Skip to content

Commit 062cc26

Browse files
committed
common : warm-up with 2 tokens - seems to work better
1 parent 35acd23 commit 062cc26

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

common/common.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -770,7 +770,7 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
770770
{
771771
LOG("warming up the model with an empty run\n");
772772

773-
const std::vector<llama_token> tmp = { llama_token_bos(lctx), };
773+
const std::vector<llama_token> tmp = { llama_token_bos(lctx), llama_token_eos(lctx), };
774774
llama_eval(lctx, tmp.data(), tmp.size(), 0, params.n_threads);
775775
llama_reset_timings(lctx);
776776
}

0 commit comments

Comments
 (0)