aditya/llama.cpp
1
0
Fork 0
mirror of https://git.adityakumar.xyz/llama.cpp.git synced 2024-11-14 00:59:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
llama.cpp/llama.cpp

1816 lines
58 KiB
C++
Raw Normal View History

Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
#include "llama.h"
#include "ggml.h"
#include <cinttypes>
#include <fstream>
#include <random>
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
#include <map>
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
#include <unordered_map>
#include <queue>
#include <regex>
#include <cassert>
Add missing header for memcpy (#386) fixed: memcpy is not defined
2023-03-22 08:55:45 +00:00
#include <cstring>
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
#if defined(_WIN32) && !defined(_POSIX_MAPPED_FILES)
#define WIN32_LEAN_AND_MEAN
#include <Windows.h>
#else
#include <sys/types.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
Add mmap support for model files
2023-03-29 00:03:43 +00:00
#endif
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
#define Min(X, Y) ((Y) > (X) ? (X) : (Y))
#define Max(X, Y) ((Y) < (X) ? (X) : (Y))
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
#define LLAMA_USE_SCRATCH
#define LLAMA_MAX_SCRATCH_BUFFERS 16
#define LLAMA_ASSERT(x) \
do { \
if (!(x)) { \
fprintf(stderr, "LLAMA_ASSERT: %s:%d: %s\n", __FILE__, __LINE__, #x); \
abort(); \
} \
} while (0)
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// determine number of model parts based on the dimension
static const std::unordered_map<int, int> LLAMA_N_PARTS = {
{ 4096, 1 },
{ 5120, 2 },
{ 6656, 4 },
{ 8192, 8 },
};
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// available llama models
enum e_model {
MODEL_UNKNOWN,
MODEL_7B,
MODEL_13B,
MODEL_30B,
MODEL_65B,
};
static const size_t MB = 1024*1024;
// computed for n_ctx == 2048
// TODO: dynamically determine these sizes
// needs modifications in ggml
static const std::map<e_model, size_t> MEM_REQ_SCRATCH0 = {
{ MODEL_7B, 512ull*MB },
{ MODEL_13B, 512ull*MB },
{ MODEL_30B, 512ull*MB },
{ MODEL_65B, 512ull*MB },
};
static const std::map<e_model, size_t> MEM_REQ_SCRATCH1 = {
{ MODEL_7B, 512ull*MB },
{ MODEL_13B, 512ull*MB },
{ MODEL_30B, 512ull*MB },
{ MODEL_65B, 512ull*MB },
};
// 2*n_embd*n_ctx*n_layer*sizeof(float16)
static const std::map<e_model, size_t> MEM_REQ_KV_SELF = {
{ MODEL_7B, 1026ull*MB },
{ MODEL_13B, 1608ull*MB },
{ MODEL_30B, 3124ull*MB },
{ MODEL_65B, 5120ull*MB },
};
// this is mostly needed for temporary mul_mat buffers to dequantize the data
// not actually needed if BLAS is disabled
static const std::map<e_model, size_t> MEM_REQ_EVAL = {
{ MODEL_7B, 768ull*MB },
{ MODEL_13B, 1024ull*MB },
{ MODEL_30B, 1280ull*MB },
{ MODEL_65B, 1536ull*MB },
};
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// default hparams (LLaMA 7B)
struct llama_hparams {
int32_t n_vocab = 32000;
int32_t n_ctx = 512; // this is provided as user input?
int32_t n_embd = 4096;
int32_t n_mult = 256;
int32_t n_head = 32;
int32_t n_layer = 32;
int32_t n_rot = 64;
int32_t f16 = 1;
};
struct llama_layer {
// normalization
struct ggml_tensor * attention_norm;
// attention
struct ggml_tensor * wq;
struct ggml_tensor * wk;
struct ggml_tensor * wv;
struct ggml_tensor * wo;
// normalization
struct ggml_tensor * ffn_norm;
// ff
struct ggml_tensor * w1;
struct ggml_tensor * w2;
struct ggml_tensor * w3;
};
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
struct llama_kv_cache {
struct ggml_tensor * k;
struct ggml_tensor * v;
struct ggml_context * ctx;
std::vector<uint8_t> buf;
int n; // number of tokens currently in the cache
};
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct llama_model {
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
e_model type = MODEL_UNKNOWN;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
llama_hparams hparams;
struct ggml_tensor * tok_embeddings;
struct ggml_tensor * norm;
struct ggml_tensor * output;
std::vector<llama_layer> layers;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// context
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct ggml_context * ctx;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// key + value cache for the self attention
// TODO: move to llama_state
struct llama_kv_cache kv_self;
// the model memory buffer
std::vector<uint8_t> buf;
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
// model memory mapped file
Always initialize mm_addr and mm_length in llama_model
2023-03-29 06:53:14 +00:00
void * mm_addr = NULL;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
uint64_t mm_length = 0;
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// tensors
int n_loaded;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
std::unordered_map<std::string, struct ggml_tensor *> tensors;
};
struct llama_vocab {
using id = int32_t;
using token = std::string;
struct token_score {
token tok;
float score;
};
std::unordered_map<token, id> token_to_id;
std::vector<token_score> id_to_token;
};
struct llama_context {
std::mt19937 rng;
int64_t t_load_us = 0;
int64_t t_start_us = 0;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
bool has_evaluated_once = false;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
int64_t t_sample_us = 0;
int64_t t_eval_us = 0;
Add timings for the prompt evaluation (#478)
2023-03-25 14:34:23 +00:00
int64_t t_p_eval_us = 0;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
int32_t n_sample = 0; // number of tokens sampled
int32_t n_eval = 0; // number of eval calls
Add timings for the prompt evaluation (#478)
2023-03-25 14:34:23 +00:00
int32_t n_p_eval = 0; // number of tokens in eval calls for the prompt (with batch size > 1)
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
llama_model model;
llama_vocab vocab;
size_t mem_per_token = 0;
// decode output (2-dimensional array: [n_tokens][n_vocab])
std::vector<float> logits;
bool logits_all = false;
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
// input embedding (1-dimensional array: [n_embd])
std::vector<float> embedding;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// memory buffers used to evaluate the model
// TODO: move in llama_state
std::vector<uint8_t> buf_compute;
std::vector<uint8_t> buf_scratch[LLAMA_MAX_SCRATCH_BUFFERS];
int buf_last = 0;
size_t buf_max_size[LLAMA_MAX_SCRATCH_BUFFERS] = { 0 };
void use_buf(struct ggml_context * ctx, int i) {
#if defined(LLAMA_USE_SCRATCH)
size_t last_size = 0;
if (i == -1) {
last_size = ggml_set_scratch(ctx, { 0, 0, nullptr, });
} else {
auto & buf = buf_scratch[i];
last_size = ggml_set_scratch(ctx, { 0, buf.size(), buf.data(), });
}
if (buf_last >= 0) {
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
buf_max_size[buf_last] = Max(buf_max_size[buf_last], last_size);
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
}
buf_last = i;
#else
(void) i;
(void) ctx;
#endif
}
size_t get_buf_max_mem(int i) const {
#if defined(LLAMA_USE_SCRATCH)
return buf_max_size[i];
#else
(void) i;
return 0;
#endif
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
};
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
//
// kv cache
//
static bool kv_cache_init(
const struct llama_hparams & hparams,
struct llama_kv_cache & cache,
ggml_type wtype,
int n_ctx) {
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int n_mem = n_layer*n_ctx;
const int n_elements = n_embd*n_mem;
Fix crash for 65B model with pre-allocated memory (#485)
2023-03-25 04:38:14 +00:00
cache.buf.resize(2u*n_elements*ggml_type_size(wtype) + 2u*MB);
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
struct ggml_init_params params;
params.mem_size = cache.buf.size();
params.mem_buffer = cache.buf.data();
Add mmap support for model files
2023-03-29 00:03:43 +00:00
params.no_alloc = false;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
cache.ctx = ggml_init(params);
if (!cache.ctx) {
fprintf(stderr, "%s: failed to allocate memory for kv cache\n", __func__);
return false;
}
cache.k = ggml_new_tensor_1d(cache.ctx, wtype, n_elements);
cache.v = ggml_new_tensor_1d(cache.ctx, wtype, n_elements);
return true;
}
static void kv_cache_free(struct llama_kv_cache & cache) {
if (cache.ctx) {
ggml_free(cache.ctx);
cache.ctx = nullptr;
}
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct llama_context_params llama_context_default_params() {
struct llama_context_params result = {
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
/*.n_ctx =*/ 512,
/*.n_parts =*/ -1,
/*.seed =*/ 0,
/*.f16_kv =*/ false,
/*.logits_all =*/ false,
/*.vocab_only =*/ false,
/*.use_mlock =*/ false,
/*.embedding =*/ false,
/*.progress_callback =*/ nullptr,
/*.progress_callback_user_data =*/ nullptr,
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
};
return result;
}
//
// model loading
//
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
static void *mmap_file(const char *fname, uint64_t *mm_length) {
#if defined(_WIN32) && !defined(_POSIX_MAPPED_FILES)
HANDLE hFile = CreateFileA(fname,
GENERIC_READ,
FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
NULL,
OPEN_EXISTING,
FILE_ATTRIBUTE_NORMAL | FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,
NULL);
if (hFile == INVALID_HANDLE_VALUE) return 0;
Initial windows support (untested)
2023-03-29 20:22:36 +00:00
LARGE_INTEGER fileSize;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fileSize.QuadPart = -1;
Initial windows support (untested)
2023-03-29 20:22:36 +00:00
GetFileSizeEx(hFile, &fileSize);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
int64_t length = fileSize.QuadPart;
Initial windows support (untested)
2023-03-29 20:22:36 +00:00
HANDLE hMapping = CreateFileMappingA(hFile, NULL, PAGE_READONLY, 0, 0, NULL);
CloseHandle(hFile);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
if (!hMapping) return 0;
void *addr = MapViewOfFile(hMapping, FILE_MAP_READ, 0, 0, 0);
Initial windows support (untested)
2023-03-29 20:22:36 +00:00
CloseHandle(hMapping);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
if (!addr) return 0;
Add mmap support for model files
2023-03-29 00:03:43 +00:00
#else
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
int fd = open(fname, O_RDONLY);
if (fd == -1) return 0;
int64_t length = lseek(fd, 0, SEEK_END);
void *addr = mmap(NULL, length, PROT_READ, MAP_SHARED, fd, 0);
close(fd);
if (addr == MAP_FAILED) return 0;
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
#endif
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
*mm_length = length;
return addr;
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
}
static void munmap_file(void * addr, size_t length) {
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
#if defined(_WIN32) && !defined(_POSIX_MAPPED_FILES)
Initial windows support (untested)
2023-03-29 20:22:36 +00:00
UnmapViewOfFile(addr);
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
#else
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
munmap(addr, length);
Add mmap support for model files
2023-03-29 00:03:43 +00:00
#endif
}
Introduce GGML migration tool for new file format If you deleted your old Meta LLaMA .pth files, then the migrate-ggml-2023-03-30-pr613.py script will allow you to convert your old ggml files into the new mmap()'able format. See #613
2023-03-30 12:42:56 +00:00
static bool report_bad_magic(const char *path, uint32_t got, uint32_t want) {
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fprintf(stderr,
Introduce GGML migration tool for new file format If you deleted your old Meta LLaMA .pth files, then the migrate-ggml-2023-03-30-pr613.py script will allow you to convert your old ggml files into the new mmap()'able format. See #613
2023-03-30 12:42:56 +00:00
"%s: invalid model file (bad magic [got %#x want %#x])\n"
"\tyou most likely need to regenerate your ggml files\n"
"\tthe benefit is you'll get 10-100x faster load times\n"
"\tsee https://github.com/ggerganov/llama.cpp/issues/91\n"
"\tuse convert-pth-to-ggml.py to regenerate from original pth\n"
"\tuse migrate-ggml-2023-03-30-pr613.py if you deleted originals\n",
path, got, want);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
return false;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
static bool llama_model_load(
const std::string & fname,
llama_context & lctx,
int n_ctx,
int n_parts,
ggml_type memory_type,
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
bool vocab_only,
llama_progress_callback progress_callback,
void *progress_callback_user_data) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
fprintf(stderr, "%s: loading model from '%s' - please wait ...\n", __func__, fname.c_str());
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
lctx.t_start_us = ggml_time_us();
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
auto & model = lctx.model;
auto & vocab = lctx.vocab;
auto fin = std::ifstream(fname, std::ios::binary);
if (!fin) {
fprintf(stderr, "%s: failed to open '%s'\n", __func__, fname.c_str());
return false;
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
std::vector<char> f_buf(1024*1024);
fin.rdbuf()->pubsetbuf(f_buf.data(), f_buf.size());
fin.seekg(0, fin.end);
const size_t file_size = fin.tellg();
fin.seekg(0);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// verify magic
{
uint32_t magic;
fin.read((char *) &magic, sizeof(magic));
if (magic == LLAMA_FILE_MAGIC_UNVERSIONED) {
py : add temporary script to convert old ggml files to newer version (#539) Co-authored-by: Jakub Horak <jakub.horak@ibawizard.net>
2023-03-28 17:55:42 +00:00
fprintf(stderr, "%s: invalid model file '%s' (too old, regenerate your model files or convert them with convert-unversioned-ggml-to-ggml.py!)\n",
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
__func__, fname.c_str());
return false;
}
if (magic != LLAMA_FILE_MAGIC) {
Introduce GGML migration tool for new file format If you deleted your old Meta LLaMA .pth files, then the migrate-ggml-2023-03-30-pr613.py script will allow you to convert your old ggml files into the new mmap()'able format. See #613
2023-03-30 12:42:56 +00:00
return report_bad_magic(fname.c_str(), magic, LLAMA_FILE_MAGIC);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
uint32_t format_version;
fin.read((char *) &format_version, sizeof(format_version));
if (format_version != LLAMA_FILE_VERSION) {
fprintf(stderr, "%s: invalid model file '%s' (unsupported format version %" PRIu32 ", expected %d)\n",
__func__, fname.c_str(), format_version, LLAMA_FILE_VERSION);
return false;
}
}
int n_ff = 0;
// load hparams
{
auto & hparams = model.hparams;
fin.read((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
//fin.read((char *) &hparams.n_ctx, sizeof(hparams.n_ctx));
fin.read((char *) &hparams.n_embd, sizeof(hparams.n_embd));
fin.read((char *) &hparams.n_mult, sizeof(hparams.n_mult));
fin.read((char *) &hparams.n_head, sizeof(hparams.n_head));
fin.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
fin.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
fin.read((char *) &hparams.f16, sizeof(hparams.f16));
hparams.n_ctx = n_ctx;
n_ff = ((2*(4*hparams.n_embd)/3 + hparams.n_mult - 1)/hparams.n_mult)*hparams.n_mult;
if (n_parts < 1) {
n_parts = LLAMA_N_PARTS.at(hparams.n_embd);
}
// temp warning to tell the user to use "--n_parts"
if (hparams.f16 == 4 && n_parts != 1) {
fprintf(stderr, "%s: GPTQ model detected - are you sure n_parts should be %d? we normally expect it to be 1\n", __func__, n_parts);
fprintf(stderr, "%s: use '--n_parts 1' if necessary\n", __func__);
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
if (hparams.n_layer == 32) {
model.type = e_model::MODEL_7B;
}
if (hparams.n_layer == 40) {
model.type = e_model::MODEL_13B;
}
if (hparams.n_layer == 60) {
model.type = e_model::MODEL_30B;
}
if (hparams.n_layer == 80) {
model.type = e_model::MODEL_65B;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
fprintf(stderr, "%s: n_vocab = %d\n", __func__, hparams.n_vocab);
fprintf(stderr, "%s: n_ctx = %d\n", __func__, hparams.n_ctx);
fprintf(stderr, "%s: n_embd = %d\n", __func__, hparams.n_embd);
fprintf(stderr, "%s: n_mult = %d\n", __func__, hparams.n_mult);
fprintf(stderr, "%s: n_head = %d\n", __func__, hparams.n_head);
fprintf(stderr, "%s: n_layer = %d\n", __func__, hparams.n_layer);
fprintf(stderr, "%s: n_rot = %d\n", __func__, hparams.n_rot);
fprintf(stderr, "%s: f16 = %d\n", __func__, hparams.f16);
fprintf(stderr, "%s: n_ff = %d\n", __func__, n_ff);
fprintf(stderr, "%s: n_parts = %d\n", __func__, n_parts);
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
fprintf(stderr, "%s: type = %d\n", __func__, model.type);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
// load vocab
{
std::string word;
vocab.id_to_token.resize(model.hparams.n_vocab);
std::vector<char> tmp(64);
for (int i = 0; i < model.hparams.n_vocab; i++) {
uint32_t len;
fin.read((char *) &len, sizeof(len));
word.resize(len);
if (len > 0) {
tmp.resize(len);
fin.read(tmp.data(), len);
word.assign(tmp.data(), len);
} else {
word.clear();
}
float score;
fin.read((char *) &score, sizeof(score));
vocab.token_to_id[word] = i;
auto &tok_score = vocab.id_to_token[i];
tok_score.tok = word;
tok_score.score = score;
}
}
if (vocab_only) {
return true;
}
// for the big tensors, we have the option to store the data in 16-bit floats or quantized
// in order to save memory and also to speed up the computation
// wtype is for per-layer weights, while vtype is for other weights
ggml_type wtype, vtype;
switch (model.hparams.f16) {
case 0: wtype = vtype = GGML_TYPE_F32; break;
case 1: wtype = vtype = GGML_TYPE_F16; break;
case 2: wtype = vtype = GGML_TYPE_Q4_0; break;
case 3: wtype = vtype = GGML_TYPE_Q4_1; break;
case 4: wtype = GGML_TYPE_Q4_1; vtype = GGML_TYPE_F16; break;
default:
{
fprintf(stderr, "%s: invalid model file '%s' (bad f16 value %d)\n",
__func__, fname.c_str(), model.hparams.f16);
return false;
}
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
// map model into memory
char *mm_addr = NULL;
model.mm_addr = mmap_file(fname.c_str(), &model.mm_length);
if (model.mm_addr == NULL) {
fprintf(stderr, "%s: failed to mmap '%s'\n", __func__, fname.c_str());
return false;
Add mmap support for model files
2023-03-29 00:03:43 +00:00
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
mm_addr = (char *)model.mm_addr;
fprintf(stderr, "%s: ggml map size = %6.2f MB\n", __func__, model.mm_length/(1024.0*1024.0));
Add mmap support for model files
2023-03-29 00:03:43 +00:00
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
auto & ctx = model.ctx;
size_t ctx_size = 0;
{
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
const auto &hparams = model.hparams;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
const int n_layer = hparams.n_layer;
ctx_size += (5 + 10*n_layer)*256; // object overhead
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fprintf(stderr, "%s: ggml ctx size = %6.2f KB\n", __func__, ctx_size/1024.0);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
// print memory requirements
{
const size_t scale = memory_type == GGML_TYPE_F32 ? 2 : 1;
// this is the total memory required to run the inference
const size_t mem_required =
ctx_size +
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
model.mm_length +
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
MEM_REQ_SCRATCH0.at(model.type) +
MEM_REQ_SCRATCH1.at(model.type) +
MEM_REQ_EVAL.at (model.type);
// this is the memory required by one llama_state
const size_t mem_required_state =
scale*MEM_REQ_KV_SELF.at(model.type);
fprintf(stderr, "%s: mem required = %7.2f MB (+ %7.2f MB per state)\n", __func__,
mem_required / 1024.0 / 1024.0, mem_required_state / 1024.0 / 1024.0);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// create the ggml context
{
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
lctx.model.buf.resize(ctx_size);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct ggml_init_params params = {
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
/*.mem_size =*/ lctx.model.buf.size(),
/*.mem_buffer =*/ lctx.model.buf.data(),
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
/*.no_alloc =*/ true,
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
};
model.ctx = ggml_init(params);
if (!model.ctx) {
fprintf(stderr, "%s: ggml_init() failed\n", __func__);
return false;
}
}
// prepare memory for the weights
{
const auto & hparams = model.hparams;
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int n_vocab = hparams.n_vocab;
model.layers.resize(n_layer);
model.tok_embeddings = ggml_new_tensor_2d(ctx, vtype, n_embd, n_vocab);
model.norm = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
model.output = ggml_new_tensor_2d(ctx, vtype, n_embd, n_vocab);
// map by name
model.tensors["tok_embeddings.weight"] = model.tok_embeddings;
model.tensors["norm.weight"] = model.norm;
model.tensors["output.weight"] = model.output;
for (int i = 0; i < n_layer; ++i) {
auto & layer = model.layers[i];
layer.attention_norm = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.wq = ggml_new_tensor_2d(ctx, wtype, n_embd, n_embd);
layer.wk = ggml_new_tensor_2d(ctx, wtype, n_embd, n_embd);
layer.wv = ggml_new_tensor_2d(ctx, wtype, n_embd, n_embd);
layer.wo = ggml_new_tensor_2d(ctx, wtype, n_embd, n_embd);
layer.ffn_norm = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
layer.w1 = ggml_new_tensor_2d(ctx, wtype, n_embd, n_ff);
layer.w2 = ggml_new_tensor_2d(ctx, wtype, n_ff, n_embd);
layer.w3 = ggml_new_tensor_2d(ctx, wtype, n_embd, n_ff);
// map by name
model.tensors["layers." + std::to_string(i) + ".attention_norm.weight"] = layer.attention_norm;
model.tensors["layers." + std::to_string(i) + ".attention.wq.weight"] = layer.wq;
model.tensors["layers." + std::to_string(i) + ".attention.wk.weight"] = layer.wk;
model.tensors["layers." + std::to_string(i) + ".attention.wv.weight"] = layer.wv;
model.tensors["layers." + std::to_string(i) + ".attention.wo.weight"] = layer.wo;
model.tensors["layers." + std::to_string(i) + ".ffn_norm.weight"] = layer.ffn_norm;
model.tensors["layers." + std::to_string(i) + ".feed_forward.w1.weight"] = layer.w1;
model.tensors["layers." + std::to_string(i) + ".feed_forward.w2.weight"] = layer.w2;
model.tensors["layers." + std::to_string(i) + ".feed_forward.w3.weight"] = layer.w3;
}
}
std::vector<uint8_t> tmp;
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
if (progress_callback) {
progress_callback(0.0, progress_callback_user_data);
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fprintf(stderr, "%s: loading tensors from '%s'\n", __func__, fname.c_str());
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
// load weights
{
size_t total_size = 0;
model.n_loaded = 0;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
while (true) {
int32_t n_dims;
int32_t length;
int32_t ftype;
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fin.read(reinterpret_cast<char *>(&n_dims), sizeof(n_dims));
fin.read(reinterpret_cast<char *>(&length), sizeof(length));
fin.read(reinterpret_cast<char *>(&ftype), sizeof(ftype));
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
if (fin.eof()) {
break;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
int32_t nelements = 1;
int32_t ne[2] = { 1, 1 };
for (int i = 0; i < n_dims; ++i) {
fin.read(reinterpret_cast<char *>(&ne[i]), sizeof(ne[i]));
nelements *= ne[i];
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
std::string name(length, 0);
fin.read(&name[0], length);
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
if (model.tensors.find(name.data()) == model.tensors.end()) {
fprintf(stderr, "%s: unknown tensor '%s' in model file\n", __func__, name.data());
return false;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
auto tensor = model.tensors[name.data()];
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
if (ggml_nelements(tensor) != nelements) {
fprintf(stderr, "%s: tensor '%s' has wrong size in model file\n", __func__, name.data());
return false;
}
if (tensor->ne[0] != ne[0] || tensor->ne[1] != ne[1]) {
fprintf(stderr, "%s: tensor '%s' has wrong shape in model file: got [%d, %d], expected [%d, %d]\n",
__func__, name.data(), tensor->ne[0], tensor->ne[1], ne[0], ne[1]);
return false;
}
if (0) {
static const char * ftype_str[] = { "f32", "f16", "q4_0", "q4_1", };
fprintf(stderr, "%24s - [%5d, %5d], type = %6s\n", name.data(), ne[0], ne[1], ftype_str[ftype]);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
switch (ftype) {
case 0: // f32
case 1: // f16
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
break;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
case 2: // q4_0
case 3: // q4_1
assert(ne[0] % 64 == 0);
break;
default:
fprintf(stderr, "%s: unknown ftype %d in model file\n", __func__, ftype);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return false;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
};
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
// load the tensor data into memory without copying or reading it
size_t offset = fin.tellg();
size_t tensor_data_size = ggml_nbytes(tensor);
offset = (offset + 31) & -32;
tensor->data = mm_addr + offset;
fin.seekg(offset + tensor_data_size);
total_size += tensor_data_size;
model.n_loaded++;
// progress
if (progress_callback) {
double current_progress = size_t(fin.tellg()) / double(file_size);
progress_callback(current_progress, progress_callback_user_data);
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
fin.close();
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
fprintf(stderr, "%s: model size = %8.2f MB / num tensors = %d\n", __func__, total_size/1024.0/1024.0, model.n_loaded);
if (model.n_loaded == 0) {
fprintf(stderr, "%s: WARN no tensors loaded from model file - assuming empty model for testing\n", __func__);
} else if (model.n_loaded != (int) model.tensors.size()) {
fprintf(stderr, "%s: ERROR not all tensors loaded from model file - expected %zu, got %d\n", __func__, model.tensors.size(), model.n_loaded);
return false;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
// loading time will be recalculate after the first eval, so
// we take page faults deferred by mmap() into consideration
lctx.t_load_us = ggml_time_us() - lctx.t_start_us;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
if (progress_callback) {
progress_callback(1.0, progress_callback_user_data);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return true;
}
// evaluate the transformer
//
// - lctx: llama context
// - tokens: new batch of tokens to process
// - n_past: the context size so far
// - n_threads: number of threads to use
//
static bool llama_eval_internal(
llama_context & lctx,
const llama_token * tokens,
const int n_tokens,
const int n_past,
const int n_threads) {
const int64_t t_start_us = ggml_time_us();
const int N = n_tokens;
const auto & model = lctx.model;
const auto & hparams = model.hparams;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
auto & kv_self = model.kv_self;
LLAMA_ASSERT(!!kv_self.ctx);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer;
const int n_ctx = hparams.n_ctx;
const int n_head = hparams.n_head;
const int n_vocab = hparams.n_vocab;
const int n_rot = hparams.n_embd/hparams.n_head;
auto & mem_per_token = lctx.mem_per_token;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
auto & buf_compute = lctx.buf_compute;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct ggml_init_params params = {
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
/*.mem_size =*/ buf_compute.size(),
/*.mem_buffer =*/ buf_compute.data(),
Add mmap support for model files
2023-03-29 00:03:43 +00:00
/*.no_alloc =*/ false,
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
};
struct ggml_context * ctx0 = ggml_init(params);
Don't interefe with BLAS for large prompts by running only 1 thread
2023-03-25 15:03:10 +00:00
// for big prompts, if BLAS is enabled, it is better to use only one thread
// otherwise, the threads are spin-lock waiting for the BLAS calls and are degrading the performance
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
ggml_cgraph gf = {};
llama : use the same threshold for OpenBLAS and ggml thread limiting (#577)
2023-03-29 16:10:07 +00:00
gf.n_threads = N >= 32 && ggml_cpu_has_blas() ? 1 : n_threads;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
memcpy(embd->data, tokens, N*ggml_element_size(embd));
struct ggml_tensor * inpL = ggml_get_rows(ctx0, model.tok_embeddings, embd);
for (int il = 0; il < n_layer; ++il) {
struct ggml_tensor * inpSA = inpL;
struct ggml_tensor * cur;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
lctx.use_buf(ctx0, 0);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// norm
{
cur = ggml_rms_norm(ctx0, inpL);
// cur = attention_norm*cur
cur = ggml_mul(ctx0,
ggml_repeat(ctx0, model.layers[il].attention_norm, cur),
cur);
}
// self-attention
{
struct ggml_tensor * Qcur = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
struct ggml_tensor * Kcur = ggml_mul_mat(ctx0, model.layers[il].wk, cur);
struct ggml_tensor * Vcur = ggml_mul_mat(ctx0, model.layers[il].wv, cur);
// store key and value to memory
if (N >= 1) {
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
struct ggml_tensor * k = ggml_view_1d(ctx0, kv_self.k, N*n_embd, (ggml_element_size(kv_self.k)*n_embd)*(il*n_ctx + n_past));
struct ggml_tensor * v = ggml_view_1d(ctx0, kv_self.v, N*n_embd, (ggml_element_size(kv_self.v)*n_embd)*(il*n_ctx + n_past));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Kcur, k));
ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Vcur, v));
}
// Q = Qcur.contiguous().view(n_embd/n_head, n_head, N).permute(0, 2, 1, 3)
struct ggml_tensor * Q =
ggml_permute(ctx0,
ggml_rope(ctx0,
ggml_cpy(ctx0,
Qcur,
ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_embd/n_head, n_head, N)),
n_past, n_rot, 0),
0, 2, 1, 3);
// K = Kmem.view(n_embd/n_head, n_head, n_past + N).permute(0, 2, 1, 3)
struct ggml_tensor * K =
ggml_permute(ctx0,
ggml_rope(ctx0,
ggml_reshape_3d(ctx0,
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
ggml_view_1d(ctx0, kv_self.k, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(kv_self.k)*n_embd),
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
n_embd/n_head, n_head, n_past + N),
n_past, n_rot, 1),
0, 2, 1, 3);
// K * Q
struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q);
// KQ_scaled = KQ / sqrt(n_embd/n_head)
struct ggml_tensor * KQ_scaled =
ggml_scale(ctx0,
KQ,
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
ggml_new_f32(ctx0, 1.0f/sqrtf(float(n_embd)/n_head)));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// KQ_masked = mask_past(KQ_scaled)
struct ggml_tensor * KQ_masked = ggml_diag_mask_inf(ctx0, KQ_scaled, n_past);
// KQ = soft_max(KQ_masked)
struct ggml_tensor * KQ_soft_max = ggml_soft_max(ctx0, KQ_masked);
// V_trans = Vmem.view(n_embd/n_head, n_head, n_past + N).permute(1, 2, 0, 3).contiguous()
struct ggml_tensor * V_trans =
Avoid the transposed X branch in the Z = X * Y matrix multiplication (#439) Should make results reproducible for different number of threads and batch sizes
2023-03-23 21:22:01 +00:00
ggml_cpy(ctx0,
ggml_permute(ctx0,
ggml_reshape_3d(ctx0,
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
ggml_view_1d(ctx0, kv_self.v, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(kv_self.v)*n_embd),
Avoid the transposed X branch in the Z = X * Y matrix multiplication (#439) Should make results reproducible for different number of threads and batch sizes
2023-03-23 21:22:01 +00:00
n_embd/n_head, n_head, n_past + N),
1, 2, 0, 3),
Fix nasty bug in ggml_compute_forward_mul_mat_f32() and reenable BLAS
2023-03-25 14:09:54 +00:00
ggml_new_tensor_3d(ctx0, kv_self.v->type, n_past + N, n_embd/n_head, n_head));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// KQV = transpose(V) * KQ_soft_max
struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V_trans, KQ_soft_max);
// KQV_merged = KQV.permute(0, 2, 1, 3)
struct ggml_tensor * KQV_merged = ggml_permute(ctx0, KQV, 0, 2, 1, 3);
// cur = KQV_merged.contiguous().view(n_embd, N)
cur = ggml_cpy(ctx0,
KQV_merged,
ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N));
// projection (no bias)
cur = ggml_mul_mat(ctx0,
model.layers[il].wo,
cur);
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
lctx.use_buf(ctx0, 1);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
struct ggml_tensor * inpFF = ggml_add(ctx0, cur, inpSA);
// feed-forward network
{
// norm
{
cur = ggml_rms_norm(ctx0, inpFF);
// cur = ffn_norm*cur
cur = ggml_mul(ctx0,
ggml_repeat(ctx0, model.layers[il].ffn_norm, cur),
cur);
}
struct ggml_tensor * tmp = ggml_mul_mat(ctx0,
model.layers[il].w3,
cur);
cur = ggml_mul_mat(ctx0,
model.layers[il].w1,
cur);
// SILU activation
cur = ggml_silu(ctx0, cur);
cur = ggml_mul(ctx0, cur, tmp);
cur = ggml_mul_mat(ctx0,
model.layers[il].w2,
cur);
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
cur = ggml_add(ctx0, cur, inpFF);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// input for next layer
inpL = cur;
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
lctx.use_buf(ctx0, 0);
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
// used at the end to optionally extract the embeddings
struct ggml_tensor * embeddings = NULL;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// norm
{
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
inpL = ggml_rms_norm(ctx0, inpL);
// inpL = norm*inpL
inpL = ggml_mul(ctx0,
ggml_repeat(ctx0, model.norm, inpL),
inpL);
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
embeddings = inpL;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
// lm_head
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
inpL = ggml_mul_mat(ctx0, model.output, inpL);
lctx.use_buf(ctx0, -1);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// logits -> probs
//inpL = ggml_soft_max(ctx0, inpL);
// run the computation
ggml_build_forward_expand(&gf, inpL);
ggml_graph_compute (ctx0, &gf);
//if (n_past%100 == 0) {
// ggml_graph_print (&gf);
// ggml_graph_dump_dot(&gf, NULL, "gpt-2.dot");
//}
//embd_w.resize(n_vocab*N);
//memcpy(embd_w.data(), ggml_get_data(inpL), sizeof(float)*n_vocab*N);
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
// extract logits
{
auto & logits_out = lctx.logits;
if (lctx.logits_all) {
logits_out.resize(n_vocab * N);
memcpy(logits_out.data(), (float *) ggml_get_data(inpL), sizeof(float)*n_vocab*N);
} else {
// return result for just the last token
logits_out.resize(n_vocab);
memcpy(logits_out.data(), (float *) ggml_get_data(inpL) + (n_vocab*(N-1)), sizeof(float)*n_vocab);
}
}
// extract embeddings
if (lctx.embedding.size()) {
auto & embedding_out = lctx.embedding;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
embedding_out.resize(n_embd);
memcpy(embedding_out.data(), (float *) ggml_get_data(embeddings) + (n_embd*(N - 1)), sizeof(float)*n_embd);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
Revert "Fix memory allocation issues and seg faults" This reverts commit 4870e455b3653f7d7769fa5772b2c90ffad088df. Will provide the correct fix later
2023-03-24 04:22:28 +00:00
if (mem_per_token == 0) {
mem_per_token = ggml_used_mem(ctx0)/N;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
#if 0
printf("\n%s: used_mem = %.3f MB, scratch -- %.3f MB %.3f MB\n", __func__,
ggml_used_mem(ctx0)/1024.0/1024.0,
lctx.get_buf_max_mem(0)/1024.0/1024.0,
lctx.get_buf_max_mem(1)/1024.0/1024.0);
#endif
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
ggml_free(ctx0);
// measure the performance only for the single-token evals
if (N == 1) {
lctx.t_eval_us += ggml_time_us() - t_start_us;
lctx.n_eval++;
}
Add timings for the prompt evaluation (#478)
2023-03-25 14:34:23 +00:00
else if (N > 1) {
lctx.t_p_eval_us += ggml_time_us() - t_start_us;
lctx.n_p_eval += N;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return true;
}
//
// tokenizer
//
static size_t utf8_len(char src) {
const size_t lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 };
uint8_t highbits = static_cast<uint8_t>(src) >> 4;
return lookup[highbits];
}
struct llama_sp_symbol {
using index = int;
index prev;
index next;
const char * text;
size_t n;
};
struct llama_sp_bigram {
struct comparator {
bool operator()(llama_sp_bigram & l, llama_sp_bigram & r) {
return (l.score < r.score) || (l.score == r.score && l.left > r.left);
}
};
using queue_storage = std::vector<llama_sp_bigram>;
using queue = std::priority_queue<llama_sp_bigram, queue_storage, comparator>;
llama_sp_symbol::index left;
llama_sp_symbol::index right;
float score;
size_t size;
};
// original implementation:
// https://github.com/ggerganov/llama.cpp/commit/074bea2eb1f1349a0118239c4152914aecaa1be4
struct llama_tokenizer {
llama_tokenizer(const llama_vocab & vocab): vocab_(vocab) {}
void tokenize(const std::string & text, std::vector<llama_vocab::id> & output) {
// split string into utf8 chars
int index = 0;
size_t offs = 0;
while (offs < text.size()) {
llama_sp_symbol sym;
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
size_t char_len = Min(text.size() - offs, utf8_len(text[offs]));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
sym.text = text.c_str() + offs;
sym.n = char_len;
offs += char_len;
sym.prev = index - 1;
sym.next = offs == text.size() ? -1 : index + 1;
index++;
symbols_.emplace_back(std::move(sym));
}
// seed the work queue with all possible 2-character tokens.
for (size_t i = 1; i < symbols_.size(); ++i) {
try_add_bigram(i - 1, i);
}
// keep substituting the highest frequency pairs for as long as we can.
while (!work_queue_.empty()) {
auto bigram = work_queue_.top();
work_queue_.pop();
auto & left_sym = symbols_[bigram.left];
auto & right_sym = symbols_[bigram.right];
// if one of the symbols already got merged, skip it.
if (left_sym.n == 0 || right_sym.n == 0 ||
left_sym.n + right_sym.n != bigram.size) {
continue;
}
// merge the right sym into the left one
left_sym.n += right_sym.n;
right_sym.n = 0;
//printf("left = '%*s' size = %zu\n", (int) left_sym.n, left_sym.text, bigram.size);
// remove the right sym from the chain
left_sym.next = right_sym.next;
if (right_sym.next >= 0) {
symbols_[right_sym.next].prev = bigram.left;
}
// find more substitutions
try_add_bigram(left_sym.prev, bigram.left);
try_add_bigram(bigram.left, left_sym.next);
}
for (int i = 0; i != -1; i = symbols_[i].next) {
auto & symbol = symbols_[i];
auto token = vocab_.token_to_id.find(std::string(symbol.text, symbol.n));
if (token == vocab_.token_to_id.end()) {
// output any symbols that did not form tokens as bytes.
for (int j = 0; j < (int) symbol.n; ++j) {
llama_vocab::id token_id = static_cast<uint8_t>(symbol.text[j]) + 3;
output.push_back(token_id);
}
} else {
output.push_back((*token).second);
}
}
}
private:
void try_add_bigram(int left, int right) {
if (left == -1 || right == -1) {
return;
}
const std::string text = std::string(symbols_[left].text, symbols_[left].n + symbols_[right].n);
auto token = vocab_.token_to_id.find(text);
if (token == vocab_.token_to_id.end()) {
return;
}
if (static_cast<size_t>((*token).second) >= vocab_.id_to_token.size()) {
return;
}
const auto &tok_score = vocab_.id_to_token[(*token).second];
llama_sp_bigram bigram;
bigram.left = left;
bigram.right = right;
bigram.score = tok_score.score;
bigram.size = text.size();
work_queue_.push(bigram);
}
const llama_vocab & vocab_;
std::vector<llama_sp_symbol> symbols_;
llama_sp_bigram::queue work_queue_;
};
static std::vector<llama_vocab::id> llama_tokenize(const llama_vocab & vocab, const std::string & text, bool bos) {
llama_tokenizer tokenizer(vocab);
std::vector<llama_vocab::id> output;
if (text.size() == 0) {
return output;
}
if (bos) {
output.push_back(1);
}
tokenizer.tokenize(text, output);
return output;
}
//
// sampling
//
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
static void sample_top_k(std::vector<std::pair<float, llama_vocab::id>> & logits_id, int top_k) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
// find the top k tokens
std::partial_sort(
logits_id.begin(),
logits_id.begin() + top_k, logits_id.end(),
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
[](const std::pair<float, llama_vocab::id> & a, const std::pair<float, llama_vocab::id> & b) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return a.first > b.first;
});
logits_id.resize(top_k);
}
static llama_vocab::id llama_sample_top_p_top_k(
llama_context & lctx,
const std::vector<llama_vocab::id> & last_n_tokens,
int top_k,
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
float top_p,
float temp,
float repeat_penalty) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
auto & rng = lctx.rng;
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
const int n_logits = lctx.model.hparams.n_vocab;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
const auto & logits = lctx.logits;
const auto * plogits = logits.data() + logits.size() - n_logits;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
std::vector<std::pair<float, llama_vocab::id>> logits_id;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
logits_id.reserve(n_logits);
{
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
const float scale = 1.0f/temp;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
for (int i = 0; i < n_logits; ++i) {
// repetition penalty from ctrl paper (https://arxiv.org/abs/1909.05858)
// credit https://github.com/facebookresearch/llama/compare/main...shawwn:llama:main
if (std::find(last_n_tokens.begin(), last_n_tokens.end(), i) != last_n_tokens.end()) {
// if score < 0 then repetition penalty has to multiplied to reduce the previous token probability
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
if (plogits[i] < 0.0f) {
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
logits_id.push_back(std::make_pair(plogits[i]*scale*repeat_penalty, i));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
} else {
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
logits_id.push_back(std::make_pair(plogits[i]*scale/repeat_penalty, i));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
} else {
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
logits_id.push_back(std::make_pair(plogits[i]*scale, i));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
}
}
sample_top_k(logits_id, top_k);
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
float maxl = -std::numeric_limits<float>::infinity();
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
for (const auto & kv : logits_id) {
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
maxl = Max(maxl, kv.first);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
// compute probs for the top k tokens
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
std::vector<float> probs;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
probs.reserve(logits_id.size());
double sum = 0.0;
for (const auto & kv : logits_id) {
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
const float p = expf(kv.first - maxl);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
probs.push_back(p);
sum += p;
}
// normalize the probs
for (auto & p : probs) {
p /= sum;
}
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
if (top_p < 1.0) {
double cumsum = 0.0;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
for (int i = 0; i < (int) probs.size(); i++) {
cumsum += probs[i];
if (cumsum >= top_p) {
probs.resize(i + 1);
logits_id.resize(i + 1);
break;
}
}
cumsum = 1.0/cumsum;
for (int i = 0; i < (int) probs.size(); i++) {
probs[i] *= cumsum;
}
}
//printf("\n");
//for (int i = 0; i < (int) 10; i++) {
// printf("%d: '%s' %f\n", i, vocab.id_to_token.at(logits_id[i].second).c_str(), probs[i]);
//}
//printf("\n\n");
//exit(0);
std::discrete_distribution<> dist(probs.begin(), probs.end());
int idx = dist(rng);
return logits_id[idx].second;
}
//
// quantization
//
// TODO: reuse code from the llama_model_load() somehow
ggml : introduce structs for the q4 data blocks (#356) * Introduce structs for the q4 data blocks * ggml : rename quant struct variables + fix ARM_NEON --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 15:56:03 +00:00
static bool llama_model_quantize_internal(const std::string & fname_inp, const std::string & fname_out, int itype) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
ggml_type type = GGML_TYPE_Q4_1;
switch (itype) {
case 2: type = GGML_TYPE_Q4_0; break;
case 3: type = GGML_TYPE_Q4_1; break;
default: fprintf(stderr, "%s: invalid quantization type %d\n", __func__, itype); return 1;
};
if (type != GGML_TYPE_Q4_0 && type != GGML_TYPE_Q4_1) {
fprintf(stderr, "%s: invalid quantization type %d\n", __func__, type);
return false;
}
llama_vocab vocab;
printf("%s: loading model from '%s'\n", __func__, fname_inp.c_str());
auto finp = std::ifstream(fname_inp, std::ios::binary);
if (!finp) {
fprintf(stderr, "%s: failed to open '%s' for reading\n", __func__, fname_inp.c_str());
return false;
}
auto fout = std::ofstream(fname_out, std::ios::binary);
if (!fout) {
fprintf(stderr, "%s: failed to open '%s' for writing\n", __func__, fname_out.c_str());
return false;
}
// verify magic
{
uint32_t magic;
finp.read((char *) &magic, sizeof(magic));
if (magic == LLAMA_FILE_MAGIC_UNVERSIONED) {
fprintf(stderr, "%s: invalid model file '%s' (too old, regenerate your model files!)\n",
__func__, fname_inp.c_str());
return false;
}
if (magic != LLAMA_FILE_MAGIC) {
Introduce GGML migration tool for new file format If you deleted your old Meta LLaMA .pth files, then the migrate-ggml-2023-03-30-pr613.py script will allow you to convert your old ggml files into the new mmap()'able format. See #613
2023-03-30 12:42:56 +00:00
return report_bad_magic(fname_inp.c_str(), magic, LLAMA_FILE_MAGIC);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
fout.write((char *) &magic, sizeof(magic));
uint32_t format_version;
finp.read((char *) &format_version, sizeof(format_version));
if (format_version != LLAMA_FILE_VERSION) {
fprintf(stderr, "%s: invalid model file '%s' (unsupported format version %" PRIu32 ", expected %d)\n",
__func__, fname_inp.c_str(), format_version, LLAMA_FILE_VERSION);
return false;
}
fout.write((char *) &format_version, sizeof(format_version));
}
llama_hparams hparams;
// load hparams
{
finp.read((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
//finp.read((char *) &hparams.n_ctx, sizeof(hparams.n_ctx));
finp.read((char *) &hparams.n_embd, sizeof(hparams.n_embd));
finp.read((char *) &hparams.n_mult, sizeof(hparams.n_mult));
finp.read((char *) &hparams.n_head, sizeof(hparams.n_head));
finp.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
finp.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
finp.read((char *) &hparams.f16, sizeof(hparams.f16));
printf("%s: n_vocab = %d\n", __func__, hparams.n_vocab);
printf("%s: n_ctx = %d\n", __func__, hparams.n_ctx);
printf("%s: n_embd = %d\n", __func__, hparams.n_embd);
printf("%s: n_mult = %d\n", __func__, hparams.n_mult);
printf("%s: n_head = %d\n", __func__, hparams.n_head);
printf("%s: n_layer = %d\n", __func__, hparams.n_layer);
printf("%s: f16 = %d\n", __func__, hparams.f16);
fout.write((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
//fout.write((char *) &hparams.n_ctx, sizeof(hparams.n_ctx));
fout.write((char *) &hparams.n_embd, sizeof(hparams.n_embd));
fout.write((char *) &hparams.n_mult, sizeof(hparams.n_mult));
fout.write((char *) &hparams.n_head, sizeof(hparams.n_head));
fout.write((char *) &hparams.n_layer, sizeof(hparams.n_layer));
fout.write((char *) &hparams.n_rot, sizeof(hparams.n_rot));
fout.write((char *) &itype, sizeof(hparams.f16));
}
// load vocab
{
const int32_t n_vocab = hparams.n_vocab;
if (n_vocab != hparams.n_vocab) {
fprintf(stderr, "%s: invalid model file '%s' (bad vocab size %d != %d)\n",
__func__, fname_inp.c_str(), n_vocab, hparams.n_vocab);
return false;
}
llama : fix compile warnings when reading the vocab
2023-03-29 19:13:12 +00:00
std::vector<char> word(32);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
vocab.id_to_token.resize(n_vocab);
for (int i = 0; i < n_vocab; i++) {
uint32_t len;
finp.read ((char *) &len, sizeof(len));
fout.write((char *) &len, sizeof(len));
word.resize(len);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
finp.read ((char *) &word[0], len);
fout.write((char *) &word[0], len);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
float score;
finp.read ((char *) &score, sizeof(score));
fout.write((char *) &score, sizeof(score));
llama : fix compile warnings when reading the vocab
2023-03-29 19:13:12 +00:00
vocab.token_to_id[word.data()] = i;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
auto &tok_score = vocab.id_to_token[i];
llama : fix compile warnings when reading the vocab
2023-03-29 19:13:12 +00:00
tok_score.tok = word.data();
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
tok_score.score = score;
}
}
// load weights
{
size_t total_size_org = 0;
size_t total_size_new = 0;
std::vector<float> work;
std::vector<uint8_t> data_u8;
std::vector<ggml_fp16_t> data_f16;
std::vector<float> data_f32;
std::vector<int64_t> hist_all(1 << 4, 0);
while (true) {
int32_t n_dims;
int32_t length;
int32_t ftype;
finp.read(reinterpret_cast<char *>(&n_dims), sizeof(n_dims));
finp.read(reinterpret_cast<char *>(&length), sizeof(length));
finp.read(reinterpret_cast<char *>(&ftype), sizeof(ftype));
if (finp.eof()) {
break;
}
int32_t nelements = 1;
int32_t ne[2] = { 1, 1 };
for (int i = 0; i < n_dims; ++i) {
finp.read (reinterpret_cast<char *>(&ne[i]), sizeof(ne[i]));
nelements *= ne[i];
}
std::string name(length, 0);
finp.read (&name[0], length);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
{
// ensure tensor data is aligned
uint64_t offset = finp.tellg();
offset = (offset + 31) & -32;
finp.seekg(offset);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
{
static const char * ftype_str[] = { "f32", "f16", "q4_0", "q4_1", };
printf("%48s - [%5d, %5d], type = %6s ", name.data(), ne[0], ne[1], ftype_str[ftype]);
}
// regexes of tensor names to be quantized
const std::vector<std::string> k_names = {
".*weight",
};
bool quantize = false;
for (const auto & s : k_names) {
if (std::regex_match(name, std::regex(s))) {
quantize = true;
break;
}
}
// quantize only 2D tensors
quantize &= (n_dims == 2);
if (quantize) {
if (ftype != 0 && ftype != 1) {
fprintf(stderr, "%s: unsupported ftype %d for integer quantization\n", __func__, ftype);
return false;
}
if (ftype == 1) {
data_f16.resize(nelements);
finp.read(reinterpret_cast<char *>(data_f16.data()), nelements * sizeof(ggml_fp16_t));
data_f32.resize(nelements);
for (int i = 0; i < nelements; ++i) {
data_f32[i] = ggml_fp16_to_fp32(data_f16[i]);
}
} else {
data_f32.resize(nelements);
finp.read(reinterpret_cast<char *>(data_f32.data()), nelements * sizeof(float));
}
ftype = itype;
} else {
const int bpe = (ftype == 0) ? sizeof(float) : sizeof(uint16_t);
data_u8.resize(nelements*bpe);
finp.read(reinterpret_cast<char *>(data_u8.data()), nelements * bpe);
}
fout.write(reinterpret_cast<char *>(&n_dims), sizeof(n_dims));
fout.write(reinterpret_cast<char *>(&length), sizeof(length));
fout.write(reinterpret_cast<char *>(&ftype), sizeof(ftype));
for (int i = 0; i < n_dims; ++i) {
fout.write(reinterpret_cast<char *>(&ne[i]), sizeof(ne[i]));
}
fout.write(&name[0], length);
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
{
// ensure tensor data is aligned
uint64_t offset = fout.tellp();
offset = (offset + 31) & -32;
fout.seekp(offset);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
if (quantize) {
printf("quantizing .. ");
work.resize(nelements); // for quantization
size_t cur_size = 0;
std::vector<int64_t> hist_cur(1 << 4, 0);
switch (type) {
case GGML_TYPE_Q4_0:
{
ggml : introduce structs for the q4 data blocks (#356) * Introduce structs for the q4 data blocks * ggml : rename quant struct variables + fix ARM_NEON --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 15:56:03 +00:00
cur_size = ggml_quantize_q4_0(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
} break;
case GGML_TYPE_Q4_1:
{
ggml : introduce structs for the q4 data blocks (#356) * Introduce structs for the q4 data blocks * ggml : rename quant struct variables + fix ARM_NEON --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 15:56:03 +00:00
cur_size = ggml_quantize_q4_1(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
} break;
default:
{
fprintf(stderr, "%s: unsupported quantization type %d\n", __func__, type);
return false;
}
}
fout.write(reinterpret_cast<char *>(work.data()), cur_size);
total_size_new += cur_size;
printf("size = %8.2f MB -> %8.2f MB | hist: ", nelements * sizeof(float)/1024.0/1024.0, cur_size/1024.0/1024.0);
for (int i = 0; i < (int) hist_cur.size(); ++i) {
hist_all[i] += hist_cur[i];
}
for (int i = 0; i < (int) hist_cur.size(); ++i) {
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
printf("%5.3f ", hist_cur[i] / float(nelements));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
printf("\n");
} else {
printf("size = %8.3f MB\n", data_u8.size()/1024.0/1024.0);
fout.write(reinterpret_cast<char *>(data_u8.data()), data_u8.size());
total_size_new += data_u8.size();
}
total_size_org += nelements * sizeof(float);
}
printf("%s: model size = %8.2f MB\n", __func__, total_size_org/1024.0/1024.0);
printf("%s: quant size = %8.2f MB\n", __func__, total_size_new/1024.0/1024.0);
{
int64_t sum_all = 0;
for (int i = 0; i < (int) hist_all.size(); ++i) {
sum_all += hist_all[i];
}
printf("%s: hist: ", __func__);
for (int i = 0; i < (int) hist_all.size(); ++i) {
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
printf("%5.3f ", hist_all[i] / float(sum_all));
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
printf("\n");
}
}
finp.close();
fout.close();
return true;
}
//
// interface implementation
//
struct llama_context * llama_init_from_file(
const char * path_model,
struct llama_context_params params) {
ggml_time_init();
llama_context * ctx = new llama_context;
Init llama_context_params properly from CLI (#370)
2023-03-22 05:45:00 +00:00
if (params.seed <= 0) {
params.seed = time(NULL);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
ctx->rng = std::mt19937(params.seed);
ctx->logits_all = params.logits_all;
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
ggml_type memory_type = params.f16_kv ? GGML_TYPE_F16 : GGML_TYPE_F32;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
if (!llama_model_load(path_model, *ctx, params.n_ctx, params.n_parts, memory_type,
Add support for file load progress reporting callbacks (#434) * File load progress reporting * Move llama_progress_handler into llama_context_params * Renames * Use seekg to find file size instead * More correct load progress * Call progress callback more frequently * Fix typo
2023-03-25 05:26:28 +00:00
params.vocab_only, params.progress_callback,
params.progress_callback_user_data)) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
fprintf(stderr, "%s: failed to load model\n", __func__);
Properly free llama_context on failure
2023-03-24 15:21:01 +00:00
llama_free(ctx);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return nullptr;
}
Properly free llama_context on failure
2023-03-24 15:21:01 +00:00
Support calling mlock() on loaded model data on Linux and macOS (#453) * Support calling mlock() on loaded model data on Linux and macOS This is enabled by a new --mlock command line option. Using mlock() disables swapping and memory compression for the model data. Doing so can be useful on systems where the model takes up a large fraction of system RAM. In my experience, macOS is quite eager to start compressing llama.cpp's memory, which then makes it halt for a few seconds while it decompresses, even with a model that uses "only" 25GB out of 32GB. Of course, this comes at the cost of forcing the system to swap or compress other processes' memory instead, so it needs to be used with care and shouldn't be enabled by default. In theory it should be possible to support this on Windows as well using VirtualLock(), but I'm not much of a Windows user. * Update llama.cpp --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:19:05 +00:00
if (params.use_mlock) {
char *err;
Ensure --mlock works properly with mmap() support
2023-03-30 08:53:36 +00:00
if (!ggml_mlock(ctx->model.ctx,
ctx->model.mm_addr,
ctx->model.mm_length,
&err)) {
Support calling mlock() on loaded model data on Linux and macOS (#453) * Support calling mlock() on loaded model data on Linux and macOS This is enabled by a new --mlock command line option. Using mlock() disables swapping and memory compression for the model data. Doing so can be useful on systems where the model takes up a large fraction of system RAM. In my experience, macOS is quite eager to start compressing llama.cpp's memory, which then makes it halt for a few seconds while it decompresses, even with a model that uses "only" 25GB out of 32GB. Of course, this comes at the cost of forcing the system to swap or compress other processes' memory instead, so it needs to be used with care and shouldn't be enabled by default. In theory it should be possible to support this on Windows as well using VirtualLock(), but I'm not much of a Windows user. * Update llama.cpp --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:19:05 +00:00
fprintf(stderr, "%s\n", err);
free(err);
Properly free llama_context on failure
2023-03-24 15:21:01 +00:00
llama_free(ctx);
Support calling mlock() on loaded model data on Linux and macOS (#453) * Support calling mlock() on loaded model data on Linux and macOS This is enabled by a new --mlock command line option. Using mlock() disables swapping and memory compression for the model data. Doing so can be useful on systems where the model takes up a large fraction of system RAM. In my experience, macOS is quite eager to start compressing llama.cpp's memory, which then makes it halt for a few seconds while it decompresses, even with a model that uses "only" 25GB out of 32GB. Of course, this comes at the cost of forcing the system to swap or compress other processes' memory instead, so it needs to be used with care and shouldn't be enabled by default. In theory it should be possible to support this on Windows as well using VirtualLock(), but I'm not much of a Windows user. * Update llama.cpp --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:19:05 +00:00
return nullptr;
}
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
// reserve memory for context buffers
{
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
if (!kv_cache_init(ctx->model.hparams, ctx->model.kv_self, memory_type, ctx->model.hparams.n_ctx)) {
fprintf(stderr, "%s: kv_cache_init() failed for self-attention cache\n", __func__);
llama_free(ctx);
return nullptr;
}
{
const size_t memory_size = ggml_nbytes(ctx->model.kv_self.k) + ggml_nbytes(ctx->model.kv_self.v);
fprintf(stderr, "%s: kv self size = %7.2f MB\n", __func__, memory_size / 1024.0 / 1024.0);
}
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
const auto & hparams = ctx->model.hparams;
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
// resized during inference
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
if (params.logits_all) {
ctx->logits.reserve(hparams.n_ctx*hparams.n_vocab);
} else {
ctx->logits.reserve(hparams.n_ctx);
}
if (params.embedding){
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
ctx->embedding.resize(hparams.n_embd);
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
}
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
ctx->buf_compute.resize(MEM_REQ_EVAL.at(ctx->model.type));
ctx->buf_scratch[0].resize(MEM_REQ_SCRATCH0.at(ctx->model.type));
ctx->buf_scratch[1].resize(MEM_REQ_SCRATCH1.at(ctx->model.type));
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return ctx;
}
void llama_free(struct llama_context * ctx) {
Reduce memory usage and allocate enough memory for largest context (#473) * Reduce memory usage and allocate enough memory for large contexts * Simpler scratch buffer usage * Reenable BLAS for quantized mul_mat * Fix number of layers in 30B and 65B * Fix KV cache size for F32
2023-03-24 21:17:37 +00:00
kv_cache_free(ctx->model.kv_self);
Properly free llama_context on failure
2023-03-24 15:21:01 +00:00
if (ctx->model.ctx) {
ggml_free(ctx->model.ctx);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
Unmap the file in llama_free
2023-03-29 06:31:26 +00:00
if (ctx->model.mm_addr) {
munmap_file(ctx->model.mm_addr, ctx->model.mm_length);
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
delete ctx;
}
int llama_model_quantize(
const char * fname_inp,
const char * fname_out,
ggml : introduce structs for the q4 data blocks (#356) * Introduce structs for the q4 data blocks * ggml : rename quant struct variables + fix ARM_NEON --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 15:56:03 +00:00
int itype) {
if (!llama_model_quantize_internal(fname_inp, fname_out, itype)) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
fprintf(stderr, "%s: failed to quantize\n", __func__);
return 1;
}
return 0;
}
int llama_eval(
struct llama_context * ctx,
const llama_token * tokens,
int n_tokens,
int n_past,
int n_threads) {
if (!llama_eval_internal(*ctx, tokens, n_tokens, n_past, n_threads)) {
fprintf(stderr, "%s: failed to eval\n", __func__);
return 1;
}
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
// get a more accurate load time, upon first eval
if (!ctx->has_evaluated_once) {
ctx->t_load_us = ggml_time_us() - ctx->t_start_us;
ctx->has_evaluated_once = true;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
return 0;
}
int llama_tokenize(
struct llama_context * ctx,
const char * text,
llama_token * tokens,
int n_max_tokens,
bool add_bos) {
auto res = llama_tokenize(ctx->vocab, text, add_bos);
if (n_max_tokens < (int) res.size()) {
fprintf(stderr, "%s: too many tokens\n", __func__);
return -((int) res.size());
}
for (size_t i = 0; i < res.size(); i++) {
tokens[i] = res[i];
}
return res.size();
}
int llama_n_vocab(struct llama_context * ctx) {
return ctx->vocab.id_to_token.size();
}
int llama_n_ctx(struct llama_context * ctx) {
return ctx->model.hparams.n_ctx;
}
Cleanup STL headers + fix embedding examples + minor stuff
2023-03-25 18:51:14 +00:00
int llama_n_embd(struct llama_context * ctx) {
return ctx->model.hparams.n_embd;
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
float * llama_get_logits(struct llama_context * ctx) {
return ctx->logits.data();
}
Add embedding mode with arg flag. Currently working (#282) * working but ugly * add arg flag, not working on embedding mode * typo * Working! Thanks to @nullhook * make params argument instead of hardcoded boolean. remove useless time check * start doing the instructions but not finished. This probably doesnt compile * Embeddings extraction support --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-24 15:05:13 +00:00
float * llama_get_embeddings(struct llama_context * ctx) {
return ctx->embedding.data();
}
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
const char * llama_token_to_str(struct llama_context * ctx, llama_token token) {
if (token >= llama_n_vocab(ctx)) {
return nullptr;
}
return ctx->vocab.id_to_token[token].tok.c_str();
}
llama_token llama_token_bos() {
return 1;
}
llama_token llama_token_eos() {
return 2;
}
llama_token llama_sample_top_p_top_k(
llama_context * ctx,
const llama_token * last_n_tokens_data,
int last_n_tokens_size,
int top_k,
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
float top_p,
float temp,
float repeat_penalty) {
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
const int64_t t_start_sample_us = ggml_time_us();
llama_token result = 0;
// TODO: avoid this ...
const auto last_n_tokens = std::vector<llama_token>(last_n_tokens_data, last_n_tokens_data + last_n_tokens_size);
result = llama_sample_top_p_top_k(
*ctx,
last_n_tokens,
top_k,
top_p,
temp,
repeat_penalty);
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
ctx->n_sample++;
return result;
}
void llama_print_timings(struct llama_context * ctx) {
const int64_t t_end_us = ggml_time_us();
Make loading weights 10-100x faster This is a breaking change that's going to give you three benefits: 1. Your inference commands should load 100x faster 2. You may be able to safely load models 2x larger 3. You can run many concurrent inference processes This was accomplished by changing the file format so we can mmap() weights directly into memory without having to read() or copy them thereby ensuring the kernel can make its file cache pages directly accessible to our inference processes; and secondly, that the file cache pages are much less likely to get evicted (which would force loads to hit disk) because they're no longer competing with memory pages that were needlessly created by gigabytes of standard i/o. The new file format supports single-file models like LLaMA 7b, and it also supports multi-file models like LLaMA 13B. Our Python tool now merges the foo.1, foo.2, etc. files back into a single file so that the C++ code which maps it doesn't need to reshape data every time. That's made llama.cpp so much simpler. Much of its load code has now been deleted. Furthermore, this change ensures that tensors are aligned properly on a 32-byte boundary. That opens the door to seeing if we can get additional performance gains on some microprocessors, by using ops that require memory alignment. Lastly note that both POSIX and the Windows platform are supported Fixes #91
2023-03-29 20:51:37 +00:00
const int32_t n_sample = Max(1, ctx->n_sample);
const int32_t n_eval = Max(1, ctx->n_eval);
const int32_t n_p_eval = Max(1, ctx->n_p_eval);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
fprintf(stderr, "\n");
all : be more strict about converting float to double (#458) * Be more strict about converting float to double * Test equivalence of round, SILU implementations Test module is commented out in CMakeLists.txt because the tests may take a long time, depending on how much the compiler optimizes. * Fix softmax in perplexity.cpp * all : prefer float over double where appropriate * perplexity : add <cmath> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-03-28 16:48:20 +00:00
fprintf(stderr, "%s: load time = %8.2f ms\n", __func__, ctx->t_load_us / 1000.0);
fprintf(stderr, "%s: sample time = %8.2f ms / %5d runs (%8.2f ms per run)\n", __func__, 1e-3 * ctx->t_sample_us, n_sample, 1e-3 * ctx->t_sample_us / n_sample);
fprintf(stderr, "%s: prompt eval time = %8.2f ms / %5d tokens (%8.2f ms per token)\n", __func__, 1e-3 * ctx->t_p_eval_us, n_p_eval, 1e-3 * ctx->t_p_eval_us / n_p_eval);
fprintf(stderr, "%s: eval time = %8.2f ms / %5d runs (%8.2f ms per run)\n", __func__, 1e-3 * ctx->t_eval_us, n_eval, 1e-3 * ctx->t_eval_us / n_eval);
fprintf(stderr, "%s: total time = %8.2f ms\n", __func__, (t_end_us - ctx->t_start_us)/1000.0);
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
void llama_reset_timings(struct llama_context * ctx) {
ctx->t_start_us = ggml_time_us();
ctx->t_sample_us = ctx->n_sample = 0;
ctx->t_eval_us = ctx->n_eval = 0;
Add timings for the prompt evaluation (#478)
2023-03-25 14:34:23 +00:00
ctx->t_p_eval_us = ctx->n_p_eval = 0;
Introduce C-style API (#370) * Major refactoring - introduce C-style API * Clean up * Add <cassert> * Add <iterator> * Add <algorithm> .... * Fix timing reporting and accumulation * Measure eval time only for single-token calls * Change llama_tokenize return meaning
2023-03-22 05:32:36 +00:00
}
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();
}
Reference in a new issue Copy permalink