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

Merge branch 'master' into compilade/lazy-convert-hf

Browse source
This commit is contained in:
Francis Couture-Harpin 2024-05-06 08:07:08 -04:00
commit 68c5ac628f
54 changed files with 1151 additions and 771 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/server/server.cpp
View file

@ -1383,9 +1383,10 @@ struct server_context {
if (!slot.params.stream && slot.stopped_word) {
const std::vector<llama_token> stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false);
size_t safe_offset = std::min(slot.generated_token_probs.size(), stop_word_toks.size());
probs = std::vector<completion_token_output>(
slot.generated_token_probs.begin(),
slot.generated_token_probs.end() - stop_word_toks.size());
slot.generated_token_probs.end() - safe_offset);
} else {
probs = std::vector<completion_token_output>(
slot.generated_token_probs.begin(),