From: compilade <redacted>
Date: Thu, 28 Mar 2024 12:05:54 +0000 (-0400)
Subject: llama : fix command-r inference when omitting outputs (#6367)
X-Git-Tag: upstream/0.0.4488~1921
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=0308f5e3d7bf9879f818b1a4ae589ff36b242af5;p=pkg%2Fggml%2Fsources%2Fllama.cpp

llama : fix command-r inference when omitting outputs (#6367)
---

diff --git a/llama.cpp b/llama.cpp
index 892d46fb..77ec9b7a 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -9152,8 +9152,9 @@ struct llm_build_context {
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                cur     = ggml_get_rows(ctx0,     cur, inp_out_ids);
+                inpL    = ggml_get_rows(ctx0,    inpL, inp_out_ids);
+                ffn_inp = ggml_get_rows(ctx0, ffn_inp, inp_out_ids);
             }
 
             struct ggml_tensor * attn_out = cur;