[BIG tensor] Adjust tolerance for matmul and bilinear

Author: hxzd5568

tester/base_config.yaml

@@ -44,7 +44,8 @@ special_accuracy_atol_rtol:
   paddle.nn.functional.interpolate: [0.5, 1.5]
   paddle.incubate.nn.functional.fused_bias_dropout_residual_layer_norm: [1.0, 1.0]
   paddle.nn.functional.softmax: [1, 0.01]
-
+  paddle.matmul: [2.0, 0.05]
+  paddle.nn.functional.bilinear: [2.0, 0.08]
 
 # All configs that report dtype diff when not in not_check_dtype list should be
 # copied to tester/api_config/5_accuracy/accuracy_gpu_error_dtype_diff.txt