context : do not clear output buffer on reserve (#13152)

Co-authored-by: pockers21 <redacted>

diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index a52b6850b465d20f7acc157f7409ef0a74ab146b..e49225aa22433ef04261c09d6ef82d5fddb0bc4e 100644 (file)
--- 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;