Fixed incorrectly applying RMS norm twice (#1925)

diff --git a/llama.cpp b/llama.cpp
index dfbb85a682baad745c620053eecc254f8c28590c..45360cea37ba5e68d8b09e9c07dcaf6655fed79e 100644 (file)
--- a/llama.cpp
+++ b/llama.cpp
@@ -1657,11 +1657,7 @@ static bool llama_eval_internal(
     {
         cur = ggml_rms_norm(ctx0, inpL);
         offload_func_nr(cur);
-        ggml_set_name(cur, "rms_norm_inpL");
-
-        cur = ggml_rms_norm(ctx0, cur);
-        offload_func_nr(cur);
-        ggml_set_name(cur, "rms_norm_after");
+        ggml_set_name(cur, "rms_norm_2");
 
         // cur = cur*norm(broadcasted)
         cur = ggml_mul(ctx0, cur, model.norm);