llama : add option to save memory in device buffers (#22679)

* llama : add option to save memory in device buffers * tests : extend llama-save-load-state
2026-05-05 06:35:07 +03:00
parent fa595462ca
commit d6e7b033a4
11 changed files with 402 additions and 58 deletions
@@ -784,7 +784,7 @@ void llama_memory_recurrent::state_write_data(llama_io_write_i & io, const std::
    const uint32_t n_layer = hparams.n_layer;

    io.write(&s_trans, sizeof(s_trans));
-    io.write(&n_layer,   sizeof(n_layer));
+    io.write(&n_layer, sizeof(n_layer));

    // Iterate and write all the R tensors first, each row is a cell
    // Get whole range at a time