Skip to content

Commit ac3b886

Browse files
llama : fix embd when offloading non-repeating layers (#1891)
1 parent 5b9ccaf commit ac3b886

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

llama.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1658,7 +1658,7 @@ static bool llama_eval_internal(
16581658

16591659
// cur = cur*norm(broadcasted)
16601660
cur = ggml_mul(ctx0, cur, model.norm);
1661-
offload_func_nr(cur);
1661+
// offload_func_nr(cur); // TODO CPU + GPU mirrored backend
16621662
ggml_set_name(cur, "result_norm");
16631663

16641664
embeddings = cur;

0 commit comments

Comments
 (0)