From b01b062ab5506f2b04d5a7c7ce43db70e6aed532 Mon Sep 17 00:00:00 2001
From: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
Date: Mon, 8 Apr 2024 00:25:54 +0200
Subject: [PATCH] llama: dbrx: fix build kv att out

---
 llama.cpp | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)

diff --git a/llama.cpp b/llama.cpp
index f397cc28f..1d165ce74 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -7155,7 +7155,7 @@ struct llm_build_context {
                 cb(Vcur, "Vcur", il);
 
                 cur = llm_build_kv(ctx0, model, hparams, kv_self, gf,
-                                   model.layers[il].wo, model.layers[il].bo,
+                                   model.layers[il].attn_out_norm, NULL,
                                    Kcur, Vcur, Qcur, KQ_mask, nullptr, n_ctx, n_tokens, kv_head, n_kv, 1.0f, cb, il);
             }
 
@@ -7235,14 +7235,6 @@ struct llm_build_context {
                 cur = moe_out;
             }
 
-            // DbrxNormAttentionNorm
-            {
-                cur = llm_build_norm(ctx0, cur, hparams,
-                                     model.layers[il].layer_out_norm, NULL,
-                                     LLM_NORM, cb, il);
-                cb(cur, "layer_out_norm", il);
-            }
-
             cur = ggml_add(ctx0, cur, ffn_inp);
             cb(cur, "ffn_out", il);