diff --git a/common/common.cpp b/common/common.cpp
index 5f10718ece90c..85c8292e4a1f1 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -101,7 +101,7 @@ int32_t get_num_physical_cores() {
     return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4;
 }
 
-void process_escapes(std::string& input) {
+void process_escapes(std::string & input) {
     std::size_t input_len = input.length();
     std::size_t output_idx = 0;
 
diff --git a/examples/batched/batched.cpp b/examples/batched/batched.cpp
index ee1f8f1bf5dd2..7aaf63ceb1a7c 100644
--- a/examples/batched/batched.cpp
+++ b/examples/batched/batched.cpp
@@ -48,6 +48,8 @@ int main(int argc, char ** argv) {
         params.prompt = "Hello my name is";
     }
 
+    process_escapes(params.prompt);
+
     // init LLM
 
     llama_backend_init();
@@ -78,7 +80,7 @@ int main(int argc, char ** argv) {
     llama_context_params ctx_params = llama_context_default_params();
 
     ctx_params.seed  = 1234;
-    ctx_params.n_ctx = n_kv_req;
+    ctx_params.n_ctx   = n_kv_req;
     ctx_params.n_batch = std::max(n_len, n_parallel);
     ctx_params.n_seq_max       = n_parallel;
     ctx_params.n_threads       = params.n_threads;
diff --git a/llama.cpp b/llama.cpp
index 1a9fe0c4d2cea..9de4a86022b1a 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -13044,6 +13044,9 @@ struct llama_context * llama_new_context_with_model(
     cparams.rope_freq_base   = params.rope_freq_base  == 0.0f ? hparams.rope_freq_base_train  : params.rope_freq_base;
     cparams.rope_freq_scale  = params.rope_freq_scale == 0.0f ? hparams.rope_freq_scale_train : params.rope_freq_scale;
 
+    // this is necessary due to kv_self.n being padded later during inference
+    cparams.n_ctx = GGML_PAD(cparams.n_ctx, 32);
+
     // with causal attention, the batch size is limited by the context size
     cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
     cparams.n_ubatch         = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);
diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index c2916c3e480e0..1998e1cbc4703 100644
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -2091,6 +2091,13 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
         }
     }
 
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32,  64, 2,  128, { 8,  1}, {1, 1}));
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32,  83, 2,  128, { 8,  1}, {4, 1}));
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32,  64, 2,   64, { 8,  1}, {4, 1}));
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32,  83, 2,   64, { 8,  1}, {4, 1}));
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32,  64, 45, 128, { 8,  1}, {4, 1}));
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 128, 45,  64, { 8,  1}, {4, 1}));
+
     for (ggml_type type_a : all_types) {
         for (ggml_type type_b : {GGML_TYPE_F32 /*, GGML_TYPE_F16 */}) {
             for (int n_mats : {2, 4, 8}) {