diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
index 2a57ef5375..341aac0610 100644
--- a/src/llama-graph.cpp
+++ b/src/llama-graph.cpp
@@ -2207,6 +2207,7 @@ ggml_tensor * llm_graph_context::build_rs(
     ggml_tensor * output_states = get_state_rows(ctx0, states, state_copy_main);
     ggml_build_forward_expand(gf, output_states);
 
+    // copy extra states which won't be changed further (between n_seqs and n_rs)
     ggml_tensor * states_extra = ggml_get_rows(ctx0, states, state_copy_extra);
     ggml_build_forward_expand(gf,
         ggml_cpy(ctx0,
diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index 7afad1b66e..2ae023aaca 100644
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -2233,7 +2233,6 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<|end_of_text|>" // granite
                         || t.first == "<EOT>"
                         || t.first == "_<EOT>"
-                        || t.first == "_<EOT>"
                         || t.first == "[EOT]" // Kimi-K2
                         || t.first == "<｜end▁of▁sentence｜>" // DeepSeek
                         || t.first == "<end_of_utterance>" // smoldocling