b1961
1387ea21 · llama : pre-allocate input tensors in a separate buffer (#5100) ·