diff --git a/compiler_opt/rl/local_data_collector.py b/compiler_opt/rl/local_data_collector.py
index 209f57f1..dee5e155 100644
--- a/compiler_opt/rl/local_data_collector.py
+++ b/compiler_opt/rl/local_data_collector.py
@@ -55,7 +55,7 @@ def __init__(
     # We remove this activity from the critical path by running it concurrently
     # with the training phase - i.e. whatever happens between successive data
     # collection calls. Subsequent runs will wait for these to finish.
-    self._reset_workers: concurrent.futures.Future = None
+    self._reset_workers: Optional[concurrent.futures.Future] = None
     self._current_work: List[Tuple[corpus.ModuleSpec, worker.WorkerFuture]] = []
     self._pool = concurrent.futures.ThreadPoolExecutor()
 
diff --git a/compiler_opt/rl/train_locally.py b/compiler_opt/rl/train_locally.py
index 149265bc..3dc0cad5 100644
--- a/compiler_opt/rl/train_locally.py
+++ b/compiler_opt/rl/train_locally.py
@@ -45,7 +45,7 @@
 flags.DEFINE_string('root_dir', os.getenv('TEST_UNDECLARED_OUTPUTS_DIR'),
                     'Root directory for writing logs/summaries/checkpoints.')
 flags.DEFINE_string('data_path', None,
-                    'Path to CNS folder containing IR files.')
+                    'Path to directory containing the corpus.')
 flags.DEFINE_integer(
     'num_workers', None,
     'Number of parallel data collection workers. `None` for max available')
@@ -98,7 +98,7 @@ def train_eval(agent_name=constant.AgentName.PPO,
   }
   saver = policy_saver.PolicySaver(policy_dict=policy_dict)
 
-  logging.info('Loading module specs from corpus.')
+  logging.info('Loading module specs from corpus at %s.', FLAGS.data_path)
   module_specs = corpus.build_modulespecs_from_datapath(
       FLAGS.data_path, problem_config.flags_to_add(),
       problem_config.flags_to_delete())