mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-27 21:46:57 +02:00
cont : fix conflict
This commit is contained in:
@@ -1465,8 +1465,10 @@ bool rpc_server::free_buffer(const rpc_msg_free_buffer_req & request) {
|
||||
}
|
||||
// Discard all cached graphs to avoid use-after-free in graph_recompute,
|
||||
// since their nodes may hold pointers to the buffer being freed.
|
||||
for (auto & sg : stored_graphs) {
|
||||
sg.graph = nullptr;
|
||||
for (auto & sgs : stored_graphs) {
|
||||
for (auto & sg : sgs) {
|
||||
sg.second.graph = nullptr;
|
||||
}
|
||||
}
|
||||
ggml_backend_buffer_free(buffer);
|
||||
buffers.erase(buffer);
|
||||
|
||||
Reference in New Issue
Block a user