Polish two error messages (#31852)

* polish two error messages

* polish details

paddle/fluid/operators/detection/polygon_box_transform_op.cu

@@ -45,7 +45,8 @@ class PolygonBoxTransformOpCUDAKernel : public framework::OpKernel<T> {
   void Compute(const framework::ExecutionContext& ctx) const override {
     PADDLE_ENFORCE_EQ(
         platform::is_gpu_place(ctx.GetPlace()), true,
-        platform::errors::InvalidArgument("It must use CUDAPlace."));
+        platform::errors::InvalidArgument(
+            "The polygon_box_transform operator needs to be executed on GPU."));
     auto* in = ctx.Input<Tensor>("Input");
     auto in_dims = in->dims();
     const T* in_data = in->data<T>();

paddle/fluid/operators/matmul_op.cc

@@ -587,7 +587,7 @@ class MatMulOp : public framework::OperatorWithKernel {
     PADDLE_ENFORCE_EQ(mat_dim_x.width_, mat_dim_y.height_,
                       platform::errors::InvalidArgument(
                           "Input X's width should be equal to the Y's height, "
-                          "but received X's shape: [%s],"
+                          "but received X's shape: [%s], "
                           "Y's shape: [%s].",
                           dim_x, dim_y));
 #endif

paddle/fluid/operators/nll_loss_op.h

@@ -36,7 +36,10 @@ static void nll_loss_1D(T* out_data, T* total_weight_data, const T* x_data,
       }
       PADDLE_ENFORCE_EQ(cur_label >= 0 && cur_label < n_classes, true,
                         platform::errors::InvalidArgument(
-                            "label should not be out of bounds."));
+                            "Label value is out of range. "
+                            "Expected label value in range of [0, %d), but "
+                            "received value is %d.",
+                            n_classes, cur_label));
 
       const auto cur_weight =
           weight_data ? weight_data[cur_label] : static_cast<T>(1);