diff options
author | Aaron Miller <apage43@ninjawhale.com> | 2023-07-01 11:14:59 -0700 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-07-01 21:14:59 +0300 |
commit | 2f8cd979ecd1fa582852e7136e92ff8990b98fd8 (patch) | |
tree | 78164d2ee6abe345271599f8b0affdafe8035e1d /llama.cpp | |
parent | 471aab6e4cb89d8ef6d043f1bc93acb6eb78ab67 (diff) |
metal : release buffers when freeing metal context (#2062)
Diffstat (limited to 'llama.cpp')
-rw-r--r-- | llama.cpp | 8 |
1 files changed, 7 insertions, 1 deletions
@@ -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; |