From 58e3866d02260ffd3baea6dc0530c528ffb66228 Mon Sep 17 00:00:00 2001 From: Younes B <49240599+younesbelkada@users.noreply.github.com> Date: Tue, 8 Jul 2025 12:30:55 +0400 Subject: [PATCH] Update src/llama-model.cpp MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Co-authored-by: Sigbjørn Skjæret --- src/llama-model.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/llama-model.cpp b/src/llama-model.cpp index 421ebe9628..cdc753965c 100644 --- a/src/llama-model.cpp +++ b/src/llama-model.cpp @@ -14868,7 +14868,7 @@ struct llm_build_falcon_h1 : public llm_graph_context { // TODO: skip computing output earlier for unused tokens y = ggml_add(ctx0, y, ggml_mul(ctx0, x, model.layers[il].ssm_d)); - y = ggml_mul(ctx0, y, ggml_silu(ctx0, ggml_cont(ctx0, z))); + y = ggml_swiglu_split(ctx0, ggml_cont(ctx0, z), y); // grouped RMS norm if (model.layers[il].ssm_norm) {