fix: compile gemm_batched.cu complete

src/kernels/gemm_batched.cu

@@ -34,16 +34,16 @@ Tensor gemm_batched_fp16(Tensor a,  // FP16 row-major [(... batch ...), M, K]
         LayoutO,
         ElementOutput,
         cutlass::arch::OpClassTensorOp,
-        cutlass::arch::Sm80,
+        cutlass::arch::Gfx928,
         cutlass::gemm::GemmShape<32, 32, 64>,
         cutlass::gemm::GemmShape<32, 32, 64>,
-        cutlass::gemm::GemmShape<16, 8, 16>,
+        cutlass::gemm::GemmShape<16, 16, 16>,
         cutlass::epilogue::thread::LinearCombination<ElementOutput,
                                                      128 / cutlass::sizeof_bits<ElementOutput>::value,
                                                      ElementOutput,
                                                      ElementOutput>,
         cutlass::gemm::threadblock::GemmBatchedIdentityThreadblockSwizzle,
-        2>;
+        1>;
 
     auto sizeA = cutlass::MatrixCoord(M, K);
     auto sizeB = cutlass::MatrixCoord(K, N);