gpt4all/gpt4all-backend/utils.cpp
Aaron Miller bbcee1ced5 New tokenizer implementation for MPT and GPT-J
Improves output quality by making these tokenizers more closely
match the behavior of the huggingface `tokenizers` based BPE
tokenizers these models were trained with.

Featuring:
 * Fixed unicode handling (via ICU)
 * Fixed BPE token merge handling
 * Complete added vocabulary handling
2023-05-30 12:05:57 -04:00

146 lines
5.0 KiB
C++

#include "utils.h"
#include "tokenizer/bpe.h"
#include "tokenizer/mpt_tokenizer_config.h"
#include "tokenizer/gptj_tokenizer_config.h"
#include <fstream>
#include <regex>
#include <stdexcept>
void get_bpecpp_tokenizer(const TokenizerType ttype, std::unique_ptr<bpecpp::BPE>& bpe, std::unique_ptr<bpecpp::AdditionalVocabAdapter>& av) {
std::vector<bpecpp::additional_vocab_item> avis;
std::unordered_map<std::string_view, uint32_t> vocab;
std::vector<std::pair<std::string_view, std::string_view>> merges;
uint32_t tok_id = 0;
switch (ttype) {
case TokenizerType::MPT_CHAT:
avis.push_back({ .id = 50277, .content = std::string_view("<|im_start|>"), .special = true });
avis.push_back({ .id = 50278, .content = std::string_view("<|im_end|>"), .special = true });
case TokenizerType::MPT:
for (auto avi_e: mpt_additional_vocab) {
avis.push_back({avi_e.id, avi_e.content.into(mpt_buffer), avi_e.special});
}
for (auto merge: mpt_merges) {
merges.push_back({merge.first.into(mpt_buffer), merge.second.into(mpt_buffer)});
}
for (auto bufref: mpt_vocab) {
vocab.insert({bufref.into(mpt_buffer), tok_id++});
}
break;
case TokenizerType::GPTJ:
for (auto avi_e: gptj_additional_vocab) {
avis.push_back({avi_e.id, avi_e.content.into(gptj_buffer), avi_e.special});
}
for (auto merge: gptj_merges) {
merges.push_back({merge.first.into(gptj_buffer), merge.second.into(gptj_buffer)});
}
for (auto bufref: gptj_vocab) {
vocab.insert({bufref.into(gptj_buffer), tok_id++});
}
break;
default:
throw std::invalid_argument("invalid tokenizer type");
}
av = std::make_unique<bpecpp::AdditionalVocabAdapter>(avis);
bpe = std::make_unique<bpecpp::BPE>(vocab, merges);
}
gpt_vocab::id gpt_sample_top_k_top_p(
const gpt_vocab & vocab,
const size_t actualVocabSize,
const int32_t * last_n_tokens_data,
int last_n_tokens_size,
const std::vector<float> logits,
int top_k,
double top_p,
double temp,
float repeat_penalty,
std::mt19937 & rng) {
int n_logits = actualVocabSize;
const auto last_n_tokens = std::vector<int32_t>(last_n_tokens_data, last_n_tokens_data + last_n_tokens_size);
const auto * plogits = logits.data() + logits.size() - n_logits;
std::vector<std::pair<double, gpt_vocab::id>> logits_id;
logits_id.reserve(n_logits);
{
const float scale = 1.0f/temp;
for (int i = 0; i < n_logits; ++i) {
// repetition penalty from ctrl paper (https://arxiv.org/abs/1909.05858)
// credit https://github.com/facebookresearch/llama/compare/main...shawwn:llama:main
if (std::find(last_n_tokens.begin(), last_n_tokens.end(), i) != last_n_tokens.end()) {
// if score < 0 then repetition penalty has to multiplied to reduce the previous token probability
if (plogits[i] < 0.0f) {
logits_id.push_back(std::make_pair(plogits[i]*scale*repeat_penalty, i));
} else {
logits_id.push_back(std::make_pair(plogits[i]*scale/repeat_penalty, i));
}
} else {
logits_id.push_back(std::make_pair(plogits[i]*scale, i));
}
}
}
// find the top K tokens
std::partial_sort(
logits_id.begin(),
logits_id.begin() + top_k, logits_id.end(),
[](const std::pair<double, gpt_vocab::id> & a, const std::pair<double, gpt_vocab::id> & b) {
return a.first > b.first;
});
logits_id.resize(top_k);
double maxl = -INFINITY;
for (const auto & kv : logits_id) {
maxl = std::max(maxl, kv.first);
}
// compute probs for the top K tokens
std::vector<double> probs;
probs.reserve(logits_id.size());
double sum = 0.0;
for (const auto & kv : logits_id) {
double p = exp(kv.first - maxl);
probs.push_back(p);
sum += p;
}
// normalize the probs
for (auto & p : probs) {
p /= sum;
}
if (top_p < 1.0f) {
double cumsum = 0.0f;
for (int i = 0; i < top_k; i++) {
cumsum += probs[i];
if (cumsum >= top_p) {
top_k = i + 1;
probs.resize(top_k);
logits_id.resize(top_k);
break;
}
}
cumsum = 1.0/cumsum;
for (int i = 0; i < (int) probs.size(); i++) {
probs[i] *= cumsum;
}
}
//printf("\n");
//for (int i = 0; i < (int) probs.size(); i++) {
// printf("%d: '%s' %f\n", i, vocab.id_to_token.at(logits_id[i].second).c_str(), probs[i]);
//}
//exit(0);
std::discrete_distribution<> dist(probs.begin(), probs.end());
int idx = dist(rng);
return logits_id[idx].second;
}