minor fix (#1007)

Author: gaoning777

content/docs/pipelines/sdk/gcp/preemptible.md

@@ -254,7 +254,7 @@ For example:
       flip = FlipCoinOp().set_gpu_limit(1, 'nvidia').apply(gcp.use_preemptible_nodepool())
     if __name__ == '__main__':
       import kfp.compiler as compiler
-      compiler.Compiler().compile(flipcoin, __file__ + '.
+      compiler.Compiler().compile(flipcoin, __file__ + '.zip')
 
 ## Comparison with Cloud AI Platform Training service
 
@@ -311,4 +311,4 @@ running preemptible VMs or GPUs:
 ## Next steps
 
 * Explore further options for [customizing Kubeflow on GCP](/docs/gke/).
-* See how to [build pipelines with the SDK](/docs/pipelines/sdk/).
+* See how to [build pipelines with the SDK](/docs/pipelines/sdk/).