happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

server : fix off-by-1 in server_tokens::size_up_to_pos() (#20279) 

* server : fix off-by-1 in server_tokens::size_up_to_pos()

* cont : fix typo [no ci]
This commit is contained in:
Georgi Gerganov 2026-03-09 16:43:38 +02:00 committed by GitHub
parent f76565db92
commit d6e1556499
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tools/server/server-common.cpp
View File

@ -276,7 +276,7 @@ llama_pos server_tokens::pos_next(int64_t n_tokens) const {
size_t server_tokens::size_up_to_pos(llama_pos max_pos) const {
if (!has_mtmd) {
return std::min((size_t)(max_pos + 1), tokens.size());
return std::min((size_t)max_pos, tokens.size());
}
size_t idx = 0;
@ -296,7 +296,7 @@ size_t server_tokens::size_up_to_pos(llama_pos max_pos) const {
idx++;
}
if (pos > max_pos) {
if (pos >= max_pos) {
break;
}
}

2
tools/server/server-common.h
View File

@ -170,7 +170,7 @@ public:
// the next position after n_tokens. if n_tokens < 0, return the next position after all tokens.
llama_pos pos_next(int64_t n_tokens = -1) const;
// number of tokens with position <= max_pos
// number of tokens with position < max_pos
size_t size_up_to_pos(llama_pos max_pos) const;
const mtmd::input_chunk_ptr & find_chunk(size_t idx) const;

8
tools/server/server-context.cpp
View File

@ -570,7 +570,7 @@ private:
std::vector<server_slot> slots;
int slots_debug = 0;
int n_empty_consequtive = 0;
int n_empty_consecutive = 0;
std::unique_ptr<server_prompt_cache> prompt_cache;
@ -2372,7 +2372,7 @@ private:
} else {
pos_next = std::min(pos_next, std::max(it->pos_min + 1, it->pos_max));
n_past = std::min(slot.prompt.tokens.size_up_to_pos(pos_next), (size_t) it->n_tokens);
SLT_WRN(slot, "restored context checkpoint (pos_min = %d, pos_max = %d, n_tokens = %" PRId64 ", size = %.3f MiB)\n", it->pos_min, it->pos_max, it->n_tokens, (float) checkpoint_size / 1024 / 1024);
SLT_WRN(slot, "restored context checkpoint (pos_min = %d, pos_max = %d, n_tokens = %" PRId64 ", n_past = %d, size = %.3f MiB)\n", it->pos_min, it->pos_max, it->n_tokens, n_past, (float) checkpoint_size / 1024 / 1024);
}
}
@ -2630,11 +2630,11 @@ private:
if (batch.n_tokens == 0) {
SRV_WRN("%s", "no tokens to decode\n");
if (++n_empty_consequtive > 3) {
if (++n_empty_consecutive > 3) {
GGML_ABORT("fatal error - please provide logs and repro in %s\n", "https://github.com/ggml-org/llama.cpp/pull/20277");
}
} else {
n_empty_consequtive = 0;
n_empty_consecutive = 0;
}
int32_t i_next = 0;