CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-04 09:32:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : enable special tokens during tokenization by default

Browse Source
This commit is contained in:
Georgi Gerganov
2023-11-24 11:10:23 +02:00
parent b94b10914c
commit c544faed74
1 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
examples/server/server.cpp
View File

@@ -618,6 +618,11 @@ struct llama_server_context
std::vector<llama_token> tokenize(const json & json_prompt, bool add_bos) const
{
// TODO: currently, we tokenize using special tokens by default
// this is not always correct (see https://github.com/ggerganov/llama.cpp/pull/4160#issuecomment-1824826216)
// but it's better compared to completely ignoring ChatML and other chat templates
const bool TMP_FORCE_SPECIAL = true;
// If `add_bos` is true, we only add BOS, when json_prompt is a string,
// or the first element of the json_prompt array is a string.
std::vector<llama_token> prompt_tokens;
@@ -633,12 +638,12 @@ struct llama_server_context
std::vector<llama_token> p;
if (first)
{
p = ::llama_tokenize(ctx, s, add_bos);
p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
first = false;
}
else
{
p = ::llama_tokenize(ctx, s, false);
p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
}
prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
}
@@ -655,7 +660,7 @@ struct llama_server_context
else
{
auto s = json_prompt.template get<std::string>();
prompt_tokens = ::llama_tokenize(ctx, s, add_bos);
prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
}
return prompt_tokens;