vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

perplexity: add additional KL-divergence statistics

Browse source
This commit is contained in:
Iwan Kawrakow 2024-01-22 18:02:50 +02:00
parent 8128fa0bd3
commit 150af7ecf7
1 changed files with 31 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

37
examples/perplexity/perplexity.cpp
View file

@ -226,7 +226,7 @@ struct kl_divergence_result {
size_t count = 0;
};
static void log_softmax(int n_vocab, const float * logits, const uint16_t * base_log_prob, int tok, kl_divergence_result & kld) {
static double log_softmax(int n_vocab, const float * logits, const uint16_t * base_log_prob, int tok, kl_divergence_result & kld) {
float max_logit = logits[0];
int imax = 0;
for (int i = 1; i < n_vocab; ++i) {
@ -269,14 +269,16 @@ static void log_softmax(int n_vocab, const float * logits, const uint16_t * base
kld.sum_kld2 += sum*sum;
++kld.count;
if (imax == imax_base) ++kld.n_same_top;
return sum;
}
static void process_logits(int n_vocab, const float * logits, const int * tokens, int n_token,
std::vector<std::thread> & workers, const std::vector<uint16_t> & base_log_probs, kl_divergence_result & kld) {
std::vector<std::thread> & workers, const std::vector<uint16_t> & base_log_probs, kl_divergence_result & kld,
float * kld_values) {
std::mutex mutex;
const int nv = 2*((n_vocab + 1)/2) + 4;
int counter = 0;
auto compute = [&mutex, &counter, &base_log_probs, &kld, n_vocab, logits, tokens, n_token, nv] () {
auto compute = [&mutex, &counter, &base_log_probs, &kld, n_vocab, logits, tokens, n_token, nv, kld_values] () {
kl_divergence_result local_kld;
while (true) {
std::unique_lock<std::mutex> lock(mutex);
@ -293,7 +295,8 @@ static void process_logits(int n_vocab, const float * logits, const int * tokens
break;
}
lock.unlock();
log_softmax(n_vocab, logits + i*n_vocab, base_log_probs.data() + i*nv, tokens[i+1], local_kld);
double v = log_softmax(n_vocab, logits + i*n_vocab, base_log_probs.data() + i*nv, tokens[i+1], local_kld);
kld_values[i] = (float)v;
}
};
for (auto & w : workers) {
@ -1628,7 +1631,7 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
in.read((char *)&n_vocab, sizeof(n_vocab));
in.read((char *)&n_chunk, sizeof(n_chunk));
if (in.fail()) {
fprintf(stderr, "%s: failed rwading n_vocab, n_chunk from %s\n", __func__, params.logits_file.c_str());
fprintf(stderr, "%s: failed reading n_vocab, n_chunk from %s\n", __func__, params.logits_file.c_str());
return;
}
if (n_vocab != llama_n_vocab(llama_get_model(ctx))) {
@ -1647,6 +1650,7 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
const bool add_bos = llama_should_add_bos_token(llama_get_model(ctx));
std::vector<uint16_t> log_probs_uint16(size_t(n_ctx - 1 - n_ctx/2) * nv);
std::vector<float> kld_values(size_t(n_ctx - 1 - n_ctx/2)*n_chunk);
std::vector<float> logits;
if (num_batches > 1) {
logits.reserve(n_ctx * n_vocab);
@ -1665,6 +1669,7 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
};
kl_divergence_result kld;
auto kld_ptr = kld_values.data();
for (int i = 0; i < n_chunk; ++i) {
const int start = i * n_ctx;
@ -1724,7 +1729,8 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
const int first = n_ctx/2;
const float * all_logits = num_batches > 1 ? logits.data() : llama_get_logits(ctx);
process_logits(n_vocab, all_logits + first*n_vocab, tokens.data() + start + first, n_ctx - 1 - first,
workers, log_probs_uint16, kld);
workers, log_probs_uint16, kld, kld_ptr);
kld_ptr += n_ctx - 1 - first;
auto ppl = mean_and_uncertainty(kld.sum_nll, kld.sum_nll2, kld.count);
auto log_ppl_ratio = mean_and_uncertainty(kld.sum_nll_diff, kld.sum_nll_diff2, kld.count);
@ -1742,6 +1748,25 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
}
printf("\n");
if (kld.count < 100) return; // we do not wish to do statistics on so few values
std::sort(kld_values.begin(), kld_values.end());
printf("===== KL-divergence statistics\n");
auto kl_div = mean_and_uncertainty(kld.sum_kld, kld.sum_kld2, kld.count);
printf("Average: %10.6f ±%10.6lf\n", kl_div.first, kl_div.second);
auto kld_median = kld_values.size()%2 == 0 ? 0.5f*(kld_values[kld_values.size()/2] + kld_values[kld_values.size()/2-1])
: kld_values[kld_values.size()/2];
printf("Median : %10.6f\n", kld_median);
printf("Minimum: %10.6f\n", kld_values.front());
printf("Maximum: %10.6f\n", kld_values.back());
const int n_1percent = nearest_int(0.01f*kld_values.size());
printf("KLD_01 : %10.6f\n", kld_values[n_1percent]);
printf("KLD_99 : %10.6f\n", kld_values[kld_values.size()-1-n_1percent]);
const int n_5percent = nearest_int(0.05f*kld_values.size());
printf("KLD_05 : %10.6f\n", kld_values[n_5percent]);
printf("KLD_95 : %10.6f\n", kld_values[kld_values.size()-1-n_5percent]);
}
int main(int argc, char ** argv) {