fix

huggingface · Jun 8, 2023 · bf31c5e · bf31c5e
1 parent 44be183
commit bf31c5e
Showing 1 changed file with 1 addition and 2 deletions.
diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
@@ -2233,8 +2233,7 @@ def _load_best_model(self):
                                 )
                                 has_been_loaded = False
                         else:
-                            # We can't do pure 8bit training using transformers.
-                            logger.warning("Could not loading a quantized checkpoint.")
+                            logger.warning("Could not load adapter model, make sure to have `peft>=0.3.0` installed")
                             has_been_loaded = False
                     else:
                         # We load the model state dict on the CPU to avoid an OOM error.