2023-03-25 20:26:40 +02:00
#include "common.h"
#include "llama.h"
2023-05-01 09:23:47 -07:00
#include "build-info.h"
2023-03-25 20:26:40 +02:00
2023-03-28 16:48:20 +00:00
#include <cmath>
2023-04-16 12:13:00 +02:00
#include <ctime>
2023-07-22 14:21:24 +02:00
#include <sstream>
2023-03-28 16:48:20 +00:00
2023-06-16 21:23:53 +03:00
#if defined(_MSC_VER)
#pragma warning(disable: 4244 4267) // possible loss of data
#endif
2023-03-28 16:48:20 +00:00
std :: vector < float > softmax ( const std :: vector < float >& logits ) {
std :: vector < float > probs ( logits . size ());
2023-03-25 20:26:40 +02:00
float max_logit = logits [ 0 ];
for ( float v : logits ) max_logit = std :: max ( max_logit , v );
double sum_exp = 0.0 ;
for ( size_t i = 0 ; i < logits . size (); i ++ ) {
// Subtract the maximum logit value from the current logit value for numerical stability
2023-03-28 16:48:20 +00:00
const float logit = logits [ i ] - max_logit ;
const float exp_logit = expf ( logit );
2023-03-25 20:26:40 +02:00
sum_exp += exp_logit ;
probs [ i ] = exp_logit ;
}
for ( size_t i = 0 ; i < probs . size (); i ++ ) probs [ i ] /= sum_exp ;
return probs ;
}
void perplexity ( llama_context * ctx , const gpt_params & params ) {
// Download: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip?ref=salesforce-research
2023-03-26 13:14:01 +00:00
// Run `./perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
2023-03-25 20:26:40 +02:00
// Output: `perplexity: 13.5106 [114/114]`
2023-05-08 17:41:54 +03:00
// BOS tokens will be added for each chunk before eval
2023-03-25 20:26:40 +02:00
auto tokens = :: llama_tokenize ( ctx , params . prompt , true );
2023-07-18 14:24:43 +03:00
const int n_chunk_max = tokens . size () / params . n_ctx ;
2023-05-08 17:41:54 +03:00
2023-07-18 14:24:43 +03:00
const int n_chunk = params . n_chunks < 0 ? n_chunk_max : std :: min ( params . n_chunks , n_chunk_max );
2023-05-08 17:41:54 +03:00
const int n_vocab = llama_n_vocab ( ctx );
const int n_batch = params . n_batch ;
2023-03-25 20:26:40 +02:00
2023-07-18 14:24:43 +03:00
int count = 0 ;
2023-03-28 16:48:20 +00:00
double nll = 0.0 ;
2023-07-18 14:24:43 +03:00
2023-05-08 17:41:54 +03:00
fprintf ( stderr , "%s: calculating perplexity over %d chunks, batch_size=%d \n " , __func__ , n_chunk , n_batch );
2023-03-25 20:26:40 +02:00
2023-05-08 17:41:54 +03:00
for ( int i = 0 ; i < n_chunk ; ++ i ) {
const int start = i * params . n_ctx ;
const int end = start + params . n_ctx ;
const int num_batches = ( params . n_ctx + n_batch - 1 ) / n_batch ;
2023-04-13 14:50:42 -07:00
std :: vector < float > logits ;
2023-05-08 17:41:54 +03:00
const auto t_start = std :: chrono :: high_resolution_clock :: now ();
2023-04-13 14:50:42 -07:00
for ( int j = 0 ; j < num_batches ; ++ j ) {
2023-05-08 17:41:54 +03:00
const int batch_start = start + j * n_batch ;
const int batch_size = std :: min ( end - batch_start , n_batch );
// save original token and restore it after eval
const auto token_org = tokens [ batch_start ];
// add BOS token for the first batch of each chunk
if ( j == 0 ) {
tokens [ batch_start ] = llama_token_bos ();
}
if ( llama_eval ( ctx , tokens . data () + batch_start , batch_size , j * n_batch , params . n_threads )) {
2023-04-13 14:50:42 -07:00
fprintf ( stderr , "%s : failed to eval \n " , __func__ );
return ;
}
2023-05-08 17:41:54 +03:00
// restore the original token in case it was set to BOS
tokens [ batch_start ] = token_org ;
const auto batch_logits = llama_get_logits ( ctx );
2023-04-13 14:50:42 -07:00
logits . insert ( logits . end (), batch_logits , batch_logits + batch_size * n_vocab );
2023-03-25 20:26:40 +02:00
}
2023-05-08 17:41:54 +03:00
const auto t_end = std :: chrono :: high_resolution_clock :: now ();
2023-03-25 20:26:40 +02:00
if ( i == 0 ) {
2023-05-08 17:41:54 +03:00
const float t_total = std :: chrono :: duration < float > ( t_end - t_start ). count ();
fprintf ( stderr , "%s: %.2f seconds per pass - ETA " , __func__ , t_total );
int total_seconds = ( int )( t_total * n_chunk );
2023-04-21 14:57:57 +02:00
if ( total_seconds >= 60 * 60 ) {
2023-05-08 17:41:54 +03:00
fprintf ( stderr , "%d hours " , total_seconds / ( 60 * 60 ));
2023-04-21 14:57:57 +02:00
total_seconds = total_seconds % ( 60 * 60 );
}
2023-05-08 17:41:54 +03:00
fprintf ( stderr , "%d minutes \n " , total_seconds / 60 );
2023-03-25 20:26:40 +02:00
}
2023-05-08 17:41:54 +03:00
2023-03-25 20:26:40 +02:00
// We get the logits for all the tokens in the context window (params.n_ctx)
// from llama_eval above. Now, based on https://huggingface.co/docs/transformers/perplexity,
2023-05-08 17:41:54 +03:00
// calculate the perplexity over the last half of the window (so the model always has
2023-03-25 20:26:40 +02:00
// some context to predict the token).
//
// We rely on the fact that attention in the forward pass only looks at previous
// tokens here, so the logits returned for each token are an accurate representation
// of what the model would have predicted at that point.
//
// Example, we have a context window of 512, we will compute perplexity for each of the
// last 256 tokens. Then, we split the input up into context window size chunks to
// process the entire prompt.
2023-04-13 14:50:42 -07:00
for ( int j = std :: min ( 512 , params . n_ctx / 2 ); j < params . n_ctx - 1 ; ++ j ) {
2023-03-25 20:26:40 +02:00
// Calculate probability of next token, given the previous ones.
2023-05-08 17:41:54 +03:00
const std :: vector < float > tok_logits (
logits . begin () + ( j + 0 ) * n_vocab ,
2023-04-13 14:50:42 -07:00
logits . begin () + ( j + 1 ) * n_vocab );
2023-05-08 17:41:54 +03:00
const float prob = softmax ( tok_logits )[ tokens [ start + j + 1 ]];
2023-03-25 20:26:40 +02:00
nll += - std :: log ( prob );
++ count ;
}
// perplexity is e^(average negative log-likelihood)
printf ( "[%d]%.4lf," , i + 1 , std :: exp ( nll / count ));
fflush ( stdout );
}
printf ( " \n " );
}
2023-07-28 20:25:36 +02:00
void hellaswag_score ( llama_context * ctx , const gpt_params & params ) {
// Calculates hellaswag score (acc_norm) from prompt
//
// Data extracted from the HellaSwag validation dataset (MIT license) https://github.com/rowanz/hellaswag/blob/master/data/hellaswag_val.jsonl
// All used data fields are preprocessed as in https://github.com/EleutherAI/lm-evaluation-harness/blob/df3da98c5405deafd519c2ddca52bb7c3fe36bef/lm_eval/tasks/hellaswag.py#L62-L68
//
// All 10042 tasks should be extracted to keep the results standardized like other implementations.
//
// Datafile layout:
// ['??'] denotes json fields
// 6 lines per task:
// ['activity_label'] + ": " +['ctx'] - The first part of the query, the context
// ['label'] - The index the best common sense ending aka gold ending
// ['endings'][0] - Endings added to the first part of the query
// ['endings'][1]
// ['endings'][2]
// ['endings'][3]
2023-07-22 14:21:24 +02:00
std :: vector < std :: string > prompt_lines ;
std :: istringstream strstream ( params . prompt );
std :: string line ;
while ( std :: getline ( strstream , line , '\n' )) {
prompt_lines . push_back ( line );
}
2023-07-28 20:25:36 +02:00
if ( prompt_lines . size () % 6 != 0 ) {
fprintf ( stderr , "%s : number of lines in prompt not a multiple of 6. \n " , __func__ );
return ;
}
size_t hs_task_count = prompt_lines . size () / 6 ;
fprintf ( stderr , "%s : loaded %lu tasks from prompt. \n " , __func__ , hs_task_count );
// This is needed as usual for LLaMA models
bool prepend_bos = true ;
// Number of tasks to use when computing the score
if ( params . hellaswag_tasks < hs_task_count ) {
hs_task_count = params . hellaswag_tasks ;
}
// The tasks should be randomized so the score stabilizes quickly.
bool randomize_tasks = true ;
// The random seed should not impact the final result if the computation is done over enough tasks, so kept hardcoded for now
std :: mt19937 rng ( 1 );
// Dataholder for hellaswag tasks
struct hs_data_t {
std :: string context ;
size_t gold_ending_idx ;
std :: string ending [ 4 ];
size_t ending_logprob_count [ 4 ];
double ending_logprob [ 4 ];
};
fprintf ( stderr , "%s : selecting %lu %s tasks. \n " , __func__ , hs_task_count , ( randomize_tasks ? "randomized" : "the first" ) );
// Select and read data from prompt lines
hs_data_t * hs_data = new hs_data_t [ hs_task_count ];
for ( size_t i = 0 ; i < hs_task_count ; i ++ ) {
size_t idx = i ;
// Select a random example of those left in the prompt
if ( randomize_tasks ) {
std :: uniform_int_distribution < size_t > dist ( 0 , prompt_lines . size () / 6 - 1 ) ;
idx = dist ( rng );
}
hs_data [ i ]. context = prompt_lines [ idx * 6 ];
hs_data [ i ]. gold_ending_idx = std :: stoi ( prompt_lines [ idx * 6 + 1 ] );
for ( size_t j = 0 ; j < 4 ; j ++ ) {
hs_data [ i ]. ending [ j ] = " " + prompt_lines [ idx * 6 + 2 + j ];
}
// Delete the selected random example from the prompt
if ( randomize_tasks ) {
prompt_lines . erase ( std :: next ( prompt_lines . begin (), idx * 6 ) , std :: next ( prompt_lines . begin (), idx * 6 + 6 ) );
}
}
fprintf ( stderr , "%s : calculating hellaswag score over selected tasks. \n " , __func__ );
printf ( " \n task \t acc_norm \n " );
double acc = 0.0f ;
2023-07-22 14:21:24 +02:00
const int n_vocab = llama_n_vocab ( ctx );
2023-07-28 20:25:36 +02:00
for ( size_t task_idx = 0 ; task_idx < hs_task_count ; task_idx ++ ) {
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// Tokenize the context to count tokens
std :: vector < int > context_embd = :: llama_tokenize ( ctx , hs_data [ task_idx ]. context , prepend_bos );
size_t context_size = context_embd . size ();
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
for ( size_t ending_idx = 0 ; ending_idx < 4 ; ending_idx ++ ) {
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// Tokenize the query
std :: vector < int > query_embd = :: llama_tokenize ( ctx , hs_data [ task_idx ]. context + hs_data [ task_idx ]. ending [ ending_idx ], prepend_bos );
size_t query_size = query_embd . size ();
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// Stop if query wont fit the ctx window
if ( query_size > ( size_t ) params . n_ctx ) {
fprintf ( stderr , "%s : number of tokens in query %lu > n_ctxl \n " , __func__ , query_size );
return ;
}
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// Speedup small evaluations by evaluating atleast 32 tokens
if ( query_size < 32 ) {
query_embd . resize ( 32 );
}
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// Evaluate the query
if ( llama_eval ( ctx , query_embd . data (), query_embd . size (), 0 , params . n_threads )) {
fprintf ( stderr , "%s : failed to eval \n " , __func__ );
return ;
}
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
const auto query_logits = llama_get_logits ( ctx );
std :: vector < float > logits ;
logits . insert ( logits . end (), query_logits , query_logits + query_size * n_vocab );
hs_data [ task_idx ]. ending_logprob_count [ ending_idx ] = 0 ;
hs_data [ task_idx ]. ending_logprob [ ending_idx ] = 0.0f ;
// Calculate the logprobs over the ending
for ( size_t j = context_size - 1 ; j < query_size - 1 ; j ++ ) {
// Calculate probability of next token, given the previous ones.
const std :: vector < float > tok_logits (
logits . begin () + ( j + 0 ) * n_vocab ,
logits . begin () + ( j + 1 ) * n_vocab );
const float prob = softmax ( tok_logits )[ query_embd [ j + 1 ]];
hs_data [ task_idx ]. ending_logprob [ ending_idx ] += std :: log ( prob );
hs_data [ task_idx ]. ending_logprob_count [ ending_idx ] ++ ;
}
// Calculate the mean token logprob for acc_norm
hs_data [ task_idx ]. ending_logprob [ ending_idx ] /= hs_data [ task_idx ]. ending_logprob_count [ ending_idx ];
// printf("task %lu, ending %lu, whole_len %lu, context_len %lu, ending_logprob_count %lu, ending_logprob %.4f\n",
// task_idx,ending_idx,whole_size,context_size, hs_data[task_idx].ending_logprob_count[ending_idx], hs_data[task_idx].ending_logprob[ending_idx] );
2023-07-22 14:21:24 +02:00
}
2023-07-28 20:25:36 +02:00
// Find the ending with maximum logprob
size_t ending_logprob_max_idx = - 1 ;
double ending_logprob_max_val = - INFINITY ;
for ( size_t j = 0 ; j < 4 ; j ++ ) {
if ( hs_data [ task_idx ]. ending_logprob [ j ] > ending_logprob_max_val ) {
ending_logprob_max_idx = j ;
ending_logprob_max_val = hs_data [ task_idx ]. ending_logprob [ j ];
}
2023-07-22 14:21:24 +02:00
}
2023-07-28 20:25:36 +02:00
// printf("max logprob ending idx %lu, gold ending idx %lu\n", ending_logprob_max_idx, hs_data[task_idx].gold_ending_idx);
2023-07-22 14:21:24 +02:00
2023-07-28 20:25:36 +02:00
// If the gold ending got the maximum logprobe add one accuracy point
if ( ending_logprob_max_idx == hs_data [ task_idx ]. gold_ending_idx ) {
acc += 1.0 ;
2023-07-22 14:21:24 +02:00
}
2023-07-28 20:25:36 +02:00
// Print the accumulated accuracy mean x 100
printf ( "%li \t %.8lf \n " , task_idx + 1 , acc / double ( task_idx + 1 ) * 100.0 );
2023-07-22 14:21:24 +02:00
fflush ( stdout );
}
2023-07-28 20:25:36 +02:00
delete [] hs_data ;
2023-07-22 14:21:24 +02:00
printf ( " \n " );
}
2023-03-25 20:26:40 +02:00
int main ( int argc , char ** argv ) {
gpt_params params ;
2023-04-13 14:50:42 -07:00
params . n_batch = 512 ;
2023-03-25 20:26:40 +02:00
if ( gpt_params_parse ( argc , argv , params ) == false ) {
return 1 ;
}
params . perplexity = true ;
2023-04-13 14:50:42 -07:00
params . n_batch = std :: min ( params . n_batch , params . n_ctx );
2023-03-25 20:26:40 +02:00
if ( params . n_ctx > 2048 ) {
2023-07-07 00:23:49 +08:00
fprintf ( stderr , "%s: warning: model might not support context sizes greater than 2048 tokens (%d specified);"
2023-03-25 20:26:40 +02:00
"expect poor results \n " , __func__ , params . n_ctx );
}
2023-05-01 09:23:47 -07:00
fprintf ( stderr , "%s: build = %d (%s) \n " , __func__ , BUILD_NUMBER , BUILD_COMMIT );
2023-06-29 21:15:15 +08:00
if ( params . seed == LLAMA_DEFAULT_SEED ) {
2023-03-25 20:26:40 +02:00
params . seed = time ( NULL );
}
2023-06-29 21:15:15 +08:00
fprintf ( stderr , "%s: seed = %u \n " , __func__ , params . seed );
2023-03-25 20:26:40 +02:00
std :: mt19937 rng ( params . seed );
if ( params . random_prompt ) {
params . prompt = gpt_random_prompt ( rng );
}
2023-07-10 11:49:56 -04:00
llama_backend_init ( params . numa );
2023-05-20 11:06:11 +03:00
2023-06-24 11:47:58 +03:00
llama_model * model ;
2023-03-25 20:26:40 +02:00
llama_context * ctx ;
2023-05-02 22:39:51 +02:00
// load the model and apply lora adapter, if any
2023-06-24 11:47:58 +03:00
std :: tie ( model , ctx ) = llama_init_from_gpt_params ( params );
if ( model == NULL ) {
2023-05-02 22:39:51 +02:00
fprintf ( stderr , "%s: error: unable to load model \n " , __func__ );
return 1 ;
2023-04-17 17:28:55 +02:00
}
2023-03-25 20:26:40 +02:00
// print system information
{
fprintf ( stderr , " \n " );
fprintf ( stderr , "system_info: n_threads = %d / %d | %s \n " ,
params . n_threads , std :: thread :: hardware_concurrency (), llama_print_system_info ());
}
2023-07-28 20:25:36 +02:00
if ( params . hellaswag ) {
hellaswag_score ( ctx , params );
2023-07-22 14:21:24 +02:00
} else {
perplexity ( ctx , params );
}
2023-03-25 20:26:40 +02:00
llama_print_timings ( ctx );
llama_free ( ctx );
2023-06-24 11:47:58 +03:00
llama_free_model ( model );
2023-03-25 20:26:40 +02:00
2023-07-10 11:49:56 -04:00
llama_backend_free ();
2023-03-25 20:26:40 +02:00
return 0 ;
}