apply suggestions from gemini

Author: chenyushuo

tests/trainer/trainer_test.py

@@ -709,13 +709,13 @@ def test_trainer(self):
                 try:
                     with open(state_dict_iteration_file, "r") as f:
                         state_dict_iteration = int(f.read().strip())
-                except Exception:
+                except (IOError, ValueError):
                     pass
             if os.path.exists(checkpoint_iteration_file):
                 try:
                     with open(checkpoint_iteration_file, "r") as f:
                         checkpoint_iteration = int(f.read().strip())
-                except Exception:
+                except (IOError, ValueError):
                     pass
 
             if state_dict_iteration > 0:
@@ -777,7 +777,7 @@ def test_trainer(self):
                             "special_tokens_map.json",
                         },
                     )
-                print(f"Checkpoint check at {state_dict_iteration} iteration passed.")
+                print(f"Checkpoint check at {checkpoint_iteration} iteration passed.")
 
             time.sleep(1)
         trainer_process.join()

trinity/manager/synchronizer.py

@@ -85,7 +85,8 @@ async def _find_latest_state_dict(self) -> None:
                 try:
                     with open(local_latest_state_dict_iteration, "r") as f:
                         latest_model_version = int(f.read().strip())
-                except Exception:
+                except (IOError, ValueError) as e:
+                    self.logger.warning(f"Failed to read or parse state dict iteration file: {e}")
                     continue
                 if latest_model_version > self.model_version:
                     self.logger.info(
@@ -98,7 +99,7 @@ async def _find_latest_state_dict(self) -> None:
                         self.config.trainer,
                     )
                     self.logger.info(
-                        f"Synchronizer has loaded model state dict from checkpoint {self.model_version}."
+                        f"Synchronizer has loaded model state dict from checkpoint {latest_model_version}."
                     )
                     await self.set_model_state_dict(model_state_dict, latest_model_version)
             await asyncio.sleep(1)