vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-08-25 08:14:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

mtmd: fix silent prompt truncation on embedded NUL (#25548)

Browse source 
* mtmd: fix silent prompt truncation on embedded NUL

mtmd_input_text carried the prompt as a bare const char* with no
length, so a NUL byte in message content cut the prompt at the
tokenizer boundary and dropped every later message plus the assistant
marker, with no log. Add an explicit text_len and thread it through,
matching llama_tokenize and the text only path.

* cleanup

---------

Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
This commit is contained in:
Pascal 2026-07-13 00:47:25 +02:00 • committed by GitHub
parent 0c4fa7a989
commit 4114ba18b2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 8 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

3
tools/mtmd/mtmd-cli.cpp
View file

@ -250,7 +250,8 @@ static int eval_message(mtmd_cli_context & ctx, common_chat_msg & msg) {
LOG_DBG("formatted_chat.prompt: %s\n", formatted_chat.c_str()); LOG_DBG("formatted_chat.prompt: %s\n", formatted_chat.c_str());
mtmd_input_text text; mtmd_input_text text;
text.text = formatted_chat.c_str(); text.text = formatted_chat.data();
text.text_len = formatted_chat.size();
text.add_special = add_bos; text.add_special = add_bos;
text.parse_special = true; text.parse_special = true;

5
tools/mtmd/mtmd.cpp
View file

@ -809,7 +809,7 @@ void mtmd_free(mtmd_context * ctx) {
struct mtmd_tokenizer { struct mtmd_tokenizer {
mtmd_context * ctx; mtmd_context * ctx;
std::string input_text; std::string input_text; // note: can contain null bytes; do not use c_str()
bool add_special; bool add_special;
bool parse_special; bool parse_special;
const llama_vocab * vocab; const llama_vocab * vocab;
@ -839,9 +839,10 @@ struct mtmd_tokenizer {
size_t n_bitmaps) : ctx(ctx) { size_t n_bitmaps) : ctx(ctx) {
add_special = text->add_special; add_special = text->add_special;
parse_special = text->parse_special; parse_special = text->parse_special;
input_text = text->text;
vocab = ctx->vocab; vocab = ctx->vocab;
input_text.assign(text->text, text->text_len);
std::vector<const mtmd_bitmap *> bitmaps(bmps, bmps + n_bitmaps); std::vector<const mtmd_bitmap *> bitmaps(bmps, bmps + n_bitmaps);
auto parts_str = split_text(input_text, ctx->media_marker); auto parts_str = split_text(input_text, ctx->media_marker);
size_t i_bm = 0; size_t i_bm = 0;

1
tools/mtmd/mtmd.h
View file

@ -67,6 +67,7 @@ struct mtmd_batch;
struct mtmd_input_text { struct mtmd_input_text {
const char * text; const char * text;
size_t text_len;
bool add_special; bool add_special;
bool parse_special; bool parse_special;
}; };

3
tools/server/server-common.cpp
View file

@ -705,7 +705,8 @@ server_tokens process_mtmd_prompt(mtmd_context * mctx, const std::string & promp
std::vector<server_tokens> inputs; std::vector<server_tokens> inputs;
// multimodal // multimodal
mtmd_input_text inp_txt = { mtmd_input_text inp_txt = {
prompt.c_str(), prompt.data(),
prompt.size(),
/* add_special */ true, /* add_special */ true,
/* parse_special */ true, /* parse_special */ true,
}; };