metal : pad n_ctx by 32 (#6177)

* metal : require ne00 >= 128 for mat-mat kernels

ggml-ci

* llama : pad n_ctx by 32

ggml-ci

1 files changed, 3 insertions, 0 deletions

diff --git a/llama.cpp b/llama.cpp
index 1a9fe0c4..9de4a860 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);