llama : fix restoring the number of outputs from state files (#6687)

diff --git a/llama.cpp b/llama.cpp
index cf95cea14856839ad1d713bdff76af72fcf464c8..a5ef2fd8fa57597276af0dd20bc47b5e85730ba3 100644 (file)
--- a/llama.cpp
+++ b/llama.cpp
@@ -15478,6 +15478,8 @@ size_t llama_state_set_data(struct llama_context * ctx, const uint8_t * src) {
                 GGML_ASSERT((uint32_t) id < ctx->cparams.n_batch);
                 ctx->output_ids[id] = i;
             }
+
+            ctx->n_outputs = n_outputs;
         }
     }