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

llama : sync gguf-llama with llama (#2613)

Browse source 
* llama : sync gguf-llama with llama

* tests : fix build + warnings (test-tokenizer-1 still fails)

* tests : fix wstring_convert

* convert : fix layer names

* llama : sync gguf-llama.cpp

* convert : update HF converter to new tokenizer voodoo magics
This commit is contained in:
Georgi Gerganov 2023-08-14 21:33:33 +03:00 • committed by GitHub
parent afc4ca2889
commit 7494c78428
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
8 changed files with 590 additions and 293 deletions
Download patch file Download diff file Expand all files Collapse all files

26
tests/test-tokenizer-1.cpp
View file

@ -8,8 +8,9 @@
#include <codecvt>
#include <map>
#include <vector>
#include <locale>
static std::string vocab_type(llama_context* ctx) {
static std::string vocab_type(llama_context * ctx) {
return llama_n_vocab(ctx) == 32000 ? "spm": "bpe";
}
@ -32,9 +33,9 @@ static std::string escape_whitespace(const std::string& text) {
return result;
}
static std::string unescape_whitespace(llama_context* ctx, const std::vector<llama_token>& tokens) {
static std::string unescape_whitespace(llama_context * ctx, const std::vector<llama_token> & tokens) {
std::string result;
for (int i = 0; i < tokens.size(); ++i) {
for (size_t i = 0; i < tokens.size(); ++i) {
result += llama_token_to_str(ctx, tokens[i]);
}
return result;
@ -85,17 +86,17 @@ int main(int argc, char **argv) {
if (tokens.size() == 1) {
if (i != tokens[0]) {
std::string backward = llama_token_to_str(ctx, tokens[0]);
fprintf(stderr, "%s : error: token %d is string %s but bpe returns token %d %s\n",
fprintf(stderr, "%s : error: token %d is string %s but bpe returns token %d %s\n",
__func__, i, llama_token_to_str(ctx, i).c_str(), tokens[0], backward.c_str());
return 2;
}
} else {
if ((vocab_type(ctx) == "spm" && i <= 258) ||
if ((vocab_type(ctx) == "spm" && i <= 258) ||
(vocab_type(ctx) == "bpe" && (i == 0 || i >= 100000))) {
fprintf(stderr, "%s : info: token %d is string %s and bpe returns tokens %s\n",
fprintf(stderr, "%s : info: token %d is string %s and bpe returns tokens %s\n",
__func__, i, llama_token_to_str(ctx, i).c_str(), unescape_whitespace(ctx, tokens).c_str());
} else {
fprintf(stderr, "%s : error: token %d is string %s but bpe returns tokens %s\n",
fprintf(stderr, "%s : error: token %d is string %s but bpe returns tokens %s\n",
__func__, i, llama_token_to_str(ctx, i).c_str(), unescape_whitespace(ctx, tokens).c_str());
return 2;
}
@ -105,10 +106,15 @@ int main(int argc, char **argv) {
std::wstring_convert<typename std::codecvt_utf8<wchar_t>, wchar_t> converter;
for (wchar_t ch = 0x0000; ch < 0xffff; ++ch) {
std::wstring wstr(1, ch);
std::string str = converter.to_bytes(wstr);
std::vector<llama_token> tokens = llama_tokenize(ctx, escape_whitespace(str).c_str(), false);
std::string str;
try {
str = converter.to_bytes(wstr);
} catch (std::exception & e) {
continue;
}
std::vector<llama_token> tokens = llama_tokenize(ctx, escape_whitespace(str), false);
if (tokens.size() == 1) {
fprintf(stderr, "%s : info: %s tokenized to %d \n",
fprintf(stderr, "%s : info: %s tokenized to %d \n",
__func__, str.c_str(), tokens[0]);
}
}