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

server : fix misplaced n_keep varible definition

Browse source
This commit is contained in:
Jared Van Bortel 2024-02-21 10:32:48 -05:00
parent f921fc3ecd
commit 6b34d50135
1 changed files with 1 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/server/server.cpp
View file

@ -1440,8 +1440,6 @@ struct llama_server_context
task.target_id = -1; task.target_id = -1;
queue_tasks.post(task); queue_tasks.post(task);
int32_t n_keep = slot.params.n_keep + add_bos_token;
for (llama_client_slot &slot : slots) for (llama_client_slot &slot : slots)
{ {
if (slot.ga_n == 1) if (slot.ga_n == 1)
@ -1449,6 +1447,7 @@ struct llama_server_context
if (slot.is_processing() && system_tokens.size() + slot.cache_tokens.size() >= (size_t) slot.n_ctx) if (slot.is_processing() && system_tokens.size() + slot.cache_tokens.size() >= (size_t) slot.n_ctx)
{ {
// Shift context // Shift context
const int n_keep = slot.params.n_keep + add_bos_token;
const int n_left = system_tokens.size() + slot.n_past - n_keep; const int n_left = system_tokens.size() + slot.n_past - n_keep;
const int n_discard = n_left / 2; const int n_discard = n_left / 2;