_token> guidance_inp;
|
|
|
|
int guidance_offset = 0;
|
|
|
|
int original_prompt_len = 0;
|
|
|
|
if (ctx_guidance) {
|
|
|
|
params.cfg_negative_prompt.insert(0, 1, ' ');
|
2023-08-26 13:45:53 +02:00
|
|
|
guidance_inp = ::llama_tokenize(ctx_guidance, params.cfg_negative_prompt, add_bos);
|
2023-07-11 18:18:43 +02:00
|
|
|
|
2023-08-26 13:45:53 +02:00
|
|
|
std::vector<llama_token> original_inp = ::llama_tokenize(ctx, params.prompt, add_bos);
|
2023-07-11 18:18:43 +02:00
|
|
|
original_prompt_len = original_inp.size();
|
|
|
|
guidance_offset = (int)guidance_inp.size() - original_prompt_len;
|
|
|
|
}
|
|
|
|
|
2023-03-22 06:32:36 +01:00
|
|
|
const int n_ctx = llama_n_ctx(ctx);
|
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
|
2023-04-28 17:59:37 +02:00
|
|
|
// 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++;
|
|
|
|
}
|
2023-05-26 04:18:01 +02:00
|
|
|
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()) {
|
2023-04-28 17:59:37 +02:00
|
|
|
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());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-06-03 13:28:45 +02:00
|
|
|
// 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);
|
|
|
|
}
|
|
|
|
|
2023-03-28 16:09:55 +02:00
|
|
|
// number of tokens to keep when resetting context
|
2023-05-01 13:54:59 +02:00
|
|
|
if (params.n_keep < 0 || params.n_keep > (int) embd_inp.size() || params.instruct) {
|
2023-03-28 16:09:55 +02:00
|
|
|
params.n_keep = (int)embd_inp.size();
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-03-19 17:37:02 +01:00
|
|
|
// prefix & suffix for instruct mode
|
2023-08-26 13:45:53 +02:00
|
|
|
const auto inp_pfx = ::llama_tokenize(ctx, "\n\n### Instruction:\n\n", add_bos);
|
2023-08-23 22:08:04 +02:00
|
|
|
const auto inp_sfx = ::llama_tokenize(ctx, "\n\n### Response:\n\n", false);
|
2023-04-14 21:58:43 +02:00
|
|
|
|
|
|
|
// in instruct mode, we inject a prefix and a suffix to each input by the user
|
|
|
|
if (params.instruct) {
|
2023-04-24 17:45:32 +02:00
|
|
|
params.interactive_first = true;
|
2023-04-14 21:58:43 +02:00
|
|
|
params.antiprompt.push_back("### Instruction:\n\n");
|
2023-03-19 17:37:02 +01:00
|
|
|
}
|
|
|
|
|
2023-05-19 19:24:59 +02:00
|
|
|
// enable interactive mode if interactive start is specified
|
|
|
|
if (params.interactive_first) {
|
2023-03-22 18:16:35 +01:00
|
|
|
params.interactive = true;
|
|
|
|
}
|
|
|
|
|
2023-03-25 16:16:50 +01:00
|
|
|
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++) {
|
2023-08-21 22:07:43 +02:00
|
|
|
fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_str(ctx, embd_inp[i]).c_str());
|
2023-03-25 16:16:50 +01:00
|
|
|
}
|
2023-07-11 18:18:43 +02:00
|
|
|
|
|
|
|
if (ctx_guidance) {
|
|
|
|
fprintf(stderr, "\n");
|
|
|
|
fprintf(stderr, "%s: negative prompt: '%s'\n", __func__, params.cfg_negative_prompt.c_str());
|
|
|
|
fprintf(stderr, "%s: number of tokens in negative prompt = %zu\n", __func__, guidance_inp.size());
|
|
|
|
for (int i = 0; i < (int) guidance_inp.size(); i++) {
|
2023-08-21 22:07:43 +02:00
|
|
|
fprintf(stderr, "%6d -> '%s'\n", guidance_inp[i], llama_token_to_str(ctx, guidance_inp[i]).c_str());
|
2023-07-11 18:18:43 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
if (params.n_keep > 0) {
|
|
|
|
fprintf(stderr, "%s: static prompt based on n_keep: '", __func__);
|
|
|
|
for (int i = 0; i < params.n_keep; i++) {
|
2023-08-21 22:07:43 +02:00
|
|
|
fprintf(stderr, "%s", llama_token_to_str(ctx, embd_inp[i]).c_str());
|
2023-03-25 20:36:22 +01:00
|
|
|
}
|
|
|
|
fprintf(stderr, "'\n");
|
|
|
|
}
|
2023-03-25 16:16:50 +01:00
|
|
|
fprintf(stderr, "\n");
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
2023-03-25 16:16:50 +01:00
|
|
|
|
2023-03-12 22:13:28 +01:00
|
|
|
if (params.interactive) {
|
2023-03-13 04:08:01 +01:00
|
|
|
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
|
2023-03-12 22:13:28 +01:00
|
|
|
struct sigaction sigint_action;
|
|
|
|
sigint_action.sa_handler = sigint_handler;
|
|
|
|
sigemptyset (&sigint_action.sa_mask);
|
2023-03-13 18:15:08 +01:00
|
|
|
sigint_action.sa_flags = 0;
|
2023-03-12 22:13:28 +01:00
|
|
|
sigaction(SIGINT, &sigint_action, NULL);
|
2023-03-15 20:56:24 +01:00
|
|
|
#elif defined (_WIN32)
|
2023-05-18 19:30:40 +02:00
|
|
|
auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
|
2023-05-03 03:01:57 +02:00
|
|
|
return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false;
|
|
|
|
};
|
2023-08-21 22:07:43 +02:00
|
|
|
SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
|
2023-03-13 04:08:01 +01:00
|
|
|
#endif
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-13 17:39:56 +01:00
|
|
|
fprintf(stderr, "%s: interactive mode on.\n", __func__);
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
if (params.antiprompt.size()) {
|
2023-03-21 17:04:43 +01:00
|
|
|
for (auto antiprompt : params.antiprompt) {
|
|
|
|
fprintf(stderr, "Reverse prompt: '%s'\n", antiprompt.c_str());
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
|
|
|
}
|
2023-03-25 13:03:19 +01:00
|
|
|
|
2023-07-25 14:19:11 +02:00
|
|
|
if (params.input_prefix_bos) {
|
|
|
|
fprintf(stderr, "Input prefix with BOS\n");
|
|
|
|
}
|
|
|
|
|
2023-03-25 13:03:19 +01:00
|
|
|
if (!params.input_prefix.empty()) {
|
|
|
|
fprintf(stderr, "Input prefix: '%s'\n", params.input_prefix.c_str());
|
|
|
|
}
|
2023-05-04 17:41:12 +02:00
|
|
|
|
|
|
|
if (!params.input_suffix.empty()) {
|
|
|
|
fprintf(stderr, "Input suffix: '%s'\n", params.input_suffix.c_str());
|
|
|
|
}
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
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);
|
2023-03-25 20:36:22 +01:00
|
|
|
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);
|
2023-03-13 17:39:56 +01:00
|
|
|
fprintf(stderr, "\n\n");
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-07-24 05:58:10 +02:00
|
|
|
grammar_parser::parse_state parsed_grammar;
|
|
|
|
llama_grammar * grammar = NULL;
|
|
|
|
if (!params.grammar.empty()) {
|
|
|
|
parsed_grammar = grammar_parser::parse(params.grammar.c_str());
|
|
|
|
// will be empty (default) if there are parse errors
|
|
|
|
if (parsed_grammar.rules.empty()) {
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
fprintf(stderr, "%s: grammar:\n", __func__);
|
|
|
|
grammar_parser::print_grammar(stderr, parsed_grammar);
|
|
|
|
fprintf(stderr, "\n");
|
|
|
|
|
|
|
|
{
|
2023-08-21 22:07:43 +02:00
|
|
|
auto it = params.logit_bias.find(llama_token_eos(ctx));
|
2023-07-24 05:58:10 +02:00
|
|
|
if (it != params.logit_bias.end() && it->second == -INFINITY) {
|
2023-08-21 22:07:43 +02:00
|
|
|
fprintf(stderr, "%s: warning: EOS token is disabled, which will cause most grammars to fail\n", __func__);
|
2023-07-24 05:58:10 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<const llama_grammar_element *> grammar_rules(parsed_grammar.c_rules());
|
|
|
|
grammar = llama_grammar_init(
|
|
|
|
grammar_rules.data(), grammar_rules.size(), parsed_grammar.symbol_ids.at("root"));
|
|
|
|
}
|
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
// TODO: replace with ring-buffer
|
|
|
|
std::vector<llama_token> last_n_tokens(n_ctx);
|
2023-03-12 10:27:42 +01:00
|
|
|
std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0);
|
|
|
|
|
2023-03-12 22:13:28 +01:00
|
|
|
if (params.interactive) {
|
2023-05-09 04:45:48 +02:00
|
|
|
const char *control_message;
|
2023-08-04 17:20:12 +02:00
|
|
|
if (params.multiline_input) {
|
2023-05-09 04:45:48 +02:00
|
|
|
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";
|
|
|
|
}
|
2023-03-13 17:39:56 +01:00
|
|
|
fprintf(stderr, "== Running in interactive mode. ==\n"
|
2023-03-15 20:56:24 +01:00
|
|
|
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
|
2023-03-12 22:13:28 +01:00
|
|
|
" - Press Ctrl+C to interject at any time.\n"
|
2023-03-13 04:08:01 +01:00
|
|
|
#endif
|
2023-05-09 04:45:48 +02:00
|
|
|
"%s\n", control_message);
|
|
|
|
|
2023-04-24 17:45:32 +02:00
|
|
|
is_interacting = params.interactive_first;
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
|
|
|
|
2023-05-10 17:37:14 +02:00
|
|
|
bool is_antiprompt = false;
|
|
|
|
bool input_echo = true;
|
|
|
|
bool need_to_save_session = !path_session.empty() && n_matching_session_tokens < embd_inp.size();
|
2023-04-28 17:59:37 +02:00
|
|
|
|
2023-05-02 19:53:52 +02:00
|
|
|
int n_past = 0;
|
|
|
|
int n_remain = params.n_predict;
|
|
|
|
int n_consumed = 0;
|
2023-04-28 17:59:37 +02:00
|
|
|
int n_session_consumed = 0;
|
2023-07-11 18:18:43 +02:00
|
|
|
int n_past_guidance = 0;
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-21 18:11:01 +01:00
|
|
|
// the first thing we will do is to output the prompt, so set color accordingly
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::prompt);
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
std::vector<llama_token> embd;
|
2023-07-11 18:18:43 +02:00
|
|
|
std::vector<llama_token> embd_guidance;
|
2023-03-25 20:36:22 +01:00
|
|
|
|
2023-06-13 19:20:07 +02:00
|
|
|
// do one empty run to warm up the model
|
|
|
|
{
|
2023-08-21 22:07:43 +02:00
|
|
|
const std::vector<llama_token> tmp = { llama_token_bos(ctx), };
|
2023-06-13 19:20:07 +02:00
|
|
|
llama_eval(ctx, tmp.data(), tmp.size(), 0, params.n_threads);
|
|
|
|
llama_reset_timings(ctx);
|
|
|
|
}
|
|
|
|
|
2023-05-19 19:24:59 +02:00
|
|
|
while ((n_remain != 0 && !is_antiprompt) || params.interactive) {
|
2023-03-10 19:40:58 +01:00
|
|
|
// predict
|
|
|
|
if (embd.size() > 0) {
|
2023-06-11 16:19:17 +02:00
|
|
|
// Note: n_ctx - 4 here is to match the logic for commandline prompt handling via
|
|
|
|
// --prompt or --file which uses the same value.
|
|
|
|
auto max_embd_size = n_ctx - 4;
|
|
|
|
// Ensure the input doesn't exceed the context size by truncating embd if necessary.
|
|
|
|
if ((int)embd.size() > max_embd_size) {
|
|
|
|
auto skipped_tokens = embd.size() - max_embd_size;
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::error);
|
2023-06-17 19:24:11 +02:00
|
|
|
printf("<<input too long: skipped %zu token%s>>", skipped_tokens, skipped_tokens != 1 ? "s" : "");
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::reset);
|
2023-06-11 16:19:17 +02:00
|
|
|
fflush(stdout);
|
|
|
|
embd.resize(max_embd_size);
|
|
|
|
}
|
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
// 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)
|
2023-04-21 20:18:09 +02:00
|
|
|
// - take half of the last (n_ctx - n_keep) tokens and recompute the logits in batches
|
2023-07-11 18:18:43 +02:00
|
|
|
if (n_past + (int) embd.size() + std::max<int>(0, guidance_offset) > n_ctx) {
|
2023-08-10 16:28:27 +02:00
|
|
|
if (params.n_predict == -2) {
|
|
|
|
fprintf(stderr, "\n\n%s: context full, stopping generation\n", __func__);
|
|
|
|
break;
|
|
|
|
}
|
2023-03-25 20:36:22 +01:00
|
|
|
|
2023-08-10 16:28:27 +02:00
|
|
|
const int n_left = n_past - params.n_keep;
|
2023-05-08 16:41:54 +02:00
|
|
|
// always keep the first token - BOS
|
|
|
|
n_past = std::max(1, params.n_keep);
|
2023-07-11 18:18:43 +02:00
|
|
|
n_past_guidance = std::max(1, params.n_keep + guidance_offset);
|
2023-03-25 20:36:22 +01:00
|
|
|
|
|
|
|
// 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());
|
|
|
|
|
2023-05-01 13:54:59 +02:00
|
|
|
// stop saving session if we run out of context
|
2023-05-10 17:37:14 +02:00
|
|
|
path_session.clear();
|
2023-04-28 17:59:37 +02:00
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
//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");
|
|
|
|
}
|
|
|
|
|
2023-04-28 17:59:37 +02:00
|
|
|
// 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()) {
|
2023-05-01 13:54:59 +02:00
|
|
|
++i;
|
2023-04-28 17:59:37 +02:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (i > 0) {
|
|
|
|
embd.erase(embd.begin(), embd.begin() + i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-04-21 20:18:09 +02:00
|
|
|
// evaluate tokens in batches
|
|
|
|
// embd is typically prepared beforehand to fit within a batch, but not always
|
2023-07-11 18:18:43 +02:00
|
|
|
|
|
|
|
if (ctx_guidance) {
|
|
|
|
int input_size = 0;
|
|
|
|
llama_token* input_buf = NULL;
|
|
|
|
|
|
|
|
if (n_past_guidance < (int) guidance_inp.size()) {
|
|
|
|
// Guidance context should have the same data with these modifications:
|
|
|
|
//
|
|
|
|
// * Replace the initial prompt
|
|
|
|
// * Shift everything by guidance_offset
|
|
|
|
embd_guidance = guidance_inp;
|
|
|
|
if (embd.begin() + original_prompt_len < embd.end()) {
|
|
|
|
embd_guidance.insert(
|
|
|
|
embd_guidance.end(),
|
|
|
|
embd.begin() + original_prompt_len,
|
|
|
|
embd.end()
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
input_buf = embd_guidance.data();
|
|
|
|
input_size = embd_guidance.size();
|
|
|
|
//fprintf(stderr, "\n---------------------\n");
|
|
|
|
//for (int i = 0; i < (int) embd_guidance.size(); i++) {
|
|
|
|
//fprintf(stderr, "%s", llama_token_to_str(ctx, embd_guidance[i]));
|
|
|
|
//}
|
|
|
|
//fprintf(stderr, "\n---------------------\n");
|
|
|
|
} else {
|
|
|
|
input_buf = embd.data();
|
|
|
|
input_size = embd.size();
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int i = 0; i < input_size; i += params.n_batch) {
|
|
|
|
int n_eval = std::min(input_size - i, params.n_batch);
|
|
|
|
if (llama_eval(ctx_guidance, input_buf + i, n_eval, n_past_guidance, params.n_threads)) {
|
|
|
|
fprintf(stderr, "%s : failed to eval\n", __func__);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
n_past_guidance += n_eval;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-04-21 20:18:09 +02:00
|
|
|
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;
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
2023-04-28 17:59:37 +02:00
|
|
|
|
|
|
|
if (embd.size() > 0 && !path_session.empty()) {
|
|
|
|
session_tokens.insert(session_tokens.end(), embd.begin(), embd.end());
|
|
|
|
n_session_consumed = session_tokens.size();
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
embd.clear();
|
2023-07-11 18:18:43 +02:00
|
|
|
embd_guidance.clear();
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-03-25 20:36:22 +01:00
|
|
|
if ((int) embd_inp.size() <= n_consumed && !is_interacting) {
|
2023-03-12 23:07:34 +01:00
|
|
|
// out of user input, sample next token
|
2023-04-29 08:51:06 +02:00
|
|
|
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;
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
const float alpha_frequency = params.frequency_penalty;
|
2023-04-29 08:51:06 +02:00
|
|
|
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;
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-04-28 17:59:37 +02:00
|
|
|
// optionally save the session on first sample (for faster prompt loading next time)
|
2023-06-07 04:10:17 +02:00
|
|
|
if (!path_session.empty() && need_to_save_session && !params.prompt_cache_ro) {
|
2023-04-28 17:59:37 +02:00
|
|
|
need_to_save_session = false;
|
|
|
|
llama_save_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.size());
|
|
|
|
}
|
|
|
|
|
2023-03-22 06:32:36 +01:00
|
|
|
llama_token id = 0;
|
2023-03-10 19:40:58 +01:00
|
|
|
|
|
|
|
{
|
2023-05-02 19:53:52 +02:00
|
|
|
auto logits = llama_get_logits(ctx);
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
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;
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
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});
|
2023-03-19 19:22:48 +01:00
|
|
|
}
|
|
|
|
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
|
|
|
|
|
2023-07-11 18:18:43 +02:00
|
|
|
if (ctx_guidance) {
|
2023-07-21 12:58:36 +02:00
|
|
|
llama_sample_classifier_free_guidance(ctx, &candidates_p, ctx_guidance, params.cfg_scale);
|
2023-07-11 18:18:43 +02:00
|
|
|
}
|
|
|
|
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
// Apply penalties
|
2023-08-21 22:07:43 +02:00
|
|
|
float nl_logit = logits[llama_token_nl(ctx)];
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
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) {
|
2023-08-21 22:07:43 +02:00
|
|
|
logits[llama_token_nl(ctx)] = nl_logit;
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
}
|
|
|
|
|
2023-07-24 05:58:10 +02:00
|
|
|
if (grammar != NULL) {
|
|
|
|
llama_sample_grammar(ctx, &candidates_p, grammar);
|
|
|
|
}
|
|
|
|
|
llama : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
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
|
2023-05-06 23:01:47 +02:00
|
|
|
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 : new sampling algorithms (#1126)
* Sample interface, new samplers.
New samplers:
- locally typical sampling
- tail free sampling
- frequency and presence penalty
- mirostat
Ignore EOS fix: -inf should be used.
* mirostat
* Added --logit-bias and --no-penalize-nl, removed std::span
* Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
Use C++11, clarify llama API documentation, rename Mirostat parameters to --mirostat_lr and --mirostat_ent, add temperature sampling for Mirostat, simplify Mirostat sampling API parameters (removed N and *k)
* Save and load example adjust
* Tests
* Windows build fix
* Windows test fix
2023-04-29 07:34:41 +02:00
|
|
|
llama_sample_temperature(ctx, &candidates_p, temp);
|
|
|
|
id = llama_sample_token(ctx, &candidates_p);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// printf("`%d`", candidates_p.size);
|
2023-03-12 10:27:42 +01:00
|
|
|
|
2023-07-24 05:58:10 +02:00
|
|
|
if (grammar != NULL) {
|
|
|
|
llama_grammar_accept_token(ctx, grammar, id);
|
|
|
|
}
|
|
|
|
|
2023-03-12 10:27:42 +01:00
|
|
|
last_n_tokens.erase(last_n_tokens.begin());
|
|
|
|
last_n_tokens.push_back(id);
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// add it to the context
|
|
|
|
embd.push_back(id);
|
2023-03-12 22:13:28 +01:00
|
|
|
|
|
|
|
// echo this to console
|
2023-05-02 18:13:26 +02:00
|
|
|
input_echo = true;
|
2023-03-12 22:13:28 +01:00
|
|
|
|
|
|
|
// decrement remaining sampling budget
|
2023-03-25 20:36:22 +01:00
|
|
|
--n_remain;
|
2023-03-10 19:40:58 +01:00
|
|
|
} else {
|
2023-03-12 23:07:34 +01:00
|
|
|
// some user input remains from prompt or interaction, forward it to processing
|
2023-03-25 20:36:22 +01:00
|
|
|
while ((int) embd_inp.size() > n_consumed) {
|
|
|
|
embd.push_back(embd_inp[n_consumed]);
|
2023-03-12 10:27:42 +01:00
|
|
|
last_n_tokens.erase(last_n_tokens.begin());
|
2023-03-25 20:36:22 +01:00
|
|
|
last_n_tokens.push_back(embd_inp[n_consumed]);
|
|
|
|
++n_consumed;
|
2023-03-19 18:46:32 +01:00
|
|
|
if ((int) embd.size() >= params.n_batch) {
|
2023-03-10 19:40:58 +01:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// display text
|
2023-05-02 18:13:26 +02:00
|
|
|
if (input_echo) {
|
2023-03-12 22:13:28 +01:00
|
|
|
for (auto id : embd) {
|
2023-08-21 22:07:43 +02:00
|
|
|
printf("%s", llama_token_to_str(ctx, id).c_str());
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
|
|
|
fflush(stdout);
|
|
|
|
}
|
2023-03-19 20:44:30 +01:00
|
|
|
// reset color to default if we there is no pending user input
|
2023-05-02 18:13:26 +02:00
|
|
|
if (input_echo && (int)embd_inp.size() == n_consumed) {
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::reset);
|
2023-03-19 20:44:30 +01:00
|
|
|
}
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-05-19 19:24:59 +02:00
|
|
|
// if not currently processing queued inputs;
|
|
|
|
if ((int) embd_inp.size() <= n_consumed) {
|
2023-03-28 16:09:55 +02:00
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
// check for reverse prompt
|
|
|
|
if (params.antiprompt.size()) {
|
2023-03-28 16:09:55 +02:00
|
|
|
std::string last_output;
|
|
|
|
for (auto id : last_n_tokens) {
|
|
|
|
last_output += llama_token_to_str(ctx, id);
|
|
|
|
}
|
2023-03-21 17:04:43 +01:00
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
is_antiprompt = false;
|
2023-03-28 16:09:55 +02:00
|
|
|
// Check if each of the reverse prompts appears at the end of the output.
|
2023-05-19 19:24:59 +02:00
|
|
|
// 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.
|
2023-04-14 21:58:43 +02:00
|
|
|
for (std::string & antiprompt : params.antiprompt) {
|
2023-05-19 19:24:59 +02:00
|
|
|
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;
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::user_input);
|
2023-05-19 19:24:59 +02:00
|
|
|
}
|
2023-04-14 21:58:43 +02:00
|
|
|
is_antiprompt = true;
|
2023-05-19 19:24:59 +02:00
|
|
|
fflush(stdout);
|
2023-03-28 16:09:55 +02:00
|
|
|
break;
|
|
|
|
}
|
2023-03-19 20:33:06 +01:00
|
|
|
}
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
2023-03-24 22:17:58 +01:00
|
|
|
|
2023-07-25 14:19:11 +02:00
|
|
|
// deal with end of text token in interactive mode
|
2023-08-21 22:07:43 +02:00
|
|
|
if (last_n_tokens.back() == llama_token_eos(ctx)) {
|
2023-07-25 14:19:11 +02:00
|
|
|
if (params.interactive) {
|
|
|
|
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());
|
|
|
|
is_antiprompt = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
is_interacting = true;
|
|
|
|
printf("\n");
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::user_input);
|
2023-07-25 14:19:11 +02:00
|
|
|
fflush(stdout);
|
|
|
|
} else if (params.instruct) {
|
|
|
|
is_interacting = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
if (n_past > 0 && is_interacting) {
|
|
|
|
if (params.instruct) {
|
2023-03-19 17:37:02 +01:00
|
|
|
printf("\n> ");
|
|
|
|
}
|
|
|
|
|
2023-07-25 14:19:11 +02:00
|
|
|
if (params.input_prefix_bos) {
|
2023-08-21 22:07:43 +02:00
|
|
|
embd_inp.push_back(llama_token_bos(ctx));
|
2023-07-25 14:19:11 +02:00
|
|
|
}
|
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
std::string buffer;
|
2023-03-25 13:03:19 +01:00
|
|
|
if (!params.input_prefix.empty()) {
|
|
|
|
buffer += params.input_prefix;
|
2023-03-25 15:22:05 +01:00
|
|
|
printf("%s", buffer.c_str());
|
2023-03-25 13:03:19 +01:00
|
|
|
}
|
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
std::string line;
|
|
|
|
bool another_line = true;
|
|
|
|
do {
|
2023-08-04 17:20:12 +02:00
|
|
|
another_line = console::readline(line, params.multiline_input);
|
2023-05-09 04:45:48 +02:00
|
|
|
buffer += line;
|
2023-04-14 21:58:43 +02:00
|
|
|
} while (another_line);
|
2023-03-21 18:11:01 +01:00
|
|
|
|
|
|
|
// done taking input, reset color
|
2023-08-04 17:20:12 +02:00
|
|
|
console::set_display(console::reset);
|
2023-03-19 17:37:02 +01:00
|
|
|
|
2023-03-28 16:09:55 +02:00
|
|
|
// 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) {
|
2023-05-04 17:41:12 +02:00
|
|
|
// append input suffix if any
|
|
|
|
if (!params.input_suffix.empty()) {
|
|
|
|
buffer += params.input_suffix;
|
|
|
|
printf("%s", params.input_suffix.c_str());
|
|
|
|
}
|
2023-03-13 00:35:51 +01:00
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
// instruct mode: insert instruction prefix
|
|
|
|
if (params.instruct && !is_antiprompt) {
|
2023-03-28 16:09:55 +02:00
|
|
|
n_consumed = embd_inp.size();
|
|
|
|
embd_inp.insert(embd_inp.end(), inp_pfx.begin(), inp_pfx.end());
|
|
|
|
}
|
2023-03-12 22:13:28 +01:00
|
|
|
|
2023-03-28 16:09:55 +02:00
|
|
|
auto line_inp = ::llama_tokenize(ctx, buffer, false);
|
|
|
|
embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end());
|
|
|
|
|
2023-04-14 21:58:43 +02:00
|
|
|
// instruct mode: insert response suffix
|
|
|
|
if (params.instruct) {
|
2023-03-28 16:09:55 +02:00
|
|
|
embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
|
|
|
|
}
|
|
|
|
|
|
|
|
n_remain -= line_inp.size();
|
|
|
|
}
|
2023-03-19 20:44:30 +01:00
|
|
|
|
2023-05-02 18:13:26 +02:00
|
|
|
input_echo = false; // do not echo this again
|
2023-03-12 22:13:28 +01:00
|
|
|
}
|
2023-03-24 22:17:58 +01:00
|
|
|
|
|
|
|
if (n_past > 0) {
|
2023-07-24 05:58:10 +02:00
|
|
|
if (is_interacting) {
|
|
|
|
// reset grammar state if we're restarting generation
|
|
|
|
if (grammar != NULL) {
|
|
|
|
llama_grammar_free(grammar);
|
|
|
|
|
2023-08-21 22:07:43 +02:00
|
|
|
std::vector<const llama_grammar_element *> grammar_rules( parsed_grammar.c_rules());
|
2023-07-24 05:58:10 +02:00
|
|
|
grammar = llama_grammar_init(
|
|
|
|
grammar_rules.data(), grammar_rules.size(),
|
|
|
|
parsed_grammar.symbol_ids.at("root"));
|
|
|
|
}
|
|
|
|
}
|
2023-03-24 22:17:58 +01:00
|
|
|
is_interacting = false;
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// end of text token
|
2023-08-21 22:07:43 +02:00
|
|
|
if (!embd.empty() && embd.back() == llama_token_eos(ctx) && !(params.instruct || params.interactive)) {
|
2023-07-25 14:19:11 +02:00
|
|
|
fprintf(stderr, " [end of text]\n");
|
|
|
|
break;
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
2023-03-19 19:31:17 +01:00
|
|
|
|
|
|
|
// In interactive mode, respect the maximum number of tokens and drop back to user input when reached.
|
2023-08-24 18:11:13 +02:00
|
|
|
// We skip this logic when n_predict == -1 (infinite) or -2 (stop at context size).
|
|
|
|
if (params.interactive && n_remain <= 0 && params.n_predict >= 0) {
|
2023-03-25 20:36:22 +01:00
|
|
|
n_remain = params.n_predict;
|
2023-03-19 19:31:17 +01:00
|
|
|
is_interacting = true;
|
|
|
|
}
|
2023-03-10 19:40:58 +01:00
|
|
|
}
|
|
|
|
|
2023-06-07 04:10:17 +02:00
|
|
|
if (!path_session.empty() && params.prompt_cache_all && !params.prompt_cache_ro) {
|
2023-05-10 17:37:14 +02:00
|
|
|
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());
|
|
|
|
}
|
|
|
|
|
2023-03-22 06:32:36 +01:00
|
|
|
llama_print_timings(ctx);
|
2023-07-11 18:18:43 +02:00
|
|
|
if (ctx_guidance) { llama_free(ctx_guidance); }
|
2023-03-22 06:32:36 +01:00
|
|
|
llama_free(ctx);
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_free_model(model);
|
2023-03-10 19:40:58 +01:00
|
|
|
|
2023-07-24 05:58:10 +02:00
|
|
|
if (grammar != NULL) {
|
|
|
|
llama_grammar_free(grammar);
|
|
|
|
}
|
2023-07-10 17:49:56 +02:00
|
|
|
llama_backend_free();
|
|
|
|
|
2023-03-10 19:40:58 +01:00
|
|
|
return 0;
|
|
|
|
}
|