From 2da64a2f8a47894d70bc87a160dcb57a3df63b6e Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Mon, 5 Jan 2026 12:34:23 +0200
Subject: [PATCH] models : fix backend assignment for Granite/Nemotron graphs
 (#18599)

* models : fix backend assignment for Granite/Nemotron graphs

* cont : add ref

* cont : move call to build_inp_embd()
---
 src/llama-graph.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
index 6cde7338c2..86c5472638 100644
--- a/src/llama-graph.cpp
+++ b/src/llama-graph.cpp
@@ -1326,6 +1326,10 @@ ggml_tensor * llm_graph_context::build_inp_embd(ggml_tensor * tok_embd) const {
 
     res->add_input(std::move(inp));
 
+    // make sure the produced embeddings are immediately materialized in the ggml graph
+    // ref: https://github.com/ggml-org/llama.cpp/pull/18599
+    ggml_build_forward_expand(gf, cur);
+
     return cur;
 }