1 files changed, 5 insertions, 0 deletions
diff --git a/ggml-cuda/quantize.cuh b/ggml-cuda/quantize.cuh
new file mode 100644
index 00000000..adb89c83
--- /dev/null
+++ b/ggml-cuda/quantize.cuh
@@ -0,0 +1,5 @@
+#include "common.cuh"
+
+#define CUDA_QUANTIZE_BLOCK_SIZE 256
+
+void quantize_row_q8_1_cuda(const float * x, void * vy, const int kx, const int ky, const int kx_padded, cudaStream_t stream);