From 9ee8aeccd73dbc4a000eb93c1c08eea843c752a3 Mon Sep 17 00:00:00 2001
From: l3utterfly <gc.pthzfoldr@gmail.com>
Date: Sun, 8 Oct 2023 00:36:20 +0800
Subject: [PATCH] fixed memory leak by freeing temporary graph during session
 load

---
 llama.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/llama.cpp b/llama.cpp
index 4d5fb7e4be308..242815af2471f 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -8405,6 +8405,10 @@ size_t llama_set_state_data(struct llama_context * ctx, uint8_t * src) {
             ggml_graph_compute_helper(ctx->work_buffer, gf, /*n_threads*/ 1);
 
             ggml_free(cpy_ctx);
+
+            // free our allocated graph
+            free(gf);
+            gf = NULL;
         }
 
         ctx->kv_self.head = kv_head;