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

llama : restore prefix space in llama tokenizer (#4081)

Browse source
This commit is contained in:
Jared Van Bortel 2023-11-15 11:34:47 -05:00 committed by GitHub
parent 1cf2850d52
commit a6fc554e26
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
llama.cpp
View file

@ -6283,7 +6283,10 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
// by modifying llm_tokenizer_x to operate with string offsets like pre-tokenizer
// and passing 'add space prefix' as bool argument
//
auto raw_text = (special ? "" : " ") + fragment.raw_text.substr(fragment.offset, fragment.length);
auto raw_text = fragment.raw_text.substr(fragment.offset, fragment.length);
if (&fragment == &fragment_buffer.front()) {
raw_text = " " + raw_text; // prefix with space if the first token is not special
}
#ifdef PRETOKENIZERDEBUG
fprintf(stderr,"TT: (%ld %ld %ld) '%s'\n", raw_text.length(), fragment.offset, fragment.length, raw_text.c_str());