diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
index b0e3f63597a76..f4e6dfe392f8c 100644
--- a/src/llama-graph.cpp
+++ b/src/llama-graph.cpp
@@ -1488,6 +1488,10 @@ ggml_tensor * llm_graph_context::build_attn(
 
     if (wo) {
         cur = build_lora_mm(wo, cur);
+        if (arch == LLM_ARCH_GLM4) {
+            // GLM4 seems to have numerical issues with half-precision accumulators
+            ggml_mul_mat_set_prec(cur, GGML_PREC_F32);
+        }
     }
 
     if (wo_b) {