Unbreak persimmon after #3837 (#4010)

diff --git a/llama.cpp b/llama.cpp
index d220ff3e9b130c3cf48fd49b26966fc3aa753e04..d682d2864d283677df636b17e05c118ddfe06f43 100644 (file)
--- a/llama.cpp
+++ b/llama.cpp
@@ -4209,7 +4209,7 @@ struct llm_build_context {
                 struct ggml_tensor * Kcur = ggml_concat(ctx0, krotated, kpass);
                 cb(Kcur, "Kcur", il);
 
-                struct ggml_tensor * Q = ggml_cont(ctx0, ggml_permute(ctx0, Qcur, 1, 2, 0, 3));
+                struct ggml_tensor * Q = ggml_cont(ctx0, ggml_permute(ctx0, Qcur, 2, 1, 0, 3));
                 cb(Q, "Q", il);
 
                 Kcur = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 2, 1, 0, 3));