From 1a3e8ad6db149f87c7134c82c284949b61c73dd2 Mon Sep 17 00:00:00 2001
From: Aaron Miller <apage43@ninjawhale.com>
Date: Fri, 30 Jun 2023 16:08:37 -0700
Subject: [PATCH] release metal buffers when freeing metal context

---
 ggml-metal.m | 4 +++-
 llama.cpp    | 8 +++++++-
 2 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/ggml-metal.m b/ggml-metal.m
index 7551231b9cf32..fd69c41fe357d 100644
--- a/ggml-metal.m
+++ b/ggml-metal.m
@@ -202,7 +202,9 @@ @implementation GGMLMetalClass
 
 void ggml_metal_free(struct ggml_metal_context * ctx) {
     fprintf(stderr, "%s: deallocating\n", __func__);
-
+    for (int i = 0; i < ctx->n_buffers; ++i) {
+        [ctx->buffers[i].metal release];
+    }
     free(ctx);
 }
 
diff --git a/llama.cpp b/llama.cpp
index 049f73e4449cd..14421cb76a198 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -253,7 +253,13 @@ struct llama_model {
 
 struct llama_context {
     llama_context(const llama_model & model, const llama_vocab & vocab) : model(model), vocab(vocab), t_load_us(model.t_load_us), t_start_us(model.t_start_us) {}
-
+#ifdef GGML_USE_METAL
+    ~llama_context() {
+        if (ctx_metal) {
+            ggml_metal_free(ctx_metal);
+        }
+    }
+#endif
     std::mt19937 rng;
 
     bool has_evaluated_once = false;