diff --git a/src/llama-context.cpp b/src/llama-context.cpp index 929678587e..c591c95be2 100644 --- a/src/llama-context.cpp +++ b/src/llama-context.cpp @@ -1057,6 +1057,8 @@ bool llama_context::set_sampler(llama_seq_id seq_id, llama_sampler * sampler) { sampling.samplers.erase(seq_id); + sched_need_reserve = true; + return true; }