Benchmarks: micro benchmarks - change cublasLtMatmulDescCreate scaleType from...

Benchmarks: micro benchmarks - change cublasLtMatmulDescCreate scaleType  from CUDA_R_32F to CUDA_R_16F in FP16 dist inference  (#732)

**Description**
change cublasLtMatmulDescCreate scaleType from CUDA_R_32F to CUDA_R_16F
in FP16 dist inference to fix cublaslt error.

superbench/benchmarks/micro_benchmarks/dist_inference_cpp/dist_inference.cu

@@ -416,8 +416,8 @@ void TestModel(int64_t m, int64_t n, int64_t k, float alpha, float beta, int32_t
     CHECK_CUBLASLT_ERROR(cublasLtMatrixLayoutCreate(&matF, CUDA_R_16F, k, n, k));
     CHECK_CUBLASLT_ERROR(cublasLtMatrixLayoutCreate(&matG, CUDA_R_16F, k, n, k));
 
-    CHECK_CUBLASLT_ERROR(cublasLtMatmulDescCreate(&matmul1, CUBLAS_COMPUTE_16F, CUDA_R_32F));
-    CHECK_CUBLASLT_ERROR(cublasLtMatmulDescCreate(&matmul2, CUBLAS_COMPUTE_16F, CUDA_R_32F));
+    CHECK_CUBLASLT_ERROR(cublasLtMatmulDescCreate(&matmul1, CUBLAS_COMPUTE_16F, CUDA_R_16F));
+    CHECK_CUBLASLT_ERROR(cublasLtMatmulDescCreate(&matmul2, CUBLAS_COMPUTE_16F, CUDA_R_16F));
 
     cublasOperation_t trans = CUBLAS_OP_N;
     CHECK_CUBLASLT_ERROR(cublasLtMatmulDescSetAttribute(matmul1, CUBLASLT_MATMUL_DESC_TRANSA, &trans, sizeof(int32_t)));