diff --git a/common/common.cpp b/common/common.cpp
index 6219f0ce..44678d7a 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -2169,8 +2169,10 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         if (bos != -1) {
             tmp.push_back(bos);
         }
-        tmp.push_back(eos);
-
+	else
+	{
+	    tmp.push_back(eos);
+	}
         if (llama_model_has_encoder(model)) {
             llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0));
             llama_token decoder_start_token_id = llama_model_decoder_start_token(model);
diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index 41b93df5..95df06dc 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -1586,7 +1586,7 @@ int main(int argc, char ** argv) {
         if (params.warmup) {
             if (t.n_prompt > 0) {
                 //test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads);
-                test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads);
+                test_prompt(ctx, 1, 0, t.n_batch, t.n_threads);
             }
             if (t.n_gen > 0) {
                 test_gen(ctx, 1, 0, t.n_threads);
diff --git a/src/llama.cpp b/src/llama.cpp
index 2ee15c7a..b2553802 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -14521,7 +14521,7 @@ static struct ggml_cgraph * llama_build_graph(
     const llama_vocab * vocab = llama_get_vocab(&lctx);
     llama_token bos = llama_token_bos_impl(*vocab);
     llama_token eos = llama_token_eos_impl(*vocab);
-    bool is_warming_up = (batch.n_tokens == 2 && batch.token[0] == bos && batch.token[1] == eos);
+    bool is_warming_up = (batch.n_tokens == 1 && batch.token[0] == bos);
     struct llm_build_context llm(lctx, batch, cb, worst_case, is_warming_up);
 
     llm.init();