llama : add ggml_cont to trigger bug with Metal

ggml-org · Sep 4, 2023 · 4a6ad90 · 4a6ad90
1 parent bd33e5a
commit 4a6ad90
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/llama.cpp b/llama.cpp
@@ -2387,11 +2387,11 @@ static struct ggml_cgraph * llm_build_llama(
             ggml_set_name(Q, "Q");
 
             struct ggml_tensor * K =
-                ggml_view_3d(ctx0, kv_self.k,
+                ggml_cont(ctx0, ggml_view_3d(ctx0, kv_self.k,
                         n_embd_head, n_past + N, n_head_kv,
                         ggml_element_size(kv_self.k)*n_embd_gqa,
                         ggml_element_size(kv_self.k)*n_embd_head,
-                        ggml_element_size(kv_self.k)*n_embd_gqa*n_ctx*il);
+                        ggml_element_size(kv_self.k)*n_embd_gqa*n_ctx*il));
             offload_func_kq(K);
             ggml_set_name(K, "K");