2023-04-30 17:51:57 +02:00
# include "llama-util.h"
2023-03-27 20:00:32 +02:00
# include "llama.h"
# include "ggml.h"
2023-05-14 17:46:19 +02:00
# ifdef GGML_USE_CUBLAS
# include "ggml-cuda.h"
# endif
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
# include <array>
2023-04-30 17:51:57 +02:00
# include <ctime>
2023-03-27 20:00:32 +02:00
# include <cinttypes>
# include <fstream>
# include <random>
# include <map>
# include <unordered_map>
# include <queue>
# include <cassert>
# include <cstring>
2023-04-10 21:59:13 +02:00
# include <climits>
# include <memory>
# include <algorithm>
# include <initializer_list>
2023-04-30 17:51:57 +02:00
# include <thread>
# include <atomic>
# include <mutex>
# include <sstream>
# include <numeric>
2023-03-27 20:00:32 +02:00
# define LLAMA_USE_SCRATCH
# define LLAMA_MAX_SCRATCH_BUFFERS 16
// available llama models
enum e_model {
MODEL_UNKNOWN ,
MODEL_7B ,
MODEL_13B ,
MODEL_30B ,
MODEL_65B ,
} ;
2023-05-23 13:04:39 +02:00
2023-03-27 20:00:32 +02:00
static const size_t MB = 1024 * 1024 ;
// computed for n_ctx == 2048
// TODO: dynamically determine these sizes
// needs modifications in ggml
2023-04-30 17:51:57 +02:00
static const std : : map < e_model , size_t > & MEM_REQ_SCRATCH0 ( )
{
2023-05-14 17:46:19 +02:00
static std : : map < e_model , size_t > k_sizes = {
2023-04-30 17:51:57 +02:00
{ MODEL_7B , 512ull * MB } ,
{ MODEL_13B , 512ull * MB } ,
{ MODEL_30B , 512ull * MB } ,
{ MODEL_65B , 1024ull * MB } ,
} ;
2023-05-14 17:46:19 +02:00
return k_sizes ;
2023-04-30 17:51:57 +02:00
}
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
static const std : : map < e_model , size_t > & MEM_REQ_SCRATCH1 ( )
{
2023-05-14 17:46:19 +02:00
static std : : map < e_model , size_t > k_sizes = {
2023-04-30 17:51:57 +02:00
{ MODEL_7B , 512ull * MB } ,
{ MODEL_13B , 512ull * MB } ,
{ MODEL_30B , 512ull * MB } ,
{ MODEL_65B , 1024ull * MB } ,
} ;
2023-05-14 17:46:19 +02:00
return k_sizes ;
2023-04-30 17:51:57 +02:00
}
2023-03-27 20:00:32 +02:00
// 2*n_embd*n_ctx*n_layer*sizeof(float16)
2023-04-30 17:51:57 +02:00
static const std : : map < e_model , size_t > & MEM_REQ_KV_SELF ( )
{
2023-05-14 17:46:19 +02:00
static std : : map < e_model , size_t > k_sizes = {
2023-04-30 17:51:57 +02:00
{ MODEL_7B , 1026ull * MB } ,
{ MODEL_13B , 1608ull * MB } ,
{ MODEL_30B , 3124ull * MB } ,
{ MODEL_65B , 5120ull * MB } ,
} ;
2023-05-14 17:46:19 +02:00
return k_sizes ;
2023-04-30 17:51:57 +02:00
}
2023-03-27 20:00:32 +02:00
// this is mostly needed for temporary mul_mat buffers to dequantize the data
// not actually needed if BLAS is disabled
2023-04-30 17:51:57 +02:00
static const std : : map < e_model , size_t > & MEM_REQ_EVAL ( )
{
2023-05-14 17:46:19 +02:00
static std : : map < e_model , size_t > k_sizes = {
2023-04-30 17:51:57 +02:00
{ MODEL_7B , 768ull * MB } ,
{ MODEL_13B , 1024ull * MB } ,
{ MODEL_30B , 1280ull * MB } ,
{ MODEL_65B , 1536ull * MB } ,
} ;
2023-05-14 17:46:19 +02:00
return k_sizes ;
2023-04-30 17:51:57 +02:00
}
2023-03-27 20:00:32 +02:00
// default hparams (LLaMA 7B)
struct llama_hparams {
2023-04-10 21:59:13 +02:00
uint32_t n_vocab = 32000 ;
uint32_t n_ctx = 512 ; // this is provided as user input?
uint32_t n_embd = 4096 ;
uint32_t n_mult = 256 ;
uint32_t n_head = 32 ;
uint32_t n_layer = 32 ;
uint32_t n_rot = 64 ;
2023-04-30 17:51:57 +02:00
enum llama_ftype ftype = LLAMA_FTYPE_MOSTLY_F16 ;
2023-04-10 21:59:13 +02:00
bool operator ! = ( const llama_hparams & other ) const {
2023-05-23 13:04:39 +02:00
return static_cast < bool > ( memcmp ( this , & other , sizeof ( llama_hparams ) ) ) ;
2023-04-10 21:59:13 +02:00
}
2023-03-27 20:00:32 +02:00
} ;
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 ;
} ;
struct llama_kv_cache {
struct ggml_tensor * k ;
struct ggml_tensor * v ;
2023-04-10 21:59:13 +02:00
struct ggml_context * ctx = NULL ;
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
llama_ctx_buffer buf ;
2023-03-27 20:00:32 +02:00
int n ; // number of tokens currently in the cache
2023-04-10 21:59:13 +02:00
~ llama_kv_cache ( ) {
if ( ctx ) {
ggml_free ( ctx ) ;
}
}
2023-03-27 20:00:32 +02:00
} ;
struct llama_model {
e_model type = MODEL_UNKNOWN ;
llama_hparams hparams ;
struct ggml_tensor * tok_embeddings ;
struct ggml_tensor * norm ;
struct ggml_tensor * output ;
std : : vector < llama_layer > layers ;
// context
2023-04-10 21:59:13 +02:00
struct ggml_context * ctx = NULL ;
2023-03-27 20:00:32 +02:00
// key + value cache for the self attention
// TODO: move to llama_state
struct llama_kv_cache kv_self ;
// the model memory buffer
2023-04-30 17:51:57 +02:00
llama_ctx_buffer buf ;
2023-04-10 21:59:13 +02:00
// model memory mapped file
std : : unique_ptr < llama_mmap > mapping ;
// objects representing data potentially being locked in memory
llama_mlock mlock_buf ;
llama_mlock mlock_mmap ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
// for quantize-stats only
std : : vector < std : : pair < std : : string , struct ggml_tensor * > > tensors_by_name ;
~ llama_model ( ) {
if ( ctx ) {
ggml_free ( ctx ) ;
}
}
2023-03-27 20:00:32 +02:00
} ;
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 ;
2023-04-10 21:59:13 +02:00
bool has_evaluated_once = false ;
2023-03-27 20:00:32 +02:00
int64_t t_sample_us = 0 ;
int64_t t_eval_us = 0 ;
int64_t t_p_eval_us = 0 ;
int32_t n_sample = 0 ; // number of tokens sampled
int32_t n_eval = 0 ; // number of eval calls
int32_t n_p_eval = 0 ; // number of tokens in eval calls for the prompt (with batch size > 1)
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 ;
// input embedding (1-dimensional array: [n_embd])
std : : vector < float > embedding ;
// memory buffers used to evaluate the model
// TODO: move in llama_state
2023-04-30 17:51:57 +02:00
llama_ctx_buffer buf_compute ;
llama_ctx_buffer buf_scratch [ LLAMA_MAX_SCRATCH_BUFFERS ] ;
2023-03-27 20:00:32 +02:00
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 ] ;
2023-04-10 21:59:13 +02:00
last_size = ggml_set_scratch ( ctx , { 0 , buf . size , buf . addr , } ) ;
2023-03-27 20:00:32 +02:00
}
if ( buf_last > = 0 ) {
buf_max_size [ buf_last ] = std : : max ( buf_max_size [ buf_last ] , last_size ) ;
}
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
}
} ;
2023-04-10 21:59:13 +02:00
template < typename T >
static T checked_mul ( T a , T b ) {
T ret = a * b ;
if ( a ! = 0 & & ret / a ! = b ) {
throw format ( " overflow multiplying %llu * %llu " ,
( unsigned long long ) a , ( unsigned long long ) b ) ;
}
return ret ;
}
static std : : string llama_format_tensor_shape ( const std : : vector < uint32_t > & ne ) {
2023-04-30 17:51:57 +02:00
char buf [ 256 ] ;
snprintf ( buf , sizeof ( buf ) , " %5u " , ne . at ( 0 ) ) ;
2023-04-10 21:59:13 +02:00
for ( size_t i = 1 ; i < ne . size ( ) ; i + + ) {
2023-04-30 17:51:57 +02:00
snprintf ( buf + strlen ( buf ) , sizeof ( buf ) - strlen ( buf ) , " x %5u " , ne . at ( i ) ) ;
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
return buf ;
2023-04-10 21:59:13 +02:00
}
static size_t llama_calc_tensor_size ( const std : : vector < uint32_t > & ne , enum ggml_type type ) {
size_t size = ggml_type_size ( type ) ;
for ( uint32_t dim : ne ) {
size = checked_mul < size_t > ( size , dim ) ;
}
return size / ggml_blck_size ( type ) ;
}
struct llama_load_tensor_shard {
std : : vector < uint32_t > ne ;
size_t size ;
enum ggml_type type ;
size_t file_idx ;
size_t file_off ;
void calc_size ( ) {
size = llama_calc_tensor_size ( ne , type ) ;
}
} ;
enum llama_split_type {
SPLIT_NONE ,
SPLIT_BY_COLUMNS ,
SPLIT_BY_ROWS
} ;
struct llama_load_tensor {
std : : vector < llama_load_tensor_shard > shards ;
std : : string name ;
enum ggml_type type = GGML_TYPE_F32 ;
llama_split_type split_type = SPLIT_NONE ;
std : : vector < uint32_t > ne ;
size_t size ;
struct ggml_tensor * ggml_tensor = NULL ;
uint8_t * data ;
llama_load_tensor ( const std : : string & name ) : name ( name ) { }
void calc_all ( ) {
calc_type ( ) ;
calc_split_type ( ) ;
calc_ne ( ) ;
calc_size ( ) ;
}
void calc_type ( ) {
const auto & first_shard = shards . at ( 0 ) ;
for ( const auto & shard : shards ) {
if ( shard . type ! = first_shard . type ) {
throw format ( " inconsistent tensor shard type in '%s' " , name . c_str ( ) ) ;
}
}
type = first_shard . type ;
}
void calc_split_type ( ) {
if ( shards . at ( 0 ) . ne . size ( ) = = 1 | | // 1D tensors are just duplicated in every file
shards . size ( ) = = 1 ) { // only one file?
split_type = SPLIT_NONE ;
} else if ( name . find ( " tok_embeddings. " ) = = 0 | |
name . find ( " .attention.wo.weight " ) ! = std : : string : : npos | |
name . find ( " .feed_forward.w2.weight " ) ! = std : : string : : npos ) {
split_type = SPLIT_BY_COLUMNS ;
} else {
split_type = SPLIT_BY_ROWS ;
}
}
void calc_ne ( ) {
const auto & first_shard = shards . at ( 0 ) ;
for ( const auto & shard : shards ) {
if ( shard . ne ! = first_shard . ne ) {
throw format ( " inconsistent tensor shard shape in '%s': first was %s, other was %s " ,
name . c_str ( ) , llama_format_tensor_shape ( first_shard . ne ) . c_str ( ) , llama_format_tensor_shape ( shard . ne ) . c_str ( ) ) ;
}
}
ne = first_shard . ne ;
LLAMA_ASSERT ( shards . size ( ) < = UINT32_MAX ) ;
uint32_t n_shards = ( uint32_t ) shards . size ( ) ;
switch ( split_type ) {
case SPLIT_NONE :
ne = first_shard . ne ;
break ;
case SPLIT_BY_COLUMNS :
ne = { checked_mul < uint32_t > ( first_shard . ne [ 0 ] , n_shards ) ,
first_shard . ne [ 1 ] } ;
break ;
case SPLIT_BY_ROWS :
ne = { first_shard . ne [ 0 ] ,
checked_mul < uint32_t > ( first_shard . ne [ 1 ] , n_shards ) } ;
break ;
}
}
void calc_size ( ) {
size = llama_calc_tensor_size ( ne , type ) ;
}
} ;
struct llama_load_tensors_map {
// tensors is kept in a separate vector to preserve file order
std : : vector < llama_load_tensor > tensors ;
std : : unordered_map < std : : string , size_t > name_to_idx ;
} ;
enum llama_file_version {
LLAMA_FILE_VERSION_GGML ,
LLAMA_FILE_VERSION_GGMF_V1 , // added version field and scores in vocab
LLAMA_FILE_VERSION_GGJT_V1 , // added padding
2023-05-14 17:46:19 +02:00
LLAMA_FILE_VERSION_GGJT_V2 , // changed quantization format
2023-05-23 13:04:39 +02:00
LLAMA_FILE_VERSION_GGJT_V3 , // changed Q4 and Q8 quantization format
2023-04-10 21:59:13 +02:00
} ;
struct llama_file_loader {
llama_file file ;
llama_file_version file_version ;
llama_hparams hparams ;
llama_vocab vocab ;
llama_file_loader ( const char * fname , size_t file_idx , llama_load_tensors_map & tensors_map )
: file ( fname , " rb " ) {
fprintf ( stderr , " llama.cpp: loading model from %s \n " , fname ) ;
read_magic ( ) ;
read_hparams ( ) ;
read_vocab ( ) ;
read_tensor_metadata ( file_idx , tensors_map ) ;
}
void read_magic ( ) {
uint32_t magic = file . read_u32 ( ) ;
2023-05-23 13:04:39 +02:00
if ( magic = = LLAMA_FILE_MAGIC_GGML ) {
file_version = LLAMA_FILE_VERSION_GGML ;
return ;
2023-04-10 21:59:13 +02:00
}
2023-05-23 13:04:39 +02:00
uint32_t version = file . read_u32 ( ) ;
switch ( magic ) {
case LLAMA_FILE_MAGIC_GGMF :
switch ( version ) {
case 1 : file_version = LLAMA_FILE_VERSION_GGMF_V1 ; return ;
}
break ;
case LLAMA_FILE_MAGIC_GGJT :
switch ( version ) {
case 1 : file_version = LLAMA_FILE_VERSION_GGJT_V1 ; return ;
case 2 : file_version = LLAMA_FILE_VERSION_GGJT_V2 ; return ;
case 3 : file_version = LLAMA_FILE_VERSION_GGJT_V3 ; return ;
}
2023-04-10 21:59:13 +02:00
}
2023-05-23 13:04:39 +02:00
throw format ( " unknown (magic, version) combination : % 08 x , % 08 x ; is this really a GGML file ? " ,
magic , version ) ;
2023-04-10 21:59:13 +02:00
}
void read_hparams ( ) {
hparams . n_vocab = file . read_u32 ( ) ;
hparams . n_embd = file . read_u32 ( ) ;
hparams . n_mult = file . read_u32 ( ) ;
hparams . n_head = file . read_u32 ( ) ;
hparams . n_layer = file . read_u32 ( ) ;
hparams . n_rot = file . read_u32 ( ) ;
2023-04-30 17:51:57 +02:00
hparams . ftype = ( enum llama_ftype ) file . read_u32 ( ) ;
2023-04-10 21:59:13 +02:00
}
void read_vocab ( ) {
vocab . id_to_token . resize ( hparams . n_vocab ) ;
for ( uint32_t i = 0 ; i < hparams . n_vocab ; i + + ) {
uint32_t len = file . read_u32 ( ) ;
std : : string word = file . read_string ( len ) ;
float score = 0.0f ;
if ( file_version > = LLAMA_FILE_VERSION_GGMF_V1 ) {
file . read_raw ( & score , sizeof ( score ) ) ;
}
vocab . token_to_id [ word ] = i ;
auto & tok_score = vocab . id_to_token [ i ] ;
tok_score . tok = std : : move ( word ) ;
tok_score . score = score ;
}
}
void read_tensor_metadata ( size_t file_idx , llama_load_tensors_map & tensors_map ) {
while ( file . tell ( ) < file . size ) {
llama_load_tensor_shard shard ;
uint32_t n_dims = file . read_u32 ( ) ;
uint32_t name_len = file . read_u32 ( ) ;
2023-04-30 17:51:57 +02:00
shard . type = ( enum ggml_type ) file . read_u32 ( ) ;
2023-04-10 21:59:13 +02:00
shard . ne . resize ( n_dims ) ;
file . read_raw ( shard . ne . data ( ) , sizeof ( shard . ne [ 0 ] ) * n_dims ) ;
std : : string name = file . read_string ( name_len ) ;
if ( n_dims < 1 | | n_dims > 2 ) {
throw format ( " llama.cpp: tensor '%s' should not be %u-dimensional " , name . c_str ( ) , n_dims ) ;
}
2023-04-30 17:51:57 +02:00
switch ( shard . type ) {
case GGML_TYPE_F32 :
case GGML_TYPE_F16 :
case GGML_TYPE_Q4_0 :
case GGML_TYPE_Q4_1 :
case GGML_TYPE_Q5_0 :
case GGML_TYPE_Q5_1 :
case GGML_TYPE_Q8_0 :
break ;
2023-04-10 21:59:13 +02:00
default : {
2023-04-30 17:51:57 +02:00
throw format ( " unrecognized tensor type %u \n " , shard . type ) ;
2023-04-10 21:59:13 +02:00
}
}
if ( file_version > = LLAMA_FILE_VERSION_GGJT_V1 ) {
// skip to the next multiple of 32 bytes
2023-05-23 13:04:39 +02:00
file . seek ( - static_cast < ptrdiff_t > ( file . tell ( ) ) & 31 , SEEK_CUR ) ;
2023-04-10 21:59:13 +02:00
}
shard . file_idx = file_idx ;
shard . file_off = file . tell ( ) ;
shard . calc_size ( ) ;
file . seek ( shard . size , SEEK_CUR ) ;
auto it = tensors_map . name_to_idx . find ( name ) ;
size_t idx ;
if ( it ! = tensors_map . name_to_idx . end ( ) ) {
idx = it - > second ;
} else {
tensors_map . tensors . emplace_back ( name ) ;
idx = tensors_map . tensors . size ( ) - 1 ;
tensors_map . name_to_idx . emplace ( name , idx ) ;
}
tensors_map . tensors . at ( idx ) . shards . push_back ( shard ) ;
}
}
} ;
struct llama_file_saver {
llama_file file ;
llama_file_loader * any_file_loader ;
2023-04-30 17:51:57 +02:00
llama_file_saver ( const char * fname , llama_file_loader * any_file_loader , enum llama_ftype new_ftype )
2023-04-10 21:59:13 +02:00
: file ( fname , " wb " ) , any_file_loader ( any_file_loader ) {
fprintf ( stderr , " llama.cpp: saving model to %s \n " , fname ) ;
write_magic ( ) ;
2023-04-30 17:51:57 +02:00
write_hparams ( new_ftype ) ;
2023-04-10 21:59:13 +02:00
write_vocab ( ) ;
}
void write_magic ( ) {
2023-05-14 17:46:19 +02:00
file . write_u32 ( LLAMA_FILE_MAGIC ) ; // magic
file . write_u32 ( LLAMA_FILE_VERSION ) ; // version
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
void write_hparams ( enum llama_ftype new_ftype ) {
2023-04-10 21:59:13 +02:00
const llama_hparams & hparams = any_file_loader - > hparams ;
file . write_u32 ( hparams . n_vocab ) ;
file . write_u32 ( hparams . n_embd ) ;
file . write_u32 ( hparams . n_mult ) ;
file . write_u32 ( hparams . n_head ) ;
file . write_u32 ( hparams . n_layer ) ;
file . write_u32 ( hparams . n_rot ) ;
2023-04-30 17:51:57 +02:00
file . write_u32 ( new_ftype ) ;
2023-04-10 21:59:13 +02:00
}
void write_vocab ( ) {
if ( any_file_loader - > file_version = = LLAMA_FILE_VERSION_GGML ) {
fprintf ( stderr , " llama.cpp: WARNING: input is an old file that doesn't have scores; will add dummy scores \n " ) ;
}
uint32_t n_vocab = any_file_loader - > hparams . n_vocab ;
for ( uint32_t i = 0 ; i < n_vocab ; i + + ) {
const auto & token_score = any_file_loader - > vocab . id_to_token . at ( i ) ;
file . write_u32 ( ( uint32_t ) token_score . tok . size ( ) ) ;
file . write_raw ( token_score . tok . data ( ) , token_score . tok . size ( ) ) ;
file . write_raw ( & token_score . score , sizeof ( token_score . score ) ) ;
}
}
void write_tensor ( llama_load_tensor & tensor , enum ggml_type new_type , const void * new_data , size_t new_size ) {
switch ( new_type ) {
2023-04-30 17:51:57 +02:00
case GGML_TYPE_F32 :
case GGML_TYPE_F16 :
case GGML_TYPE_Q4_0 :
case GGML_TYPE_Q4_1 :
case GGML_TYPE_Q5_0 :
case GGML_TYPE_Q5_1 :
case GGML_TYPE_Q8_0 :
break ;
2023-04-10 21:59:13 +02:00
default : LLAMA_ASSERT ( false ) ;
}
file . write_u32 ( ( uint32_t ) tensor . ne . size ( ) ) ;
file . write_u32 ( ( uint32_t ) tensor . name . size ( ) ) ;
2023-04-30 17:51:57 +02:00
file . write_u32 ( new_type ) ;
2023-04-10 21:59:13 +02:00
file . write_raw ( tensor . ne . data ( ) , sizeof ( tensor . ne [ 0 ] ) * tensor . ne . size ( ) ) ;
file . write_raw ( tensor . name . data ( ) , tensor . name . size ( ) ) ;
2023-05-23 13:04:39 +02:00
file . seek ( - static_cast < ptrdiff_t > ( file . tell ( ) ) & 31 , SEEK_CUR ) ;
2023-04-10 21:59:13 +02:00
LLAMA_ASSERT ( new_size = = llama_calc_tensor_size ( tensor . ne , new_type ) ) ;
file . write_raw ( new_data , new_size ) ;
}
} ;
struct llama_model_loader {
std : : vector < std : : unique_ptr < llama_file_loader > > file_loaders ;
llama_load_tensors_map tensors_map ;
bool use_mmap ;
size_t num_ggml_tensors_created = 0 ;
struct ggml_context * ggml_ctx = NULL ;
std : : unique_ptr < llama_mmap > mapping ;
llama_model_loader ( const std : : string & fname_base , bool use_mmap , bool vocab_only ) {
2023-05-14 17:46:19 +02:00
auto * first_file = new llama_file_loader ( fname_base . c_str ( ) , 0 , tensors_map ) ;
2023-04-10 21:59:13 +02:00
file_loaders . emplace_back ( first_file ) ;
uint32_t n_parts = vocab_only ? 1 : guess_n_parts ( ) ;
for ( uint32_t i = 1 ; i < n_parts ; i + + ) {
std : : string fname = fname_base + " . " + std : : to_string ( i ) ;
2023-05-14 17:46:19 +02:00
auto * ith_file = new llama_file_loader ( fname . c_str ( ) , i , tensors_map ) ;
2023-04-10 21:59:13 +02:00
file_loaders . emplace_back ( ith_file ) ;
if ( ith_file - > hparams ! = first_file - > hparams ) {
throw format ( " llama.cpp: hparams inconsistent between files " ) ;
}
}
if ( ! llama_mmap : : SUPPORTED ) {
use_mmap = false ;
}
if ( use_mmap & & alignment_prevents_mmap ( ) ) {
fprintf ( stderr , " llama.cpp: can't use mmap because tensors are not aligned; convert to new format to avoid this \n " ) ;
use_mmap = false ;
}
this - > use_mmap = use_mmap ;
for ( llama_load_tensor & lt : tensors_map . tensors ) {
lt . calc_all ( ) ;
}
}
bool alignment_prevents_mmap ( ) {
for ( const llama_load_tensor & lt : tensors_map . tensors ) {
for ( const llama_load_tensor_shard & shard : lt . shards ) {
if ( shard . file_off & 3 ) {
return true ;
}
}
}
return false ;
}
uint32_t guess_n_parts ( ) const {
auto it = tensors_map . name_to_idx . find ( " tok_embeddings.weight " ) ;
if ( it = = tensors_map . name_to_idx . end ( ) ) {
throw std : : string ( " missing tok_embeddings.weight " ) ;
}
const llama_load_tensor & lt = tensors_map . tensors . at ( it - > second ) ;
return file_loaders . at ( 0 ) - > hparams . n_embd / lt . shards . at ( 0 ) . ne . at ( 0 ) ;
}
void calc_sizes ( size_t * ctx_size_p , size_t * mmapped_size_p ) const {
* ctx_size_p = * mmapped_size_p = 0 ;
for ( const llama_load_tensor & lt : tensors_map . tensors ) {
* ctx_size_p + = sizeof ( struct ggml_tensor ) + GGML_OBJECT_SIZE ;
* ( use_mmap ? mmapped_size_p : ctx_size_p ) + = lt . size ;
}
}
2023-05-23 13:04:39 +02:00
struct ggml_tensor * get_tensor ( const std : : string & name , const std : : vector < uint32_t > & ne , ggml_backend backend ) {
2023-04-10 21:59:13 +02:00
auto it = tensors_map . name_to_idx . find ( name ) ;
if ( it = = tensors_map . name_to_idx . end ( ) ) {
throw format ( " llama.cpp: tensor '%s' is missing from model " , name . c_str ( ) ) ;
}
llama_load_tensor & lt = tensors_map . tensors . at ( it - > second ) ;
if ( lt . ne ! = ne ) {
throw format ( " llama.cpp: tensor '%s' has wrong shape; expected %s, got %s " ,
name . c_str ( ) , llama_format_tensor_shape ( ne ) . c_str ( ) , llama_format_tensor_shape ( lt . ne ) . c_str ( ) ) ;
}
2023-04-30 17:51:57 +02:00
2023-05-23 13:04:39 +02:00
return get_tensor_for ( lt , backend ) ;
2023-04-10 21:59:13 +02:00
}
2023-05-23 13:04:39 +02:00
struct ggml_tensor * get_tensor_for ( llama_load_tensor & lt , ggml_backend backend ) {
2023-04-10 21:59:13 +02:00
struct ggml_tensor * tensor ;
if ( lt . ne . size ( ) = = 2 ) {
tensor = ggml_new_tensor_2d ( ggml_ctx , lt . type , lt . ne . at ( 0 ) , lt . ne . at ( 1 ) ) ;
} else {
LLAMA_ASSERT ( lt . ne . size ( ) = = 1 ) ;
tensor = ggml_new_tensor_1d ( ggml_ctx , lt . type , lt . ne . at ( 0 ) ) ;
}
2023-05-14 17:46:19 +02:00
ggml_set_name ( tensor , lt . name . c_str ( ) ) ;
2023-04-10 21:59:13 +02:00
LLAMA_ASSERT ( lt . ggml_tensor = = NULL ) ; // if this fails, we called get_tensor twice on the same tensor
2023-05-23 13:04:39 +02:00
tensor - > backend = backend ;
2023-04-10 21:59:13 +02:00
lt . ggml_tensor = tensor ;
num_ggml_tensors_created + + ;
return tensor ;
}
2023-05-14 17:46:19 +02:00
void done_getting_tensors ( ) const {
2023-04-10 21:59:13 +02:00
if ( num_ggml_tensors_created ! = tensors_map . tensors . size ( ) ) {
throw std : : string ( " llama.cpp: file contained more tensors than expected " ) ;
}
}
void load_all_data ( llama_progress_callback progress_callback , void * progress_callback_user_data , llama_mlock * lmlock ) {
size_t data_size = 0 ;
2023-05-23 13:04:39 +02:00
size_t prefetch_size = 0 ;
2023-04-10 21:59:13 +02:00
for ( const llama_load_tensor & lt : tensors_map . tensors ) {
data_size + = lt . size ;
2023-05-23 13:04:39 +02:00
if ( lt . ggml_tensor - > backend = = GGML_BACKEND_CPU ) {
prefetch_size + = lt . size ;
}
2023-04-10 21:59:13 +02:00
}
if ( use_mmap ) {
2023-05-23 13:04:39 +02:00
mapping . reset ( new llama_mmap ( & file_loaders . at ( 0 ) - > file , prefetch_size ) ) ;
2023-04-10 21:59:13 +02:00
if ( ! lmlock ) {
// Don't call the callback since the actual loading will be lazy
// and we can't measure it.
progress_callback = NULL ;
}
if ( lmlock ) {
lmlock - > init ( mapping - > addr ) ;
}
}
size_t done_size = 0 ;
for ( llama_load_tensor & lt : tensors_map . tensors ) {
2023-05-23 13:04:39 +02:00
if ( lt . ggml_tensor - > backend ! = GGML_BACKEND_CPU ) {
continue ;
}
2023-04-10 21:59:13 +02:00
if ( progress_callback ) {
progress_callback ( ( float ) done_size / data_size , progress_callback_user_data ) ;
}
LLAMA_ASSERT ( lt . ggml_tensor ) ; // unused tensors should have been caught by load_data already
lt . data = ( uint8_t * ) lt . ggml_tensor - > data ;
load_data_for ( lt ) ;
lt . ggml_tensor - > data = lt . data ;
done_size + = lt . size ;
if ( use_mmap & & lmlock ) {
lmlock - > grow_to ( done_size ) ;
}
}
}
void load_data_for ( llama_load_tensor & lt ) {
if ( use_mmap ) {
LLAMA_ASSERT ( lt . shards . size ( ) = = 1 ) ;
lt . data = ( uint8_t * ) mapping - > addr + lt . shards . at ( 0 ) . file_off ;
} else if ( lt . split_type = = SPLIT_NONE ) {
llama_file & file = file_loaders . at ( lt . shards . at ( 0 ) . file_idx ) - > file ;
file . seek ( lt . shards . at ( 0 ) . file_off , SEEK_SET ) ;
file . read_raw ( lt . data , lt . size ) ;
} else if ( lt . split_type = = SPLIT_BY_ROWS ) {
size_t offset = 0 ;
for ( llama_load_tensor_shard & shard : lt . shards ) {
llama_file & file = file_loaders . at ( shard . file_idx ) - > file ;
file . seek ( shard . file_off , SEEK_SET ) ;
file . read_raw ( lt . data + offset , shard . size ) ;
offset + = shard . size ;
}
LLAMA_ASSERT ( offset = = lt . size ) ;
} else if ( lt . split_type = = SPLIT_BY_COLUMNS ) {
// Let's load the data into temporary buffers to ensure the OS performs large loads.
2023-05-14 17:46:19 +02:00
std : : vector < llama_buffer > tmp_bufs ( lt . shards . size ( ) ) ;
2023-04-10 21:59:13 +02:00
for ( size_t i = 0 ; i < lt . shards . size ( ) ; i + + ) {
llama_load_tensor_shard & shard = lt . shards . at ( i ) ;
llama_file & file = file_loaders . at ( shard . file_idx ) - > file ;
file . seek ( shard . file_off , SEEK_SET ) ;
tmp_bufs . at ( i ) . resize ( shard . size ) ;
file . read_raw ( tmp_bufs . at ( i ) . addr , shard . size ) ;
}
// Then reshape.
size_t num_rows = lt . ne . at ( 1 ) ;
size_t per_shard_row_size = lt . shards . at ( 0 ) . size / num_rows ;
size_t out_offset = 0 ;
for ( size_t row = 0 ; row < num_rows ; row + + ) {
for ( llama_buffer & tmp_buf : tmp_bufs ) {
memcpy ( lt . data + out_offset ,
tmp_buf . addr + row * per_shard_row_size ,
per_shard_row_size ) ;
out_offset + = per_shard_row_size ;
}
}
LLAMA_ASSERT ( out_offset = = lt . size ) ;
}
if ( 0 ) {
print_checksum ( lt ) ;
}
}
static void print_checksum ( llama_load_tensor & lt ) {
uint32_t sum = 0 ;
for ( size_t i = 0 ; i < lt . size ; i + + ) {
uint8_t byte = lt . data [ i ] ;
sum = byte + ( sum < < 6 ) + ( sum < < 16 ) - sum ; // sdbm hash
}
fprintf ( stderr , " %s checksum: %#08x (%s, size %zu) \n " , lt . name . c_str ( ) , sum ,
llama_format_tensor_shape ( lt . ne ) . c_str ( ) , lt . size ) ;
}
} ;
2023-03-27 20:00:32 +02:00
//
// 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 ;
2023-04-30 17:51:57 +02:00
const int64_t n_mem = n_layer * n_ctx ;
2023-04-10 21:59:13 +02:00
const int64_t n_elements = n_embd * n_mem ;
2023-03-27 20:00:32 +02:00
cache . buf . resize ( 2u * n_elements * ggml_type_size ( wtype ) + 2u * MB ) ;
struct ggml_init_params params ;
2023-04-10 21:59:13 +02:00
params . mem_size = cache . buf . size ;
params . mem_buffer = cache . buf . addr ;
params . no_alloc = false ;
2023-03-27 20:00:32 +02:00
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 ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( cache . k , " cache_k " ) ;
ggml_set_name ( cache . v , " cache_v " ) ;
2023-03-27 20:00:32 +02:00
return true ;
}
struct llama_context_params llama_context_default_params ( ) {
struct llama_context_params result = {
/*.n_ctx =*/ 512 ,
2023-05-14 17:46:19 +02:00
/*.gpu_layers =*/ 0 ,
/*.seed =*/ - 1 ,
2023-05-23 13:04:39 +02:00
/*.f16_kv =*/ true ,
2023-03-27 20:00:32 +02:00
/*.logits_all =*/ false ,
/*.vocab_only =*/ false ,
2023-04-10 21:59:13 +02:00
/*.use_mmap =*/ true ,
2023-03-27 20:00:32 +02:00
/*.use_mlock =*/ false ,
/*.embedding =*/ false ,
/*.progress_callback =*/ nullptr ,
/*.progress_callback_user_data =*/ nullptr ,
} ;
return result ;
}
2023-04-10 21:59:13 +02:00
bool llama_mmap_supported ( ) {
return llama_mmap : : SUPPORTED ;
}
bool llama_mlock_supported ( ) {
return llama_mlock : : SUPPORTED ;
}
2023-05-23 13:04:39 +02:00
void llama_init_backend ( ) {
ggml_time_init ( ) ;
// needed to initialize f16 tables
{
struct ggml_init_params params = { 0 , NULL , false } ;
struct ggml_context * ctx = ggml_init ( params ) ;
ggml_free ( ctx ) ;
}
}
int64_t llama_time_us ( ) {
return ggml_time_us ( ) ;
}
2023-03-27 20:00:32 +02:00
//
// model loading
//
2023-04-10 21:59:13 +02:00
static const char * llama_file_version_name ( llama_file_version version ) {
switch ( version ) {
case LLAMA_FILE_VERSION_GGML : return " 'ggml' (old version with low tokenizer quality and no mmap support) " ;
case LLAMA_FILE_VERSION_GGMF_V1 : return " ggmf v1 (old version with no mmap support) " ;
2023-05-14 17:46:19 +02:00
case LLAMA_FILE_VERSION_GGJT_V1 : return " ggjt v1 (pre #1405) " ;
2023-05-23 13:04:39 +02:00
case LLAMA_FILE_VERSION_GGJT_V2 : return " ggjt v2 (pre #1508) " ;
case LLAMA_FILE_VERSION_GGJT_V3 : return " ggjt v3 (latest) " ;
2023-04-10 21:59:13 +02:00
}
2023-05-14 17:46:19 +02:00
return " unknown " ;
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
static const char * llama_ftype_name ( enum llama_ftype ftype ) {
switch ( ftype ) {
case LLAMA_FTYPE_ALL_F32 : return " all F32 " ;
case LLAMA_FTYPE_MOSTLY_F16 : return " mostly F16 " ;
case LLAMA_FTYPE_MOSTLY_Q4_0 : return " mostly Q4_0 " ;
case LLAMA_FTYPE_MOSTLY_Q4_1 : return " mostly Q4_1 " ;
case LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 :
return " mostly Q4_1, some F16 " ;
case LLAMA_FTYPE_MOSTLY_Q5_0 : return " mostly Q5_0 " ;
case LLAMA_FTYPE_MOSTLY_Q5_1 : return " mostly Q5_1 " ;
case LLAMA_FTYPE_MOSTLY_Q8_0 : return " mostly Q8_0 " ;
default : return " unknown, may not work " ;
}
}
2023-04-10 21:59:13 +02:00
static const char * llama_model_type_name ( e_model type ) {
switch ( type ) {
case MODEL_7B : return " 7B " ;
case MODEL_13B : return " 13B " ;
case MODEL_30B : return " 30B " ;
case MODEL_65B : return " 65B " ;
default : LLAMA_ASSERT ( false ) ;
}
}
static void llama_model_load_internal (
2023-03-27 20:00:32 +02:00
const std : : string & fname ,
llama_context & lctx ,
int n_ctx ,
2023-05-14 17:46:19 +02:00
int n_gpu_layers ,
2023-03-27 20:00:32 +02:00
ggml_type memory_type ,
2023-04-10 21:59:13 +02:00
bool use_mmap ,
bool use_mlock ,
2023-03-27 20:00:32 +02:00
bool vocab_only ,
llama_progress_callback progress_callback ,
2023-04-10 21:59:13 +02:00
void * progress_callback_user_data ) {
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
lctx . t_start_us = ggml_time_us ( ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
std : : unique_ptr < llama_model_loader > ml ( new llama_model_loader ( fname , use_mmap , vocab_only ) ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
lctx . vocab = std : : move ( ml - > file_loaders . at ( 0 ) - > vocab ) ;
2023-03-27 20:00:32 +02:00
auto & model = lctx . model ;
2023-04-10 21:59:13 +02:00
model . hparams = ml - > file_loaders . at ( 0 ) - > hparams ;
llama_file_version file_version = ml - > file_loaders . at ( 0 ) - > file_version ;
auto & hparams = model . hparams ;
uint32_t n_ff = ( ( 2 * ( 4 * hparams . n_embd ) / 3 + hparams . n_mult - 1 ) / hparams . n_mult ) * hparams . n_mult ;
2023-03-27 20:00:32 +02:00
{
2023-04-10 21:59:13 +02:00
switch ( hparams . n_layer ) {
case 32 : model . type = e_model : : MODEL_7B ; break ;
case 40 : model . type = e_model : : MODEL_13B ; break ;
case 60 : model . type = e_model : : MODEL_30B ; break ;
case 80 : model . type = e_model : : MODEL_65B ; break ;
2023-03-27 20:00:32 +02:00
}
hparams . n_ctx = n_ctx ;
}
{
2023-04-10 21:59:13 +02:00
fprintf ( stderr , " %s: format = %s \n " , __func__ , llama_file_version_name ( file_version ) ) ;
fprintf ( stderr , " %s: n_vocab = %u \n " , __func__ , hparams . n_vocab ) ;
fprintf ( stderr , " %s: n_ctx = %u \n " , __func__ , hparams . n_ctx ) ;
fprintf ( stderr , " %s: n_embd = %u \n " , __func__ , hparams . n_embd ) ;
fprintf ( stderr , " %s: n_mult = %u \n " , __func__ , hparams . n_mult ) ;
fprintf ( stderr , " %s: n_head = %u \n " , __func__ , hparams . n_head ) ;
fprintf ( stderr , " %s: n_layer = %u \n " , __func__ , hparams . n_layer ) ;
fprintf ( stderr , " %s: n_rot = %u \n " , __func__ , hparams . n_rot ) ;
2023-04-30 17:51:57 +02:00
fprintf ( stderr , " %s: ftype = %u (%s) \n " , __func__ , hparams . ftype , llama_ftype_name ( hparams . ftype ) ) ;
2023-04-10 21:59:13 +02:00
fprintf ( stderr , " %s: n_ff = %u \n " , __func__ , n_ff ) ;
fprintf ( stderr , " %s: n_parts = %zu \n " , __func__ , ml - > file_loaders . size ( ) ) ;
fprintf ( stderr , " %s: model size = %s \n " , __func__ , llama_model_type_name ( model . type ) ) ;
2023-03-27 20:00:32 +02:00
}
2023-05-23 13:04:39 +02:00
if ( file_version < LLAMA_FILE_VERSION_GGJT_V2 ) {
2023-05-14 17:46:19 +02:00
if ( hparams . ftype ! = LLAMA_FTYPE_ALL_F32 & &
hparams . ftype ! = LLAMA_FTYPE_MOSTLY_F16 & &
hparams . ftype ! = LLAMA_FTYPE_MOSTLY_Q8_0 ) {
2023-05-23 13:04:39 +02:00
throw format ( " this format is no longer supported (see https://github.com/ggerganov/llama.cpp/pull/1405) " ) ;
}
}
if ( file_version < LLAMA_FILE_VERSION_GGJT_V3 ) {
if ( hparams . ftype = = LLAMA_FTYPE_MOSTLY_Q4_0 | |
hparams . ftype = = LLAMA_FTYPE_MOSTLY_Q4_1 | |
hparams . ftype = = LLAMA_FTYPE_MOSTLY_Q8_0 ) {
throw format ( " this format is no longer supported (see https://github.com/ggerganov/llama.cpp/pull/1508) " ) ;
2023-05-14 17:46:19 +02:00
}
}
2023-03-27 20:00:32 +02:00
if ( vocab_only ) {
2023-04-10 21:59:13 +02:00
return ;
2023-03-27 20:00:32 +02:00
}
auto & ctx = model . ctx ;
2023-05-14 17:46:19 +02:00
size_t ctx_size ;
size_t mmapped_size ;
2023-04-10 21:59:13 +02:00
ml - > calc_sizes ( & ctx_size , & mmapped_size ) ;
2023-05-23 13:04:39 +02:00
fprintf ( stderr , " %s: ggml ctx size = %7.2f MB \n " , __func__ , ctx_size / 1024.0 / 1024.0 ) ;
2023-03-27 20:00:32 +02:00
// create the ggml context
{
lctx . model . buf . resize ( ctx_size ) ;
2023-04-10 21:59:13 +02:00
if ( use_mlock ) {
lctx . model . mlock_buf . init ( lctx . model . buf . addr ) ;
lctx . model . mlock_buf . grow_to ( lctx . model . buf . size ) ;
}
2023-03-27 20:00:32 +02:00
struct ggml_init_params params = {
2023-04-10 21:59:13 +02:00
/*.mem_size =*/ lctx . model . buf . size ,
/*.mem_buffer =*/ lctx . model . buf . addr ,
/*.no_alloc =*/ ml - > use_mmap ,
2023-03-27 20:00:32 +02:00
} ;
model . ctx = ggml_init ( params ) ;
if ( ! model . ctx ) {
2023-04-10 21:59:13 +02:00
throw format ( " ggml_init() failed " ) ;
2023-03-27 20:00:32 +02:00
}
}
2023-05-23 13:04:39 +02:00
# ifdef GGML_USE_CUBLAS
2023-06-25 15:13:50 +02:00
# define LLAMA_BACKEND_OFFLOAD GGML_BACKEND_GPU
2023-05-23 13:04:39 +02:00
# else
# define LLAMA_BACKEND_OFFLOAD GGML_BACKEND_CPU
# endif
2023-03-27 20:00:32 +02:00
// prepare memory for the weights
2023-05-23 13:04:39 +02:00
size_t vram_total = 0 ;
2023-03-27 20:00:32 +02:00
{
2023-04-10 21:59:13 +02:00
const uint32_t n_embd = hparams . n_embd ;
const uint32_t n_layer = hparams . n_layer ;
const uint32_t n_vocab = hparams . n_vocab ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
ml - > ggml_ctx = ctx ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
model . tok_embeddings = ml - > get_tensor ( " tok_embeddings.weight " , { n_embd , n_vocab } , GGML_BACKEND_CPU ) ;
model . norm = ml - > get_tensor ( " norm.weight " , { n_embd } , GGML_BACKEND_CPU ) ;
// "output" tensor
{
ggml_backend backend_output ;
if ( n_gpu_layers > int ( n_layer ) ) { // NOLINT
backend_output = LLAMA_BACKEND_OFFLOAD ;
} else {
backend_output = GGML_BACKEND_CPU ;
}
model . output = ml - > get_tensor ( " output.weight " , { n_embd , n_vocab } , backend_output ) ;
}
const int i_gpu_start = n_layer - n_gpu_layers ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
model . layers . resize ( n_layer ) ;
for ( uint32_t i = 0 ; i < n_layer ; + + i ) {
2023-05-23 13:04:39 +02:00
const ggml_backend backend = int ( i ) < i_gpu_start ? GGML_BACKEND_CPU : LLAMA_BACKEND_OFFLOAD ;
2023-03-27 20:00:32 +02:00
auto & layer = model . layers [ i ] ;
2023-04-10 21:59:13 +02:00
std : : string layers_i = " layers. " + std : : to_string ( i ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
layer . attention_norm = ml - > get_tensor ( layers_i + " .attention_norm.weight " , { n_embd } , backend ) ;
layer . wq = ml - > get_tensor ( layers_i + " .attention.wq.weight " , { n_embd , n_embd } , backend ) ;
layer . wk = ml - > get_tensor ( layers_i + " .attention.wk.weight " , { n_embd , n_embd } , backend ) ;
layer . wv = ml - > get_tensor ( layers_i + " .attention.wv.weight " , { n_embd , n_embd } , backend ) ;
layer . wo = ml - > get_tensor ( layers_i + " .attention.wo.weight " , { n_embd , n_embd } , backend ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
layer . ffn_norm = ml - > get_tensor ( layers_i + " .ffn_norm.weight " , { n_embd } , backend ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
layer . w1 = ml - > get_tensor ( layers_i + " .feed_forward.w1.weight " , { n_embd , n_ff } , backend ) ;
layer . w2 = ml - > get_tensor ( layers_i + " .feed_forward.w2.weight " , { n_ff , n_embd } , backend ) ;
layer . w3 = ml - > get_tensor ( layers_i + " .feed_forward.w3.weight " , { n_embd , n_ff } , backend ) ;
2023-03-27 20:00:32 +02:00
2023-06-25 15:13:50 +02:00
if ( backend = = GGML_BACKEND_GPU ) {
2023-05-23 13:04:39 +02:00
vram_total + =
ggml_nbytes ( layer . attention_norm ) + ggml_nbytes ( layer . wq ) + ggml_nbytes ( layer . wk ) +
ggml_nbytes ( layer . wv ) + ggml_nbytes ( layer . wo ) + ggml_nbytes ( layer . attention_norm ) +
ggml_nbytes ( layer . w1 ) + ggml_nbytes ( layer . w2 ) + ggml_nbytes ( layer . w3 ) ;
}
2023-03-27 20:00:32 +02:00
}
}
2023-04-10 21:59:13 +02:00
ml - > done_getting_tensors ( ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
// print memory requirements
{
const size_t scale = memory_type = = GGML_TYPE_F32 ? 2 : 1 ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
// this is the total memory required to run the inference
const size_t mem_required =
ctx_size +
mmapped_size - vram_total + // weights in VRAM not in memory
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 ) ;
2023-03-27 20:00:32 +02:00
2023-05-14 17:46:19 +02:00
# ifdef GGML_USE_CUBLAS
const int n_gpu = std : : min ( n_gpu_layers , int ( hparams . n_layer ) ) ;
fprintf ( stderr , " %s: [cublas] offloading %d layers to GPU \n " , __func__ , n_gpu ) ;
2023-05-23 13:04:39 +02:00
if ( n_gpu_layers > ( int ) hparams . n_layer ) {
fprintf ( stderr , " %s: [cublas] offloading output layer to GPU \n " , __func__ ) ;
}
fprintf ( stderr , " %s: [cublas] total VRAM used: %zu MB \n " , __func__ , vram_total / 1024 / 1024 ) ;
# else
( void ) n_gpu_layers ;
# endif
}
2023-05-14 17:46:19 +02:00
2023-05-23 13:04:39 +02:00
// populate `tensors_by_name`
for ( llama_load_tensor & lt : ml - > tensors_map . tensors ) {
model . tensors_by_name . emplace_back ( lt . name , lt . ggml_tensor ) ;
}
2023-05-14 17:46:19 +02:00
2023-05-23 13:04:39 +02:00
ml - > load_all_data ( progress_callback , progress_callback_user_data , use_mlock ? & lctx . model . mlock_mmap : NULL ) ;
2023-05-14 17:46:19 +02:00
2023-05-23 13:04:39 +02:00
# ifdef GGML_USE_CUBLAS
{
size_t done_size = 0 ;
size_t data_size = 0 ;
for ( llama_load_tensor & lt : ml - > tensors_map . tensors ) {
data_size + = lt . size ;
if ( lt . ggml_tensor - > backend = = GGML_BACKEND_CPU ) {
done_size + = lt . size ;
}
2023-05-14 17:46:19 +02:00
}
2023-05-23 13:04:39 +02:00
for ( llama_load_tensor & lt : ml - > tensors_map . tensors ) {
2023-06-25 15:13:50 +02:00
if ( lt . ggml_tensor - > backend ! = GGML_BACKEND_GPU ) {
2023-05-23 13:04:39 +02:00
continue ;
}
if ( progress_callback ) {
progress_callback ( ( float ) done_size / data_size , progress_callback_user_data ) ;
}
ggml_cuda_load_data ( fname . c_str ( ) , lt . ggml_tensor , lt . shards . at ( 0 ) . file_off ) ;
done_size + = lt . size ;
2023-05-14 17:46:19 +02:00
}
2023-05-23 13:04:39 +02:00
}
# endif // GGML_USE_CUBLAS
2023-05-14 17:46:19 +02:00
2023-05-23 13:04:39 +02:00
if ( progress_callback ) {
progress_callback ( 1.0f , progress_callback_user_data ) ;
2023-05-14 17:46:19 +02:00
}
2023-05-23 13:04:39 +02:00
model . mapping = std : : move ( ml - > mapping ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
// 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 ;
}
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
static bool llama_model_load (
const std : : string & fname ,
llama_context & lctx ,
int n_ctx ,
2023-05-14 17:46:19 +02:00
int n_gpu_layers ,
2023-04-10 21:59:13 +02:00
ggml_type memory_type ,
bool use_mmap ,
bool use_mlock ,
bool vocab_only ,
llama_progress_callback progress_callback ,
void * progress_callback_user_data ) {
try {
2023-05-14 17:46:19 +02:00
llama_model_load_internal ( fname , lctx , n_ctx , n_gpu_layers , memory_type , use_mmap , use_mlock ,
2023-04-10 21:59:13 +02:00
vocab_only , progress_callback , progress_callback_user_data ) ;
return true ;
} catch ( const std : : string & err ) {
fprintf ( stderr , " error loading model: %s \n " , err . c_str ( ) ) ;
return false ;
2023-03-27 20:00:32 +02:00
}
}
// 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 ,
2023-09-07 11:33:12 +02:00
int n_threads ) {
2023-05-14 17:46:19 +02:00
// enforce that the first token is BOS
if ( n_past = = 0 & & tokens [ 0 ] ! = llama_token_bos ( ) ) {
fprintf ( stderr , " %s: first token must be BOS \n " , __func__ ) ;
return false ;
}
2023-03-27 20:00:32 +02:00
const int64_t t_start_us = ggml_time_us ( ) ;
const int N = n_tokens ;
const auto & model = lctx . model ;
const auto & hparams = model . hparams ;
2023-05-14 17:46:19 +02:00
const auto & kv_self = model . kv_self ;
2023-03-27 20:00:32 +02:00
LLAMA_ASSERT ( ! ! kv_self . ctx ) ;
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 ;
2023-09-07 11:33:12 +02:00
const float eps = 5e-6 f ; // TODO: take from hparams
2023-03-27 20:00:32 +02:00
auto & mem_per_token = lctx . mem_per_token ;
auto & buf_compute = lctx . buf_compute ;
struct ggml_init_params params = {
2023-04-10 21:59:13 +02:00
/*.mem_size =*/ buf_compute . size ,
/*.mem_buffer =*/ buf_compute . addr ,
/*.no_alloc =*/ false ,
2023-03-27 20:00:32 +02:00
} ;
struct ggml_context * ctx0 = ggml_init ( params ) ;
// 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
ggml_cgraph gf = { } ;
2023-09-07 11:33:12 +02:00
n_threads = N > = 32 & & ggml_cpu_has_blas ( ) & & ! ggml_cpu_has_gpublas ( ) ? 1 : n_threads ;
2023-03-27 20:00:32 +02:00
struct ggml_tensor * embd = ggml_new_tensor_1d ( ctx0 , GGML_TYPE_I32 , N ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( embd , " embd " ) ;
2023-03-27 20:00:32 +02:00
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 ;
lctx . use_buf ( ctx0 , 0 ) ;
// norm
{
2023-09-07 11:33:12 +02:00
cur = ggml_rms_norm ( ctx0 , inpL , eps ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
// cur = cur*attention_norm(broadcasted)
cur = ggml_mul ( ctx0 , cur , model . layers [ il ] . attention_norm ) ;
2023-03-27 20:00:32 +02:00
}
// self-attention
{
2023-04-10 21:59:13 +02:00
// compute Q and K and RoPE them
2023-07-03 18:24:01 +02:00
struct ggml_tensor * Qcur = ggml_rope_inplace ( ctx0 , ggml_reshape_3d ( ctx0 , ggml_mul_mat ( ctx0 , model . layers [ il ] . wq , cur ) , n_embd / n_head , n_head , N ) , n_past , n_rot , 0 , 0 ) ;
struct ggml_tensor * Kcur = ggml_rope_inplace ( ctx0 , ggml_reshape_3d ( ctx0 , ggml_mul_mat ( ctx0 , model . layers [ il ] . wk , cur ) , n_embd / n_head , n_head , N ) , n_past , n_rot , 0 , 0 ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( Qcur , " Qcur " ) ;
ggml_set_name ( Kcur , " Kcur " ) ;
2023-03-27 20:00:32 +02:00
// store key and value to memory
2023-04-10 21:59:13 +02:00
{
// compute the transposed [N, n_embd] V matrix
struct ggml_tensor * Vcur = ggml_transpose ( ctx0 , ggml_reshape_2d ( ctx0 , ggml_mul_mat ( ctx0 , model . layers [ il ] . wv , cur ) , n_embd , N ) ) ;
2023-03-27 20:00:32 +02:00
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 ) ) ;
2023-04-10 21:59:13 +02:00
struct ggml_tensor * v = ggml_view_2d ( ctx0 , kv_self . v , N , n_embd ,
( n_ctx ) * ggml_element_size ( kv_self . v ) ,
( il * n_ctx ) * ggml_element_size ( kv_self . v ) * n_embd + n_past * ggml_element_size ( kv_self . v ) ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
// important: storing RoPE-ed version of K in the KV cache!
2023-03-27 20:00:32 +02:00
ggml_build_forward_expand ( & gf , ggml_cpy ( ctx0 , Kcur , k ) ) ;
ggml_build_forward_expand ( & gf , ggml_cpy ( ctx0 , Vcur , v ) ) ;
}
struct ggml_tensor * Q =
ggml_permute ( ctx0 ,
2023-04-10 21:59:13 +02:00
Qcur ,
2023-03-27 20:00:32 +02:00
0 , 2 , 1 , 3 ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( Q , " Q " ) ;
2023-03-27 20:00:32 +02:00
struct ggml_tensor * K =
ggml_permute ( ctx0 ,
2023-04-10 21:59:13 +02:00
ggml_reshape_3d ( ctx0 ,
ggml_view_1d ( ctx0 , kv_self . k , ( n_past + N ) * n_embd , il * n_ctx * ggml_element_size ( kv_self . k ) * n_embd ) ,
n_embd / n_head , n_head , n_past + N ) ,
2023-03-27 20:00:32 +02:00
0 , 2 , 1 , 3 ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( K , " K " ) ;
2023-03-27 20:00:32 +02:00
// K * Q
struct ggml_tensor * KQ = ggml_mul_mat ( ctx0 , K , Q ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( KQ , " KQ " ) ;
2023-03-27 20:00:32 +02:00
// KQ_scaled = KQ / sqrt(n_embd/n_head)
2023-05-14 17:46:19 +02:00
struct ggml_tensor * KQ_scale = ggml_new_f32 ( ctx0 , 1.0f / sqrtf ( float ( n_embd ) / n_head ) ) ;
ggml_set_name ( KQ_scale , " 1/sqrt(n_embd/n_head) " ) ;
// KQ_scaled shape [n_past + N, N, n_head, 1]
struct ggml_tensor * KQ_scaled = ggml_scale_inplace ( ctx0 , KQ , KQ_scale ) ;
ggml_set_name ( KQ_scaled , " KQ_scaled " ) ;
2023-03-27 20:00:32 +02:00
// KQ_masked = mask_past(KQ_scaled)
2023-05-14 17:46:19 +02:00
struct ggml_tensor * KQ_masked = ggml_diag_mask_inf_inplace ( ctx0 , KQ_scaled , n_past ) ;
ggml_set_name ( KQ_masked , " KQ_masked " ) ;
2023-03-27 20:00:32 +02:00
// KQ = soft_max(KQ_masked)
2023-05-14 17:46:19 +02:00
struct ggml_tensor * KQ_soft_max = ggml_soft_max_inplace ( ctx0 , KQ_masked ) ;
ggml_set_name ( KQ_soft_max , " KQ_soft_max " ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
// split cached V into n_head heads
struct ggml_tensor * V =
ggml_view_3d ( ctx0 , kv_self . v ,
n_past + N , n_embd / n_head , n_head ,
n_ctx * ggml_element_size ( kv_self . v ) ,
n_ctx * ggml_element_size ( kv_self . v ) * n_embd / n_head ,
il * n_ctx * ggml_element_size ( kv_self . v ) * n_embd ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( V , " V " ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
# if 1
struct ggml_tensor * KQV = ggml_mul_mat ( ctx0 , V , KQ_soft_max ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( KQV , " KQV " ) ;
2023-04-10 21:59:13 +02:00
# else
// make V contiguous in memory to speed up the matmul, however we waste time on the copy
// on M1 this is faster for the perplexity computation, but ~5% slower for the single-token generation
// is there a better way?
struct ggml_tensor * V_cont = ggml_cpy ( ctx0 , V , ggml_new_tensor_3d ( ctx0 , kv_self . v - > type , n_past + N , n_embd / n_head , n_head ) ) ;
struct ggml_tensor * KQV = ggml_mul_mat ( ctx0 , V_cont , KQ_soft_max ) ;
# endif
2023-03-27 20:00:32 +02:00
// KQV_merged = KQV.permute(0, 2, 1, 3)
struct ggml_tensor * KQV_merged = ggml_permute ( ctx0 , KQV , 0 , 2 , 1 , 3 ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( KQV_merged , " KQV_merged " ) ;
2023-03-27 20:00:32 +02:00
// 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 ) ) ;
2023-05-14 17:46:19 +02:00
ggml_set_name ( cur , " KQV_merged_contiguous " ) ;
2023-03-27 20:00:32 +02:00
// projection (no bias)
cur = ggml_mul_mat ( ctx0 ,
model . layers [ il ] . wo ,
cur ) ;
}
lctx . use_buf ( ctx0 , 1 ) ;
struct ggml_tensor * inpFF = ggml_add ( ctx0 , cur , inpSA ) ;
// feed-forward network
{
// norm
{
2023-09-07 11:33:12 +02:00
cur = ggml_rms_norm ( ctx0 , inpFF , eps ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
// cur = cur*ffn_norm(broadcasted)
cur = ggml_mul ( ctx0 , cur , model . layers [ il ] . ffn_norm ) ;
2023-03-27 20:00:32 +02:00
}
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 ) ;
}
cur = ggml_add ( ctx0 , cur , inpFF ) ;
// input for next layer
inpL = cur ;
}
lctx . use_buf ( ctx0 , 0 ) ;
// used at the end to optionally extract the embeddings
struct ggml_tensor * embeddings = NULL ;
// norm
{
2023-09-07 11:33:12 +02:00
inpL = ggml_rms_norm ( ctx0 , inpL , eps ) ;
2023-03-27 20:00:32 +02:00
2023-05-23 13:04:39 +02:00
// inpL = inpL*norm(broadcasted)
inpL = ggml_mul ( ctx0 , inpL , model . norm ) ;
2023-03-27 20:00:32 +02:00
embeddings = inpL ;
}
// lm_head
inpL = ggml_mul_mat ( ctx0 , model . output , inpL ) ;
lctx . use_buf ( ctx0 , - 1 ) ;
// logits -> probs
2023-05-14 17:46:19 +02:00
//inpL = ggml_soft_max_inplace(ctx0, inpL);
2023-03-27 20:00:32 +02:00
// run the computation
2023-09-07 11:33:12 +02:00
ggml_build_forward_expand ( & gf , inpL ) ;
ggml_graph_compute_with_ctx ( ctx0 , & gf , n_threads ) ;
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
# ifdef GGML_PERF
2023-04-10 21:59:13 +02:00
// print timing information per ggml operation (for debugging purposes)
// requires GGML_PERF to be defined
2023-04-30 17:51:57 +02:00
ggml_graph_print ( & gf ) ;
# endif
2023-04-10 21:59:13 +02:00
// plot the computation graph in dot format (for debugging purposes)
2023-03-27 20:00:32 +02:00
//if (n_past%100 == 0) {
2023-04-10 21:59:13 +02:00
// ggml_graph_dump_dot(&gf, NULL, "llama.dot");
2023-03-27 20:00:32 +02:00
//}
//embd_w.resize(n_vocab*N);
//memcpy(embd_w.data(), ggml_get_data(inpL), sizeof(float)*n_vocab*N);
2023-05-08 19:59:21 +02:00
// update kv token count
lctx . model . kv_self . n = n_past + N ;
2023-03-27 20:00:32 +02:00
// 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
2023-05-14 17:46:19 +02:00
if ( ! lctx . embedding . empty ( ) ) {
2023-03-27 20:00:32 +02:00
auto & embedding_out = lctx . embedding ;
embedding_out . resize ( n_embd ) ;
memcpy ( embedding_out . data ( ) , ( float * ) ggml_get_data ( embeddings ) + ( n_embd * ( N - 1 ) ) , sizeof ( float ) * n_embd ) ;
}
if ( mem_per_token = = 0 ) {
mem_per_token = ggml_used_mem ( ctx0 ) / N ;
}
#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
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 + + ;
}
else if ( N > 1 ) {
lctx . t_p_eval_us + = ggml_time_us ( ) - t_start_us ;
lctx . n_p_eval + = N ;
}
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 ;
} ;
2023-05-14 17:46:19 +02:00
static_assert ( std : : is_trivially_copyable < llama_sp_symbol > : : value , " llama_sp_symbol is not trivially copyable " ) ;
2023-03-27 20:00:32 +02:00
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 ;
size_t char_len = std : : min ( text . size ( ) - offs , utf8_len ( text [ offs ] ) ) ;
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 + + ;
2023-05-14 17:46:19 +02:00
symbols_ . emplace_back ( sym ) ;
2023-03-27 20:00:32 +02:00
}
// 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 ;
2023-05-14 17:46:19 +02:00
if ( text . empty ( ) ) {
2023-03-27 20:00:32 +02:00
return output ;
}
if ( bos ) {
2023-05-14 17:46:19 +02:00
output . push_back ( llama_token_bos ( ) ) ;
2023-03-27 20:00:32 +02:00
}
tokenizer . tokenize ( text , output ) ;
return output ;
}
//
// sampling
//
2023-04-30 17:51:57 +02:00
void llama_sample_softmax ( struct llama_context * ctx , llama_token_data_array * candidates ) {
assert ( candidates - > size > 0 ) ;
const int64_t t_start_sample_us = ggml_time_us ( ) ;
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
// Sort the logits in descending order
if ( ! candidates - > sorted ) {
std : : sort ( candidates - > data , candidates - > data + candidates - > size , [ ] ( const llama_token_data & a , const llama_token_data & b ) {
return a . logit > b . logit ;
} ) ;
candidates - > sorted = true ;
}
float max_l = candidates - > data [ 0 ] . logit ;
float cum_sum = 0.0f ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
float p = expf ( candidates - > data [ i ] . logit - max_l ) ;
candidates - > data [ i ] . p = p ;
cum_sum + = p ;
}
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
candidates - > data [ i ] . p / = cum_sum ;
}
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
2023-03-27 20:00:32 +02:00
}
2023-04-30 17:51:57 +02:00
void llama_sample_top_k ( struct llama_context * ctx , llama_token_data_array * candidates , int k , size_t min_keep ) {
const int64_t t_start_sample_us = ggml_time_us ( ) ;
k = std : : max ( k , ( int ) min_keep ) ;
k = std : : min ( k , ( int ) candidates - > size ) ;
// Sort scores in descending order
if ( ! candidates - > sorted ) {
auto comp = [ ] ( const llama_token_data & a , const llama_token_data & b ) {
return a . logit > b . logit ;
} ;
if ( k = = ( int ) candidates - > size ) {
std : : sort ( candidates - > data , candidates - > data + candidates - > size , comp ) ;
} else {
std : : partial_sort ( candidates - > data , candidates - > data + k , candidates - > data + candidates - > size , comp ) ;
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
candidates - > sorted = true ;
}
candidates - > size = k ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
}
2023-04-10 21:59:13 +02:00
2023-04-30 17:51:57 +02:00
void llama_sample_top_p ( struct llama_context * ctx , llama_token_data_array * candidates , float p , size_t min_keep ) {
if ( p > = 1.0f ) {
return ;
}
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
const int64_t t_start_sample_us = ggml_time_us ( ) ;
llama_sample_softmax ( ctx , candidates ) ;
// Compute the cumulative probabilities
float cum_sum = 0.0f ;
size_t last_idx = candidates - > size ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
cum_sum + = candidates - > data [ i ] . p ;
// Check if the running sum is greater than p or if we have kept at least min_keep tokens
if ( cum_sum > p & & i > = min_keep ) {
last_idx = i ;
break ;
2023-03-27 20:00:32 +02:00
}
}
2023-04-30 17:51:57 +02:00
// Resize the output vector to keep only the top-p tokens
candidates - > size = last_idx ;
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
void llama_sample_tail_free ( struct llama_context * ctx , llama_token_data_array * candidates , float z , size_t min_keep ) {
if ( z > = 1.0f | | candidates - > size < = 2 ) {
return ;
2023-03-27 20:00:32 +02:00
}
2023-04-30 17:51:57 +02:00
const int64_t t_start_sample_us = ggml_time_us ( ) ;
llama_sample_softmax ( nullptr , candidates ) ;
// Compute the first and second derivatives
std : : vector < float > first_derivatives ( candidates - > size - 1 ) ;
std : : vector < float > second_derivatives ( candidates - > size - 2 ) ;
for ( size_t i = 0 ; i < first_derivatives . size ( ) ; + + i ) {
first_derivatives [ i ] = candidates - > data [ i ] . p - candidates - > data [ i + 1 ] . p ;
}
for ( size_t i = 0 ; i < second_derivatives . size ( ) ; + + i ) {
second_derivatives [ i ] = first_derivatives [ i ] - first_derivatives [ i + 1 ] ;
2023-03-27 20:00:32 +02:00
}
2023-04-30 17:51:57 +02:00
// Calculate absolute value of second derivatives
for ( size_t i = 0 ; i < second_derivatives . size ( ) ; + + i ) {
second_derivatives [ i ] = abs ( second_derivatives [ i ] ) ;
}
// Normalize the second derivatives
float second_derivatives_sum = std : : accumulate ( second_derivatives . begin ( ) , second_derivatives . end ( ) , 0.0f ) ;
for ( float & value : second_derivatives ) {
value / = second_derivatives_sum ;
}
float cum_sum = 0.0f ;
size_t last_idx = candidates - > size ;
for ( size_t i = 0 ; i < second_derivatives . size ( ) ; + + i ) {
cum_sum + = second_derivatives [ i ] ;
// Check if the running sum is greater than z or if we have kept at least min_keep tokens
if ( cum_sum > z & & i > = min_keep ) {
last_idx = i ;
break ;
2023-03-27 20:00:32 +02:00
}
}
2023-04-30 17:51:57 +02:00
// Resize the output vector to keep only the tokens above the tail location
candidates - > size = last_idx ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
void llama_sample_typical ( struct llama_context * ctx , llama_token_data_array * candidates , float p , size_t min_keep ) {
// Reference implementation:
// https://github.com/huggingface/transformers/compare/main...cimeister:typical-sampling:typical-pr
if ( p > = 1.0f ) {
return ;
}
const int64_t t_start_sample_us = ggml_time_us ( ) ;
// Compute the softmax of logits and calculate entropy
llama_sample_softmax ( nullptr , candidates ) ;
float entropy = 0.0f ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
entropy + = - candidates - > data [ i ] . p * logf ( candidates - > data [ i ] . p ) ;
}
// Compute the absolute difference between negative log probability and entropy for each candidate
std : : vector < float > shifted_scores ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
float shifted_score = fabsf ( - logf ( candidates - > data [ i ] . p ) - entropy ) ;
shifted_scores . push_back ( shifted_score ) ;
}
// Sort tokens based on the shifted_scores and their corresponding indices
std : : vector < size_t > indices ( candidates - > size ) ;
std : : iota ( indices . begin ( ) , indices . end ( ) , 0 ) ;
std : : sort ( indices . begin ( ) , indices . end ( ) , [ & ] ( size_t a , size_t b ) {
return shifted_scores [ a ] < shifted_scores [ b ] ;
} ) ;
// Compute the cumulative probabilities
float cum_sum = 0.0f ;
size_t last_idx = indices . size ( ) ;
for ( size_t i = 0 ; i < indices . size ( ) ; + + i ) {
size_t idx = indices [ i ] ;
cum_sum + = candidates - > data [ idx ] . p ;
// Check if the running sum is greater than typical or if we have kept at least min_keep tokens
if ( cum_sum > p & & i > = min_keep - 1 ) {
last_idx = i + 1 ;
break ;
}
}
// Resize the output vector to keep only the locally typical tokens
std : : vector < llama_token_data > new_candidates ;
for ( size_t i = 0 ; i < last_idx ; + + i ) {
size_t idx = indices [ i ] ;
new_candidates . push_back ( candidates - > data [ idx ] ) ;
}
// Replace the data in candidates with the new_candidates data
std : : copy ( new_candidates . begin ( ) , new_candidates . end ( ) , candidates - > data ) ;
candidates - > size = new_candidates . size ( ) ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
void llama_sample_temperature ( struct llama_context * ctx , llama_token_data_array * candidates_p , float temp ) {
const int64_t t_start_sample_us = ggml_time_us ( ) ;
for ( size_t i = 0 ; i < candidates_p - > size ; + + i ) {
candidates_p - > data [ i ] . logit / = temp ;
}
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
2023-05-14 17:46:19 +02:00
void llama_sample_repetition_penalty ( struct llama_context * ctx , llama_token_data_array * candidates , const llama_token * last_tokens , size_t last_tokens_size , float penalty ) {
2023-04-30 17:51:57 +02:00
if ( last_tokens_size = = 0 | | penalty = = 1.0f ) {
return ;
}
const int64_t t_start_sample_us = ggml_time_us ( ) ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
2023-05-14 17:46:19 +02:00
const auto * token_iter = std : : find ( last_tokens , last_tokens + last_tokens_size , candidates - > data [ i ] . id ) ;
2023-04-30 17:51:57 +02:00
if ( token_iter = = last_tokens + last_tokens_size ) {
continue ;
}
// The academic publication that described this technique actually just only divided, but that would cause tokens with negative logits to become more likely, which is obviously wrong.
// This is common fix for this problem, which is to multiply by the penalty instead of dividing.
if ( candidates - > data [ i ] . logit < = 0 ) {
candidates - > data [ i ] . logit * = penalty ;
} else {
candidates - > data [ i ] . logit / = penalty ;
}
}
candidates - > sorted = false ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
2023-05-14 17:46:19 +02:00
void llama_sample_frequency_and_presence_penalties ( struct llama_context * ctx , llama_token_data_array * candidates , const llama_token * last_tokens_p , size_t last_tokens_size , float alpha_frequency , float alpha_presence ) {
2023-04-30 17:51:57 +02:00
if ( last_tokens_size = = 0 | | ( alpha_frequency = = 0.0f & & alpha_presence = = 0.0f ) ) {
return ;
}
const int64_t t_start_sample_us = ggml_time_us ( ) ;
// Create a frequency map to count occurrences of each token in last_tokens
std : : unordered_map < llama_token , int > token_count ;
for ( size_t i = 0 ; i < last_tokens_size ; + + i ) {
token_count [ last_tokens_p [ i ] ] + + ;
}
// Apply frequency and presence penalties to the candidates
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
auto token_iter = token_count . find ( candidates - > data [ i ] . id ) ;
if ( token_iter = = token_count . end ( ) ) {
continue ;
}
int count = token_iter - > second ;
candidates - > data [ i ] . logit - = float ( count ) * alpha_frequency + float ( count > 0 ) * alpha_presence ;
}
candidates - > sorted = false ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
}
llama_token llama_sample_token_mirostat ( struct llama_context * ctx , llama_token_data_array * candidates , float tau , float eta , int m , float * mu ) {
assert ( ctx ) ;
auto N = float ( llama_n_vocab ( ctx ) ) ;
int64_t t_start_sample_us ;
t_start_sample_us = ggml_time_us ( ) ;
llama_sample_softmax ( nullptr , candidates ) ;
// Estimate s_hat using the most probable m tokens
float s_hat = 0.0 ;
float sum_ti_bi = 0.0 ;
float sum_ti_sq = 0.0 ;
for ( size_t i = 0 ; i < size_t ( m - 1 ) & & i < candidates - > size - 1 ; + + i ) {
float t_i = logf ( float ( i + 2 ) / float ( i + 1 ) ) ;
float b_i = logf ( candidates - > data [ i ] . p / candidates - > data [ i + 1 ] . p ) ;
sum_ti_bi + = t_i * b_i ;
sum_ti_sq + = t_i * t_i ;
}
s_hat = sum_ti_bi / sum_ti_sq ;
// Compute k from the estimated s_hat and target surprise value
float epsilon_hat = s_hat - 1 ;
float k = powf ( ( epsilon_hat * powf ( 2 , * mu ) ) / ( 1 - powf ( N , - epsilon_hat ) ) , 1 / s_hat ) ;
// Sample the next word X using top-k sampling
2023-05-14 17:46:19 +02:00
llama_sample_top_k ( nullptr , candidates , int ( k ) , 1 ) ;
2023-04-30 17:51:57 +02:00
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
llama_token X = llama_sample_token ( ctx , candidates ) ;
t_start_sample_us = ggml_time_us ( ) ;
// Compute error as the difference between observed surprise and target surprise value
size_t X_idx = std : : distance ( candidates - > data , std : : find_if ( candidates - > data , candidates - > data + candidates - > size , [ & ] ( const llama_token_data & candidate ) {
return candidate . id = = X ;
} ) ) ;
float observed_surprise = - log2f ( candidates - > data [ X_idx ] . p ) ;
float e = observed_surprise - tau ;
// Update mu using the learning rate and error
* mu = * mu - eta * e ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
ctx - > n_sample + + ;
}
return X ;
}
llama_token llama_sample_token_mirostat_v2 ( struct llama_context * ctx , llama_token_data_array * candidates , float tau , float eta , float * mu ) {
assert ( ctx ) ;
int64_t t_start_sample_us ;
t_start_sample_us = ggml_time_us ( ) ;
llama_sample_softmax ( ctx , candidates ) ;
// Truncate the words with surprise values greater than mu
candidates - > size = std : : distance ( candidates - > data , std : : find_if ( candidates - > data , candidates - > data + candidates - > size , [ & ] ( const llama_token_data & candidate ) {
return - log2f ( candidate . p ) > * mu ;
} ) ) ;
// Normalize the probabilities of the remaining words
llama_sample_softmax ( ctx , candidates ) ;
// Sample the next word X from the remaining words
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
llama_token X = llama_sample_token ( ctx , candidates ) ;
t_start_sample_us = ggml_time_us ( ) ;
// Compute error as the difference between observed surprise and target surprise value
size_t X_idx = std : : distance ( candidates - > data , std : : find_if ( candidates - > data , candidates - > data + candidates - > size , [ & ] ( const llama_token_data & candidate ) {
return candidate . id = = X ;
} ) ) ;
float observed_surprise = - log2f ( candidates - > data [ X_idx ] . p ) ;
float e = observed_surprise - tau ;
// Update mu using the learning rate and error
* mu = * mu - eta * e ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
}
return X ;
}
llama_token llama_sample_token_greedy ( struct llama_context * ctx , llama_token_data_array * candidates ) {
const int64_t t_start_sample_us = ggml_time_us ( ) ;
// Find max element
2023-05-14 17:46:19 +02:00
auto * max_iter = std : : max_element ( candidates - > data , candidates - > data + candidates - > size , [ ] ( const llama_token_data & a , const llama_token_data & b ) {
2023-04-30 17:51:57 +02:00
return a . logit < b . logit ;
} ) ;
llama_token result = max_iter - > id ;
if ( ctx ) {
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
ctx - > n_sample + + ;
}
return result ;
}
llama_token llama_sample_token ( struct llama_context * ctx , llama_token_data_array * candidates ) {
assert ( ctx ) ;
const int64_t t_start_sample_us = ggml_time_us ( ) ;
llama_sample_softmax ( nullptr , candidates ) ;
std : : vector < float > probs ;
probs . reserve ( candidates - > size ) ;
for ( size_t i = 0 ; i < candidates - > size ; + + i ) {
probs . push_back ( candidates - > data [ i ] . p ) ;
}
2023-03-27 20:00:32 +02:00
std : : discrete_distribution < > dist ( probs . begin ( ) , probs . end ( ) ) ;
2023-04-30 17:51:57 +02:00
auto & rng = ctx - > rng ;
2023-03-27 20:00:32 +02:00
int idx = dist ( rng ) ;
2023-04-30 17:51:57 +02:00
llama_token result = candidates - > data [ idx ] . id ;
ctx - > t_sample_us + = ggml_time_us ( ) - t_start_sample_us ;
ctx - > n_sample + + ;
return result ;
2023-03-27 20:00:32 +02:00
}
//
// quantization
//
2023-04-30 17:51:57 +02:00
static void llama_model_quantize_internal ( const std : : string & fname_inp , const std : : string & fname_out , enum llama_ftype ftype , int nthread ) {
2023-04-10 21:59:13 +02:00
ggml_type quantized_type ;
2023-04-30 17:51:57 +02:00
switch ( ftype ) {
case LLAMA_FTYPE_MOSTLY_Q4_0 : quantized_type = GGML_TYPE_Q4_0 ; break ;
case LLAMA_FTYPE_MOSTLY_Q4_1 : quantized_type = GGML_TYPE_Q4_1 ; break ;
case LLAMA_FTYPE_MOSTLY_Q5_0 : quantized_type = GGML_TYPE_Q5_0 ; break ;
case LLAMA_FTYPE_MOSTLY_Q5_1 : quantized_type = GGML_TYPE_Q5_1 ; break ;
case LLAMA_FTYPE_MOSTLY_Q8_0 : quantized_type = GGML_TYPE_Q8_0 ; break ;
default : throw format ( " invalid output file type %d \n " , ftype ) ;
2023-03-27 20:00:32 +02:00
} ;
2023-04-30 17:51:57 +02:00
if ( nthread < = 0 ) {
nthread = std : : thread : : hardware_concurrency ( ) ;
}
2023-05-14 17:46:19 +02:00
std : : unique_ptr < llama_model_loader > model_loader ( new llama_model_loader ( fname_inp , /*use_mmap*/ false ,
2023-04-10 21:59:13 +02:00
/*vocab_only*/ false ) ) ;
2023-04-30 17:51:57 +02:00
llama_file_saver file_saver ( fname_out . c_str ( ) , model_loader - > file_loaders . at ( 0 ) . get ( ) , ftype ) ;
2023-04-10 21:59:13 +02:00
size_t total_size_org = 0 ;
size_t total_size_new = 0 ;
std : : vector < int64_t > hist_all ( 1 < < 4 , 0 ) ;
2023-04-30 17:51:57 +02:00
std : : vector < std : : thread > workers ;
std : : mutex mutex ;
2023-04-10 21:59:13 +02:00
size_t idx = 0 ;
for ( llama_load_tensor & tensor : model_loader - > tensors_map . tensors ) {
llama_buffer read_data ;
read_data . resize ( tensor . size ) ;
tensor . data = read_data . addr ;
model_loader - > load_data_for ( tensor ) ;
2023-04-30 17:51:57 +02:00
printf ( " [%4zu/%4zu] %36s - %16s, type = %6s, " ,
2023-04-10 21:59:13 +02:00
+ + idx , model_loader - > tensors_map . tensors . size ( ) ,
tensor . name . c_str ( ) , llama_format_tensor_shape ( tensor . ne ) . c_str ( ) ,
2023-04-30 17:51:57 +02:00
ggml_type_name ( tensor . type ) ) ;
2023-04-10 21:59:13 +02:00
// This used to be a regex, but <regex> has an extreme cost to compile times.
bool quantize = tensor . name . rfind ( " weight " ) = = tensor . name . size ( ) - 6 ; // ends with 'weight'?
// quantize only 2D tensors
quantize & = ( tensor . ne . size ( ) = = 2 ) ;
2023-04-30 17:51:57 +02:00
// uncomment this to keep the output layer in FP16
//if (tensor.name == "output.weight") {
// quantize = false;
//}
2023-04-10 21:59:13 +02:00
enum ggml_type new_type ;
void * new_data ;
size_t new_size ;
llama_buffer work ;
if ( ! quantize ) {
new_type = tensor . type ;
new_data = tensor . data ;
new_size = tensor . size ;
printf ( " size = %8.3f MB \n " , tensor . size / 1024.0 / 1024.0 ) ;
} else {
new_type = quantized_type ;
float * f32_data ;
size_t nelements = tensor . ne . at ( 0 ) * tensor . ne . at ( 1 ) ;
llama_buffer f32_conv_buf ;
if ( tensor . type = = GGML_TYPE_F32 ) {
f32_data = ( float * ) tensor . data ;
} else if ( tensor . type = = GGML_TYPE_F16 ) {
f32_conv_buf . resize ( nelements * sizeof ( float ) ) ;
f32_data = ( float * ) f32_conv_buf . addr ;
2023-05-14 17:46:19 +02:00
const auto * f16_data = ( const ggml_fp16_t * ) tensor . data ;
2023-04-10 21:59:13 +02:00
for ( size_t i = 0 ; i < nelements ; i + + ) {
f32_data [ i ] = ggml_fp16_to_fp32 ( f16_data [ i ] ) ;
2023-03-27 20:00:32 +02:00
}
} else {
2023-04-30 17:51:57 +02:00
throw format ( " type %s unsupported for integer quantization " , ggml_type_name ( tensor . type ) ) ;
2023-03-27 20:00:32 +02:00
}
2023-04-10 21:59:13 +02:00
printf ( " quantizing .. " ) ;
fflush ( stdout ) ;
work . resize ( nelements * 4 ) ; // upper bound on size
new_data = work . addr ;
std : : vector < int64_t > hist_cur ( 1 < < 4 , 0 ) ;
2023-04-30 17:51:57 +02:00
int chunk_size = 32 * 512 ;
const int nchunk = ( nelements + chunk_size - 1 ) / chunk_size ;
const int nthread_use = nthread > 1 ? std : : max ( 1 , std : : min ( nthread , nchunk ) ) : 1 ;
if ( nthread_use < 2 ) {
new_size = ggml_quantize_chunk ( new_type , f32_data , new_data , 0 , nelements , hist_cur . data ( ) ) ;
} else {
size_t counter = 0 ;
new_size = 0 ;
auto compute = [ & mutex , & counter , & hist_cur , & new_size , new_type , f32_data , new_data , nelements , chunk_size ] ( ) {
std : : vector < int64_t > local_hist ;
size_t local_size = 0 ;
while ( true ) {
std : : unique_lock < std : : mutex > lock ( mutex ) ;
size_t first = counter ; counter + = chunk_size ;
if ( first > = nelements ) {
if ( ! local_hist . empty ( ) ) {
2023-05-14 17:46:19 +02:00
for ( int j = 0 ; j < int ( local_hist . size ( ) ) ; + + j ) {
hist_cur [ j ] + = local_hist [ j ] ;
}
2023-04-30 17:51:57 +02:00
new_size + = local_size ;
}
break ;
}
lock . unlock ( ) ;
size_t last = std : : min ( nelements , first + chunk_size ) ;
2023-05-14 17:46:19 +02:00
if ( local_hist . empty ( ) ) {
local_hist . resize ( hist_cur . size ( ) , 0 ) ;
}
2023-04-30 17:51:57 +02:00
local_size + = ggml_quantize_chunk ( new_type , f32_data , new_data , first , last - first , local_hist . data ( ) ) ;
}
} ;
2023-05-14 17:46:19 +02:00
if ( ( int ) workers . size ( ) < nthread_use - 1 ) {
workers . resize ( nthread_use - 1 ) ;
}
for ( int it = 0 ; it < nthread_use - 1 ; + + it ) {
workers [ it ] = std : : thread ( compute ) ;
}
2023-04-30 17:51:57 +02:00
compute ( ) ;
2023-05-14 17:46:19 +02:00
for ( int it = 0 ; it < nthread_use - 1 ; + + it ) {
workers [ it ] . join ( ) ;
}
2023-03-27 20:00:32 +02:00
}
2023-04-10 21:59:13 +02:00
printf ( " size = %8.2f MB -> %8.2f MB | hist: " , tensor . size / 1024.0 / 1024.0 , new_size / 1024.0 / 1024.0 ) ;
for ( size_t i = 0 ; i < hist_cur . size ( ) ; i + + ) {
hist_all [ i ] + = hist_cur [ i ] ;
2023-03-27 20:00:32 +02:00
}
2023-04-10 21:59:13 +02:00
for ( size_t i = 0 ; i < hist_cur . size ( ) ; i + + ) {
printf ( " %5.3f " , hist_cur [ i ] / float ( nelements ) ) ;
2023-03-27 20:00:32 +02:00
}
printf ( " \n " ) ;
}
2023-04-10 21:59:13 +02:00
total_size_org + = tensor . size ;
total_size_new + = new_size ;
file_saver . write_tensor ( tensor , new_type , new_data , new_size ) ;
2023-03-27 20:00:32 +02:00
}
2023-04-10 21:59:13 +02:00
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 ) ;
2023-03-27 20:00:32 +02:00
2023-04-10 21:59:13 +02:00
{
int64_t sum_all = 0 ;
for ( size_t i = 0 ; i < hist_all . size ( ) ; i + + ) {
sum_all + = hist_all [ i ] ;
}
printf ( " %s: hist: " , __func__ ) ;
for ( size_t i = 0 ; i < hist_all . size ( ) ; i + + ) {
printf ( " %5.3f " , hist_all [ i ] / float ( sum_all ) ) ;
}
printf ( " \n " ) ;
}
2023-03-27 20:00:32 +02:00
}
//
// 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 ;
2023-05-14 17:46:19 +02:00
if ( params . seed < 0 ) {
2023-03-27 20:00:32 +02:00
params . seed = time ( NULL ) ;
}
2023-04-10 21:59:13 +02:00
unsigned cur_percentage = 0 ;
if ( params . progress_callback = = NULL ) {
params . progress_callback_user_data = & cur_percentage ;
params . progress_callback = [ ] ( float progress , void * ctx ) {
unsigned * cur_percentage_p = ( unsigned * ) ctx ;
unsigned percentage = ( unsigned ) ( 100 * progress ) ;
while ( percentage > * cur_percentage_p ) {
2023-05-23 13:04:39 +02:00
* cur_percentage_p = percentage ;
2023-04-10 21:59:13 +02:00
fprintf ( stderr , " . " ) ;
fflush ( stderr ) ;
if ( percentage > = 100 ) {
fprintf ( stderr , " \n " ) ;
}
}
} ;
}
2023-03-27 20:00:32 +02:00
ctx - > rng = std : : mt19937 ( params . seed ) ;
ctx - > logits_all = params . logits_all ;
ggml_type memory_type = params . f16_kv ? GGML_TYPE_F16 : GGML_TYPE_F32 ;
2023-05-14 17:46:19 +02:00
if ( ! llama_model_load ( path_model , * ctx , params . n_ctx , params . n_gpu_layers , memory_type ,
2023-04-10 21:59:13 +02:00
params . use_mmap , params . use_mlock , params . vocab_only ,
params . progress_callback , params . progress_callback_user_data ) ) {
2023-03-27 20:00:32 +02:00
fprintf ( stderr , " %s: failed to load model \n " , __func__ ) ;
llama_free ( ctx ) ;
return nullptr ;
}
// reserve memory for context buffers
2023-04-10 21:59:13 +02:00
if ( ! params . vocab_only ) {
2023-03-27 20:00:32 +02:00
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 ) ;
}
const auto & hparams = ctx - > model . hparams ;
// resized during inference
if ( params . logits_all ) {
ctx - > logits . reserve ( hparams . n_ctx * hparams . n_vocab ) ;
} else {
2023-04-30 17:51:57 +02:00
ctx - > logits . reserve ( hparams . n_vocab ) ;
2023-03-27 20:00:32 +02:00
}
if ( params . embedding ) {
ctx - > embedding . resize ( hparams . n_embd ) ;
}
2023-04-30 17:51:57 +02:00
ctx - > buf_compute . resize ( MEM_REQ_EVAL ( ) . at ( ctx - > model . type ) ) ;
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
ctx - > buf_scratch [ 0 ] . resize ( MEM_REQ_SCRATCH0 ( ) . at ( ctx - > model . type ) ) ;
ctx - > buf_scratch [ 1 ] . resize ( MEM_REQ_SCRATCH1 ( ) . at ( ctx - > model . type ) ) ;
2023-03-27 20:00:32 +02:00
}
return ctx ;
}
void llama_free ( struct llama_context * ctx ) {
delete ctx ;
}
int llama_model_quantize (
const char * fname_inp ,
const char * fname_out ,
2023-04-30 17:51:57 +02:00
enum llama_ftype ftype ,
int nthread ) {
2023-04-10 21:59:13 +02:00
try {
2023-04-30 17:51:57 +02:00
llama_model_quantize_internal ( fname_inp , fname_out , ftype , nthread ) ;
2023-04-10 21:59:13 +02:00
return 0 ;
} catch ( const std : : string & err ) {
fprintf ( stderr , " %s: failed to quantize: %s \n " , __func__ , err . c_str ( ) ) ;
2023-03-27 20:00:32 +02:00
return 1 ;
}
2023-04-10 21:59:13 +02:00
}
2023-03-27 20:00:32 +02:00
2023-04-30 17:51:57 +02:00
int llama_apply_lora_from_file_internal ( struct llama_context * ctx , const char * path_lora , const char * path_base_model , int n_threads ) {
fprintf ( stderr , " %s: applying lora adapter from '%s' - please wait ... \n " , __func__ , path_lora ) ;
auto & model = ctx - > model ;
const int64_t t_start_lora_us = ggml_time_us ( ) ;
auto fin = std : : ifstream ( path_lora , std : : ios : : binary ) ;
if ( ! fin ) {
fprintf ( stderr , " %s: failed to open '%s' \n " , __func__ , path_lora ) ;
return 1 ;
}
// verify magic and version
{
uint32_t magic ;
fin . read ( ( char * ) & magic , sizeof ( magic ) ) ;
2023-05-23 13:04:39 +02:00
if ( magic ! = LLAMA_FILE_MAGIC_GGLA ) {
2023-04-30 17:51:57 +02:00
fprintf ( stderr , " %s: bad file magic \n " , __func__ ) ;
return 1 ;
}
uint32_t format_version ;
fin . read ( ( char * ) & format_version , sizeof ( format_version ) ) ;
if ( format_version ! = 1 ) {
fprintf ( stderr , " %s: unsupported file version \n " , __func__ ) ;
return 1 ;
}
}
int32_t lora_r ;
int32_t lora_alpha ;
fin . read ( ( char * ) & lora_r , sizeof ( lora_r ) ) ;
fin . read ( ( char * ) & lora_alpha , sizeof ( lora_alpha ) ) ;
float scaling = ( float ) lora_alpha / ( float ) lora_r ;
fprintf ( stderr , " %s: r = %d, alpha = %d, scaling = %.2f \n " , __func__ , lora_r , lora_alpha , scaling ) ;
// create a temporary ggml context to store the lora tensors
// todo: calculate size from biggest possible tensor
std : : vector < uint8_t > lora_buf ( 1024ull * 1024ull * 1024ull ) ;
struct ggml_init_params params ;
params . mem_size = lora_buf . size ( ) ;
params . mem_buffer = lora_buf . data ( ) ;
params . no_alloc = false ;
ggml_context * lora_ctx = ggml_init ( params ) ;
std : : unordered_map < std : : string , struct ggml_tensor * > lora_tensors ;
// create a name -> tensor map of the model to accelerate lookups
std : : unordered_map < std : : string , struct ggml_tensor * > model_tensors ;
for ( auto & kv : model . tensors_by_name ) {
model_tensors . insert ( kv ) ;
}
// load base model
std : : unique_ptr < llama_model_loader > model_loader ;
ggml_context * base_ctx = NULL ;
llama_buffer base_buf ;
if ( path_base_model ) {
fprintf ( stderr , " %s: loading base model from '%s' \n " , __func__ , path_base_model ) ;
model_loader . reset ( new llama_model_loader ( path_base_model , /*use_mmap*/ true , /*vocab_only*/ false ) ) ;
2023-05-14 17:46:19 +02:00
size_t ctx_size ;
size_t mmapped_size ;
2023-04-30 17:51:57 +02:00
model_loader - > calc_sizes ( & ctx_size , & mmapped_size ) ;
base_buf . resize ( ctx_size ) ;
ggml_init_params base_params ;
base_params . mem_size = base_buf . size ;
base_params . mem_buffer = base_buf . addr ;
base_params . no_alloc = model_loader - > use_mmap ;
base_ctx = ggml_init ( base_params ) ;
model_loader - > ggml_ctx = base_ctx ;
// maybe this should in llama_model_loader
if ( model_loader - > use_mmap ) {
2023-05-23 13:04:39 +02:00
model_loader - > mapping . reset ( new llama_mmap ( & model_loader - > file_loaders . at ( 0 ) - > file , /* prefetch */ 0 ) ) ;
2023-04-30 17:51:57 +02:00
}
}
// read tensors and apply
bool warned = false ;
int n_tensors = 0 ;
while ( true ) {
int32_t n_dims ;
int32_t length ;
int32_t ftype ;
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 ) ) ;
if ( fin . eof ( ) ) {
break ;
}
int32_t ne [ 2 ] = { 1 , 1 } ;
for ( int i = 0 ; i < n_dims ; + + i ) {
fin . read ( reinterpret_cast < char * > ( & ne [ i ] ) , sizeof ( ne [ i ] ) ) ;
}
2023-05-14 17:46:19 +02:00
std : : string name ;
{
char buf [ 1024 ] ;
fin . read ( buf , length ) ;
name = std : : string ( buf , length ) ;
}
2023-04-30 17:51:57 +02:00
// check for lora suffix and get the type of tensor
const std : : string lora_suffix = " .lora " ;
size_t pos = name . rfind ( lora_suffix ) ;
if ( pos = = std : : string : : npos ) {
fprintf ( stderr , " %s: error: '%s' is not a lora tensor \n " , __func__ , name . c_str ( ) ) ;
return 1 ;
}
std : : string lora_type = name . substr ( pos + lora_suffix . length ( ) ) ;
std : : string base_name = name ;
base_name . erase ( pos ) ;
// fprintf(stderr, "%s: %s => %s (lora type %s) ", __func__, name.c_str(),base_name.c_str(), lora_type.c_str());
2023-05-14 17:46:19 +02:00
if ( model_tensors . find ( base_name ) = = model_tensors . end ( ) ) {
2023-04-30 17:51:57 +02:00
fprintf ( stderr , " %s: unknown tensor '%s' in lora adapter \n " , __func__ , name . data ( ) ) ;
return 1 ;
}
// create ggml tensor
ggml_type wtype ;
switch ( ftype ) {
case 0 : wtype = GGML_TYPE_F32 ; break ;
case 1 : wtype = GGML_TYPE_F16 ; break ;
default :
{
fprintf ( stderr , " %s: invalid tensor data type '%d' \n " ,
__func__ , ftype ) ;
return false ;
}
}
ggml_tensor * lora_tensor ;
if ( n_dims = = 2 ) {
lora_tensor = ggml_new_tensor_2d ( lora_ctx , wtype , ne [ 0 ] , ne [ 1 ] ) ;
}
else {
fprintf ( stderr , " %s: unsupported tensor dimension %d \n " , __func__ , n_dims ) ;
return 1 ;
}
// load tensor data
size_t offset = fin . tellg ( ) ;
size_t tensor_data_size = ggml_nbytes ( lora_tensor ) ;
offset = ( offset + 31 ) & - 32 ;
fin . seekg ( offset ) ;
fin . read ( ( char * ) lora_tensor - > data , tensor_data_size ) ;
lora_tensors [ name ] = lora_tensor ;
// check if we have both A and B tensors and apply
if ( lora_tensors . find ( base_name + " .loraA " ) ! = lora_tensors . end ( ) & &
lora_tensors . find ( base_name + " .loraB " ) ! = lora_tensors . end ( ) ) {
ggml_tensor * dest_t = model_tensors [ base_name ] ;
ggml_tensor * base_t ;
if ( model_loader ) {
// load from base model
if ( model_loader - > tensors_map . name_to_idx . find ( base_name ) = = model_loader - > tensors_map . name_to_idx . end ( ) ) {
fprintf ( stderr , " %s: error: tensor '%s' not found in base model \n " , __func__ , base_name . c_str ( ) ) ;
return 1 ;
}
size_t idx = model_loader - > tensors_map . name_to_idx [ base_name ] ;
llama_load_tensor & lt = model_loader - > tensors_map . tensors [ idx ] ;
2023-05-23 13:04:39 +02:00
base_t = model_loader - > get_tensor ( base_name , { ( uint32_t ) dest_t - > ne [ 0 ] , ( uint32_t ) dest_t - > ne [ 1 ] } , GGML_BACKEND_CPU ) ;
2023-04-30 17:51:57 +02:00
lt . data = ( uint8_t * ) lt . ggml_tensor - > data ;
model_loader - > load_data_for ( lt ) ;
lt . ggml_tensor - > data = lt . data ;
}
else {
base_t = dest_t ;
}
if ( ggml_is_quantized ( base_t - > type ) ) {
if ( ! warned ) {
fprintf ( stderr , " %s: warning: using a lora adapter with a quantized model may result in poor quality, "
" use a f16 or f32 base model with --lora-base \n " , __func__ ) ;
warned = true ;
}
}
ggml_tensor * loraA = lora_tensors [ base_name + " .loraA " ] ;
ggml_tensor * loraB = lora_tensors [ base_name + " .loraB " ] ;
if ( base_t - > ne [ 0 ] ! = loraA - > ne [ 1 ] | | base_t - > ne [ 1 ] ! = loraB - > ne [ 1 ] ) {
fprintf ( stderr , " %s: incompatible tensor dimensions (% " PRId64 " and % " PRId64 " ); "
" are you sure that this adapter is for this model? \n " , __func__ , base_t - > ne [ 0 ] , loraA - > ne [ 1 ] ) ;
return 1 ;
}
// w = w + BA*s
ggml_tensor * BA = ggml_mul_mat ( lora_ctx , loraA , loraB ) ;
if ( scaling ! = 1.0f ) {
ggml_tensor * scale_tensor = ggml_new_f32 ( lora_ctx , scaling ) ;
2023-05-14 17:46:19 +02:00
BA = ggml_scale_inplace ( lora_ctx , BA , scale_tensor ) ;
2023-04-30 17:51:57 +02:00
}
ggml_tensor * r ;
if ( base_t = = dest_t ) {
r = ggml_add_inplace ( lora_ctx , dest_t , BA ) ;
}
else {
r = ggml_add ( lora_ctx , base_t , BA ) ;
r = ggml_cpy ( lora_ctx , r , dest_t ) ;
}
struct ggml_cgraph gf = ggml_build_forward ( r ) ;
2023-09-07 11:33:12 +02:00
ggml_graph_compute_with_ctx ( lora_ctx , & gf , n_threads ) ;
2023-04-30 17:51:57 +02:00
// we won't need these tensors again, reset the context to save memory
ggml_free ( lora_ctx ) ;
lora_ctx = ggml_init ( params ) ;
lora_tensors . clear ( ) ;
n_tensors + + ;
2023-05-14 17:46:19 +02:00
if ( n_tensors % 4 = = 0 ) {
2023-04-30 17:51:57 +02:00
fprintf ( stderr , " . " ) ;
2023-05-14 17:46:19 +02:00
}
2023-04-30 17:51:57 +02:00
}
}
// TODO: this should be in a destructor, it will leak on failure
ggml_free ( lora_ctx ) ;
if ( base_ctx ) {
ggml_free ( base_ctx ) ;
}
const int64_t t_lora_us = ggml_time_us ( ) - t_start_lora_us ;
fprintf ( stderr , " done (%.2f ms) \n " , t_lora_us / 1000.0 ) ;
return 0 ;
2023-04-10 21:59:13 +02:00
}
2023-04-30 17:51:57 +02:00
int llama_apply_lora_from_file ( struct llama_context * ctx , const char * path_lora , const char * path_base_model , int n_threads ) {
try {
return llama_apply_lora_from_file_internal ( ctx , path_lora , path_base_model , n_threads ) ;
} catch ( const std : : string & err ) {
fprintf ( stderr , " %s: failed to apply lora adapter: %s \n " , __func__ , err . c_str ( ) ) ;
return 1 ;
}
2023-04-10 21:59:13 +02:00
}
2023-05-14 17:46:19 +02:00
int llama_get_kv_cache_token_count ( const struct llama_context * ctx ) {
2023-04-10 21:59:13 +02:00
return ctx - > model . kv_self . n ;
}
2023-05-14 17:46:19 +02:00
# define LLAMA_MAX_RNG_STATE (64*1024)
2023-04-30 17:51:57 +02:00
void llama_set_rng_seed ( struct llama_context * ctx , int seed ) {
2023-05-14 17:46:19 +02:00
if ( seed < 0 ) {
2023-04-30 17:51:57 +02:00
seed = time ( NULL ) ;
}
ctx - > rng . seed ( seed ) ;
}
2023-05-08 19:59:21 +02:00
// Returns the *maximum* size of the state
2023-05-14 17:46:19 +02:00
size_t llama_get_state_size ( const struct llama_context * ctx ) {
2023-04-30 17:51:57 +02:00
// we don't know size of rng until we actually serialize it. so reserve more than enough memory for its serialized state.
// for reference, std::mt19937(1337) serializes to 6701 bytes.
const size_t s_rng_size = sizeof ( size_t ) ;
const size_t s_rng = LLAMA_MAX_RNG_STATE ;
const size_t s_logits_capacity = sizeof ( size_t ) ;
const size_t s_logits_size = sizeof ( size_t ) ;
const size_t s_logits = ctx - > logits . capacity ( ) * sizeof ( float ) ;
const size_t s_embedding_size = sizeof ( size_t ) ;
const size_t s_embedding = ctx - > embedding . size ( ) * sizeof ( float ) ;
const size_t s_kv_size = sizeof ( size_t ) ;
const size_t s_kv_ntok = sizeof ( int ) ;
const size_t s_kv = ctx - > model . kv_self . buf . size ;
const size_t s_total = (
+ s_rng_size
+ s_rng
+ s_logits_capacity
+ s_logits_size
+ s_logits
+ s_embedding_size
+ s_embedding
+ s_kv_size
+ s_kv_ntok
+ s_kv
) ;
return s_total ;
}
// Copies the state to the specified destination address
2023-05-14 17:46:19 +02:00
size_t llama_copy_state_data ( struct llama_context * ctx , uint8_t * dst ) {
uint8_t * out = dst ;
2023-04-30 17:51:57 +02:00
// copy rng
{
std : : stringstream rng_ss ;
rng_ss < < ctx - > rng ;
const size_t rng_size = rng_ss . str ( ) . size ( ) ;
char rng_buf [ LLAMA_MAX_RNG_STATE ] ;
memset ( & rng_buf [ 0 ] , 0 , LLAMA_MAX_RNG_STATE ) ;
memcpy ( & rng_buf [ 0 ] , rng_ss . str ( ) . data ( ) , rng_ss . str ( ) . size ( ) ) ;
memcpy ( out , & rng_size , sizeof ( rng_size ) ) ; out + = sizeof ( rng_size ) ;
memcpy ( out , & rng_buf [ 0 ] , LLAMA_MAX_RNG_STATE ) ; out + = LLAMA_MAX_RNG_STATE ;
}
// copy logits
{
const size_t logits_cap = ctx - > logits . capacity ( ) ;
const size_t logits_size = ctx - > logits . size ( ) ;
memcpy ( out , & logits_cap , sizeof ( logits_cap ) ) ; out + = sizeof ( logits_cap ) ;
memcpy ( out , & logits_size , sizeof ( logits_size ) ) ; out + = sizeof ( logits_size ) ;
if ( logits_size ) {
memcpy ( out , ctx - > logits . data ( ) , logits_size * sizeof ( float ) ) ;
}
out + = logits_cap * sizeof ( float ) ;
}
// copy embeddings
{
const size_t embedding_size = ctx - > embedding . size ( ) ;
memcpy ( out , & embedding_size , sizeof ( embedding_size ) ) ; out + = sizeof ( embedding_size ) ;
if ( embedding_size ) {
memcpy ( out , ctx - > embedding . data ( ) , embedding_size * sizeof ( float ) ) ;
out + = embedding_size * sizeof ( float ) ;
}
}
// copy kv cache
{
2023-05-08 19:59:21 +02:00
const auto & kv_self = ctx - > model . kv_self ;
const auto & hparams = ctx - > model . hparams ;
const int n_layer = hparams . n_layer ;
const int n_embd = hparams . n_embd ;
const int n_ctx = hparams . n_ctx ;
const size_t kv_size = kv_self . buf . size ;
2023-04-30 17:51:57 +02:00
const int kv_ntok = llama_get_kv_cache_token_count ( ctx ) ;
memcpy ( out , & kv_size , sizeof ( kv_size ) ) ; out + = sizeof ( kv_size ) ;
memcpy ( out , & kv_ntok , sizeof ( kv_ntok ) ) ; out + = sizeof ( kv_ntok ) ;
if ( kv_size ) {
2023-05-08 19:59:21 +02:00
const size_t elt_size = ggml_element_size ( kv_self . k ) ;
2023-05-14 17:46:19 +02:00
2023-05-08 19:59:21 +02:00
char buffer [ 4096 ] ;
2023-05-14 17:46:19 +02:00
2023-05-08 19:59:21 +02:00
ggml_context * cpy_ctx = ggml_init ( { sizeof ( buffer ) , buffer , /* no_alloc */ true } ) ;
ggml_cgraph gf { } ;
ggml_tensor * kout3d = ggml_new_tensor_3d ( cpy_ctx , kv_self . k - > type , n_embd , kv_ntok , n_layer ) ;
kout3d - > data = out ;
out + = ggml_nbytes ( kout3d ) ;
ggml_tensor * vout3d = ggml_new_tensor_3d ( cpy_ctx , kv_self . v - > type , kv_ntok , n_embd , n_layer ) ;
vout3d - > data = out ;
out + = ggml_nbytes ( vout3d ) ;
ggml_tensor * k3d = ggml_view_3d ( cpy_ctx , kv_self . k ,
n_embd , kv_ntok , n_layer ,
elt_size * n_embd , elt_size * n_embd * n_ctx , 0 ) ;
ggml_tensor * v3d = ggml_view_3d ( cpy_ctx , kv_self . v ,
kv_ntok , n_embd , n_layer ,
elt_size * n_ctx , elt_size * n_ctx * n_embd , 0 ) ;
ggml_build_forward_expand ( & gf , ggml_cpy ( cpy_ctx , k3d , kout3d ) ) ;
ggml_build_forward_expand ( & gf , ggml_cpy ( cpy_ctx , v3d , vout3d ) ) ;
2023-09-07 11:33:12 +02:00
ggml_graph_compute_with_ctx ( cpy_ctx , & gf , 1 ) ;
2023-05-14 17:46:19 +02:00
ggml_free ( cpy_ctx ) ;
2023-04-30 17:51:57 +02:00
}
}
2023-05-14 17:46:19 +02:00
const size_t written = out - dst ;
2023-05-08 19:59:21 +02:00
const size_t max_size = llama_get_state_size ( ctx ) ;
2023-04-30 17:51:57 +02:00
2023-05-08 19:59:21 +02:00
LLAMA_ASSERT ( written < = max_size ) ;
2023-04-30 17:51:57 +02:00
return written ;
}
// Sets the state reading from the specified source address
2023-05-23 13:04:39 +02:00
size_t llama_set_state_data ( struct llama_context * ctx , uint8_t * src ) {
uint8_t * inp = src ;
2023-04-30 17:51:57 +02:00
// set rng
{
size_t rng_size ;
char rng_buf [ LLAMA_MAX_RNG_STATE ] ;
2023-05-14 17:46:19 +02:00
memcpy ( & rng_size , inp , sizeof ( rng_size ) ) ; inp + = sizeof ( rng_size ) ;
memcpy ( & rng_buf [ 0 ] , inp , LLAMA_MAX_RNG_STATE ) ; inp + = LLAMA_MAX_RNG_STATE ;
2023-04-30 17:51:57 +02:00
std : : stringstream rng_ss ;
rng_ss . str ( std : : string ( & rng_buf [ 0 ] , rng_size ) ) ;
rng_ss > > ctx - > rng ;
LLAMA_ASSERT ( rng_ss . fail ( ) = = false ) ;
}
// set logits
{
size_t logits_cap ;
size_t logits_size ;
2023-05-14 17:46:19 +02:00
memcpy ( & logits_cap , inp , sizeof ( logits_cap ) ) ; inp + = sizeof ( logits_cap ) ;
memcpy ( & logits_size , inp , sizeof ( logits_size ) ) ; inp + = sizeof ( logits_size ) ;
2023-04-30 17:51:57 +02:00
LLAMA_ASSERT ( ctx - > logits . capacity ( ) = = logits_cap ) ;
if ( logits_size ) {
ctx - > logits . resize ( logits_size ) ;
2023-05-14 17:46:19 +02:00
memcpy ( ctx - > logits . data ( ) , inp , logits_size * sizeof ( float ) ) ;
2023-04-30 17:51:57 +02:00
}
2023-05-14 17:46:19 +02:00
inp + = logits_cap * sizeof ( float ) ;
2023-04-30 17:51:57 +02:00
}
// set embeddings
{
size_t embedding_size ;
2023-05-14 17:46:19 +02:00
memcpy ( & embedding_size , inp , sizeof ( embedding_size ) ) ; inp + = sizeof ( embedding_size ) ;
2023-04-30 17:51:57 +02:00
LLAMA_ASSERT ( ctx - > embedding . capacity ( ) = = embedding_size ) ;
if ( embedding_size ) {
2023-05-14 17:46:19 +02:00
memcpy ( ctx - > embedding . data ( ) , inp , embedding_size * sizeof ( float ) ) ;
inp + = embedding_size * sizeof ( float ) ;
2023-04-30 17:51:57 +02:00
}
}
// set kv cache
{
2023-05-08 19:59:21 +02:00
const auto & kv_self = ctx - > model . kv_self ;
const auto & hparams = ctx - > model . hparams ;
const int n_layer = hparams . n_layer ;
const int n_embd = hparams . n_embd ;
const int n_ctx = hparams . n_ctx ;
2023-04-30 17:51:57 +02:00
size_t kv_size ;
int kv_ntok ;
2023-05-14 17:46:19 +02:00
memcpy ( & kv_size , inp , sizeof ( kv_size ) ) ; inp + = sizeof ( kv_size ) ;
memcpy ( & kv_ntok , inp , sizeof ( kv_ntok ) ) ; inp + = sizeof ( kv_ntok ) ;
2023-04-30 17:51:57 +02:00
if ( kv_size ) {
2023-05-08 19:59:21 +02:00
LLAMA_ASSERT ( kv_self . buf . size = = kv_size ) ;
const size_t elt_size = ggml_element_size ( kv_self . k ) ;
2023-05-14 17:46:19 +02:00
2023-05-08 19:59:21 +02:00
char buffer [ 4096 ] ;
2023-05-14 17:46:19 +02:00
2023-05-08 19:59:21 +02:00
ggml_context * cpy_ctx = ggml_init ( { sizeof ( buffer ) , buffer , /* no_alloc */ true } ) ;
ggml_cgraph gf { } ;
ggml_tensor * kin3d = ggml_new_tensor_3d ( cpy_ctx , kv_self . k - > type , n_embd , kv_ntok , n_layer ) ;
2023-05-14 17:46:19 +02:00
kin3d - > data = ( void * ) inp ;
inp + = ggml_nbytes ( kin3d ) ;
2023-05-08 19:59:21 +02:00
ggml_tensor * vin3d = ggml_new_tensor_3d ( cpy_ctx , kv_self . v - > type , kv_ntok , n_embd , n_layer ) ;
2023-05-14 17:46:19 +02:00
vin3d - > data = ( void * ) inp ;
inp + = ggml_nbytes ( vin3d ) ;
2023-04-30 17:51:57 +02:00
2023-05-08 19:59:21 +02:00
ggml_tensor * k3d = ggml_view_3d ( cpy_ctx , kv_self . k ,
n_embd , kv_ntok , n_layer ,
elt_size * n_embd , elt_size * n_embd * n_ctx , 0 ) ;
2023-04-30 17:51:57 +02:00
2023-05-08 19:59:21 +02:00
ggml_tensor * v3d = ggml_view_3d ( cpy_ctx , kv_self . v ,
kv_ntok , n_embd , n_layer ,
elt_size * n_ctx , elt_size * n_ctx * n_embd , 0 ) ;
2023-04-30 17:51:57 +02:00
2023-05-08 19:59:21 +02:00
ggml_build_forward_expand ( & gf , ggml_cpy ( cpy_ctx , kin3d , k3d ) ) ;
ggml_build_forward_expand ( & gf , ggml_cpy ( cpy_ctx , vin3d , v3d ) ) ;
2023-09-07 11:33:12 +02:00
ggml_graph_compute_with_ctx ( cpy_ctx , & gf , 1 ) ;
2023-04-30 17:51:57 +02:00
2023-05-14 17:46:19 +02:00
ggml_free ( cpy_ctx ) ;
2023-04-30 17:51:57 +02:00
}
ctx - > model . kv_self . n = kv_ntok ;
}
2023-05-14 17:46:19 +02:00
const size_t nread = inp - src ;
2023-05-08 19:59:21 +02:00
const size_t max_size = llama_get_state_size ( ctx ) ;
2023-04-30 17:51:57 +02:00
2023-05-08 19:59:21 +02:00
LLAMA_ASSERT ( nread < = max_size ) ;
2023-04-30 17:51:57 +02:00
return nread ;
2023-03-27 20:00:32 +02:00
}
2023-05-14 17:46:19 +02:00
bool llama_load_session_file ( struct llama_context * ctx , const char * path_session , llama_token * tokens_out , size_t n_token_capacity , size_t * n_token_count_out ) {
llama_file file ( path_session , " rb " ) ;
// sanity checks
{
const uint32_t magic = file . read_u32 ( ) ;
const uint32_t version = file . read_u32 ( ) ;
if ( magic ! = LLAMA_SESSION_MAGIC | | version ! = LLAMA_SESSION_VERSION ) {
fprintf ( stderr , " %s : unknown (magic, version) for session file: %08x, %08x \n " , __func__ , magic , version ) ;
return false ;
}
llama_hparams session_hparams ;
file . read_raw ( & session_hparams , sizeof ( llama_hparams ) ) ;
if ( session_hparams ! = ctx - > model . hparams ) {
fprintf ( stderr , " %s : model hparams didn't match from session file! \n " , __func__ ) ;
return false ;
}
}
// load the prompt
{
const uint32_t n_token_count = file . read_u32 ( ) ;
if ( n_token_count > n_token_capacity ) {
fprintf ( stderr , " %s : token count in session file exceeded capacity! %u > %zu \n " , __func__ , n_token_count , n_token_capacity ) ;
return false ;
}
file . read_raw ( tokens_out , sizeof ( llama_token ) * n_token_count ) ;
* n_token_count_out = n_token_count ;
}
// restore the context state
{
const size_t n_state_size_cur = file . size - file . tell ( ) ;
const size_t n_state_size_max = llama_get_state_size ( ctx ) ;
if ( n_state_size_cur > n_state_size_max ) {
fprintf ( stderr , " %s : the state size in session file is too big! max %zu, got %zu \n " , __func__ , n_state_size_max , n_state_size_cur ) ;
return false ;
}
std : : vector < uint8_t > state_data ( n_state_size_max ) ;
file . read_raw ( state_data . data ( ) , n_state_size_cur ) ;
llama_set_state_data ( ctx , state_data . data ( ) ) ;
}
return true ;
}
bool llama_save_session_file ( struct llama_context * ctx , const char * path_session , const llama_token * tokens , size_t n_token_count ) {
llama_file file ( path_session , " wb " ) ;
file . write_u32 ( LLAMA_SESSION_MAGIC ) ;
file . write_u32 ( LLAMA_SESSION_VERSION ) ;
file . write_raw ( & ctx - > model . hparams , sizeof ( llama_hparams ) ) ;
// save the prompt
file . write_u32 ( ( uint32_t ) n_token_count ) ;
file . write_raw ( tokens , sizeof ( llama_token ) * n_token_count ) ;
// save the context state
{
const size_t n_state_size_max = llama_get_state_size ( ctx ) ;
std : : vector < uint8_t > state_data ( n_state_size_max ) ;
const size_t n_state_size_cur = llama_copy_state_data ( ctx , state_data . data ( ) ) ;
file . write_raw ( state_data . data ( ) , n_state_size_cur ) ;
}
return true ;
}
2023-03-27 20:00:32 +02:00
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 ;
}
2023-05-14 17:46:19 +02:00
2023-04-10 21:59:13 +02:00
// get a more accurate load time, upon first eval
2023-05-14 17:46:19 +02:00
// TODO: fix this
2023-04-10 21:59:13 +02:00
if ( ! ctx - > has_evaluated_once ) {
ctx - > t_load_us = ggml_time_us ( ) - ctx - > t_start_us ;
ctx - > has_evaluated_once = true ;
}
2023-05-14 17:46:19 +02:00
2023-03-27 20:00:32 +02:00
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 ( ) ;
}
2023-05-14 17:46:19 +02:00
int llama_n_vocab ( const struct llama_context * ctx ) {
2023-03-27 20:00:32 +02:00
return ctx - > vocab . id_to_token . size ( ) ;
}
2023-05-14 17:46:19 +02:00
int llama_n_ctx ( const struct llama_context * ctx ) {
2023-03-27 20:00:32 +02:00
return ctx - > model . hparams . n_ctx ;
}
2023-05-14 17:46:19 +02:00
int llama_n_embd ( const struct llama_context * ctx ) {
2023-03-27 20:00:32 +02:00
return ctx - > model . hparams . n_embd ;
}
float * llama_get_logits ( struct llama_context * ctx ) {
return ctx - > logits . data ( ) ;
}
float * llama_get_embeddings ( struct llama_context * ctx ) {
return ctx - > embedding . data ( ) ;
}
2023-05-14 17:46:19 +02:00
const char * llama_token_to_str ( const struct llama_context * ctx , llama_token token ) {
2023-03-27 20:00:32 +02:00
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 ;
}
2023-04-30 17:51:57 +02:00
llama_token llama_token_nl ( ) {
return 13 ;
2023-03-27 20:00:32 +02:00
}
void llama_print_timings ( struct llama_context * ctx ) {
const int64_t t_end_us = ggml_time_us ( ) ;
const int32_t n_sample = std : : max ( 1 , ctx - > n_sample ) ;
const int32_t n_eval = std : : max ( 1 , ctx - > n_eval ) ;
const int32_t n_p_eval = std : : max ( 1 , ctx - > n_p_eval ) ;
fprintf ( stderr , " \n " ) ;
2023-04-10 21:59:13 +02:00
fprintf ( stderr , " %s: load time = %8.2f ms \n " , __func__ , ctx - > t_load_us / 1000.0 ) ;
2023-05-14 17:46:19 +02:00
fprintf ( stderr , " %s: sample time = %8.2f ms / %5d runs (%8.2f ms per token) \n " , __func__ , 1e-3 * ctx - > t_sample_us , n_sample , 1e-3 * ctx - > t_sample_us / n_sample ) ;
2023-04-10 21:59:13 +02:00
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 ) ;
2023-05-14 17:46:19 +02:00
fprintf ( stderr , " %s: eval time = %8.2f ms / %5d runs (%8.2f ms per token) \n " , __func__ , 1e-3 * ctx - > t_eval_us , n_eval , 1e-3 * ctx - > t_eval_us / n_eval ) ;
2023-04-10 21:59:13 +02:00
fprintf ( stderr , " %s: total time = %8.2f ms \n " , __func__ , ( t_end_us - ctx - > t_start_us ) / 1000.0 ) ;
2023-03-27 20:00:32 +02:00
}
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 ;
ctx - > t_p_eval_us = ctx - > n_p_eval = 0 ;
}
const char * llama_print_system_info ( void ) {
static std : : string s ;
s = " " ;
2023-04-30 17:51:57 +02:00
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 + = " AVX512_VBMI = " + std : : to_string ( ggml_cpu_has_avx512_vbmi ( ) ) + " | " ;
s + = " AVX512_VNNI = " + std : : to_string ( ggml_cpu_has_avx512_vnni ( ) ) + " | " ;
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 ( ) ) + " | " ;
2023-03-27 20:00:32 +02:00
return s . c_str ( ) ;
}
2023-04-10 21:59:13 +02:00
// For internal test use
std : : vector < std : : pair < std : : string , struct ggml_tensor * > > & llama_internal_get_tensor_map ( struct llama_context * ctx ) {
return ctx - > model . tensors_by_name ;
}