8000 Fixed incorrectly applying RMS norm twice (#1925) · CodeLinaro/llama.cpp@0ede372 · GitHub
[go: up one dir, main page]
Skip to content

Commit 0ede372

Browse files
JohannesGaesslerJohannesGaessler
authored
Fixed incorrectly applying RMS norm twice (ggml-org#1925)
1 parent 8596af4 commit 0ede372

File tree

1 file changed

+1
-5
lines changed

1 file changed

+1
-5
lines changed

‎llama.cpp

Lines changed: 1 addition & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1657,11 +1657,7 @@ static bool llama_eval_internal(
16571657
{
16581658
cur = ggml_rms_norm(ctx0, inpL);
16591659
offload_func_nr(cur);
1660-
ggml_set_name(cur, "rms_norm_inpL");
1661-
1662-
cur = ggml_rms_norm(ctx0, cur);
1663-
offload_func_nr(cur);
1664-
ggml_set_name(cur, "rms_norm_after");
1660+
ggml_set_name(cur, "rms_norm_2");
16651661

16661662
// cur = cur*norm(broadcasted)
16671663
cur = ggml_mul(ctx0, cur, model.norm);

0 commit comments

Comments
 (0)
0