From 0373486dbc0dccbdcb3b5fdd65759d88cec06196 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 26 Aug 2025 17:45:17 +0300
Subject: [PATCH] graph : fix assert in memory-less build_attn (#15590)

ggml-ci
---
 src/llama-graph.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
index 6419d739bd..b928e9e16e 100644
--- a/src/llama-graph.cpp
+++ b/src/llama-graph.cpp
@@ -1376,7 +1376,7 @@ ggml_tensor * llm_graph_context::build_attn(
 
     // [TAG_NO_CACHE_PAD]
     // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams
-    assert(!ubatch.equal_seqs());
+    assert(!ubatch.equal_seqs() || (k_cur->ne[3] == 1 && k_cur->ne[3] == ubatch.n_seqs_unq));
 
     ggml_tensor * q = q_cur;
     ggml_tensor * k = k_cur;