CUDA: fix compilation with GGML_CUDA_F16 (#14837)

diff --git a/ggml/src/ggml-cuda/convert.cu b/ggml/src/ggml-cuda/convert.cu
index 1b4a71bab074c9f8f2e941909becdd9392fc3e60..15c927861f03d65e01ccf01837135d7f772a3286 100644 (file)
--- a/ggml/src/ggml-cuda/convert.cu
+++ b/ggml/src/ggml-cuda/convert.cu
@@ -31,8 +31,8 @@ static __global__ void dequantize_block(const void * __restrict__ vx, dst_t * __
     dequantize_kernel(vx, ib, iqs, v);
 
     const int64_t iy0 = ((i03*ne02 + i02)*ne01 + i01)*ne00 + iybs + iqs;
-    y[iy0 + 0]        = v.x;
-    y[iy0 + y_offset] = v.y;
+    y[iy0 + 0]        = float(v.x);
+    y[iy0 + y_offset] = float(v.y);
 }
 
 template <bool need_check>