Convert vector to f16 for dequantize mul mat vec (#1913)

* Convert vector to f16 for dmmv

* compile option

* Added compilation option description to README

* Changed cmake CUDA_ARCHITECTURES from "OFF" to "native"

1 files changed, 1 insertions, 1 deletions

diff --git a/llama.cpp b/llama.cpp
index 2105e32..5401db0 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -1620,7 +1620,7 @@ static bool llama_eval_internal(
                     model.layers[il].w1,
                     cur);
             offload_func(cur);
-            ggml_set_name(cur, "result_w2");
+            ggml_set_name(cur, "result_w1");
 
             // SILU activation
             cur = ggml_silu(ctx0, cur);