From fb0471d1753824e75474c24f82fbdd54c94dceda Mon Sep 17 00:00:00 2001 From: pockers21 <134406831+pockers21@users.noreply.github.com> Date: Mon, 28 Apr 2025 06:45:40 -0700 Subject: [PATCH] context : do not clear output buffer on reserve (#13152) Co-authored-by: pockers21 --- src/llama-context.cpp | 2 -- 1 file changed, 2 deletions(-) diff --git a/src/llama-context.cpp b/src/llama-context.cpp index a52b6850..e49225aa 100644 --- a/src/llama-context.cpp +++ b/src/llama-context.cpp @@ -1536,8 +1536,6 @@ int32_t llama_context::output_reserve(int32_t n_outputs) { // set all ids as invalid (negative) std::fill(output_ids.begin(), output_ids.end(), -1); - ggml_backend_buffer_clear(buf_output.get(), 0); - this->n_outputs = 0; this->n_outputs_max = n_outputs_max;