diff --git a/common/common.cpp b/common/common.cpp
index 5f10718ec..85c8292e4 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 5fb2cb603..7aaf63ceb 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();
diff --git a/ggml-metal.m b/ggml-metal.m
index e9598ddff..109e5fe6b 100644
--- a/ggml-metal.m
+++ b/ggml-metal.m
@@ -1383,7 +1383,7 @@ static enum ggml_status ggml_metal_graph_compute(
                             !ggml_is_transposed(src0) &&
                             !ggml_is_transposed(src1) &&
                             src1t == GGML_TYPE_F32 &&
-                            ne00 % 32 == 0 && ne00 >= 128 &&
+                            ne00 % 32 == 0 && ne00 >= 64 &&
                             (ne11 > ne11_mm_min || (ggml_is_quantized(src0t) && ne12 > 1))) {
                             //printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
 
@@ -1698,7 +1698,7 @@ static enum ggml_status ggml_metal_graph_compute(
                         //       indirect matrix multiplication
                         // !!!
                         if ([ctx->device supportsFamily:MTLGPUFamilyApple7] &&
-                            ne20 % 32 == 0 && ne20 >= 128 &&
+                            ne20 % 32 == 0 && ne20 >= 64 &&
                             ne11 > ne11_mm_min) {
 
                             id<MTLComputePipelineState> pipeline = nil;
diff --git a/llama.cpp b/llama.cpp
index 1a9fe0c4d..9de4a8602 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);