root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-11 21:39:52 +00:00
Code Issues Actions 12 Packages Projects Releases Wiki Activity

server : fix handling of characters that span multiple tokens when streaming (#4446)

Browse Source
This commit is contained in:
shibe2 2023-12-13 23:57:15 +04:00 committed by GitHub
parent 4d98d9a656
commit 948ff137ec
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 19 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
examples/server/server.cpp
View File

@ -376,7 +376,6 @@ struct llama_client_slot
int32_t num_prompt_tokens = 0;
int32_t num_prompt_tokens_processed = 0;
int32_t multibyte_pending = 0;
json prompt;
std::string generated_text;
@ -425,7 +424,6 @@ struct llama_client_slot
stopped_word = false;
stopped_limit = false;
stopping_word = "";
multibyte_pending = 0;
n_past = 0;
sent_count = 0;
sent_token_probs_index = 0;
@ -992,35 +990,36 @@ struct llama_server_context
slot.generated_text += token_str;
slot.has_next_token = true;
if (slot.multibyte_pending > 0)
// check if there is incomplete UTF-8 character at the end
bool incomplete = false;
for (unsigned i = 1; i < 5 && i <= slot.generated_text.size(); ++i)
{
slot.multibyte_pending -= token_str.size();
unsigned char c = slot.generated_text[slot.generated_text.size() - i];
if ((c & 0xC0) == 0x80)
{
// continuation byte: 10xxxxxx
continue;
}
else if (token_str.size() == 1)
{
const char c = token_str[0];
// 2-byte characters: 110xxxxx 10xxxxxx
if ((c & 0xE0) == 0xC0)
{
slot.multibyte_pending = 1;
// 3-byte characters: 1110xxxx 10xxxxxx 10xxxxxx
// 2-byte character: 110xxxxx ...
incomplete = i < 2;
}
else if ((c & 0xF0) == 0xE0)
{
slot.multibyte_pending = 2;
// 4-byte characters: 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
// 3-byte character: 1110xxxx ...
incomplete = i < 3;
}
else if ((c & 0xF8) == 0xF0)
{
slot.multibyte_pending = 3;
}
else
{
slot.multibyte_pending = 0;
// 4-byte character: 11110xxx ...
incomplete = i < 4;
}
// else 1-byte character or invalid byte
break;
}
if (slot.multibyte_pending == 0)
if (!incomplete)
{
size_t pos = std::min(slot.sent_count, slot.generated_text.size());
const std::string str_test = slot.generated_text.substr(pos);
@ -1055,7 +1054,7 @@ struct llama_server_context
}
}
if (slot.multibyte_pending > 0 && !slot.has_next_token)
if (incomplete)
{
slot.has_next_token = true;
}