[Trainer] Correct behavior of _load_best_model for PEFT models (#24103)

* v1

* some refactor

- add ST format as well

* fix

* add `ADAPTER_WEIGHTS_NAME` & `ADAPTER_SAFE_WEIGHTS_NAME`

Author: younesbelkada

src/transformers/trainer.py

@@ -134,6 +134,8 @@
 )
 from .training_args import OptimizerNames, ParallelMode, TrainingArguments
 from .utils import (
+    ADAPTER_SAFE_WEIGHTS_NAME,
+    ADAPTER_WEIGHTS_NAME,
     CONFIG_NAME,
     SAFE_WEIGHTS_INDEX_NAME,
     SAFE_WEIGHTS_NAME,
@@ -2177,11 +2179,20 @@ def _load_best_model(self):
         logger.info(f"Loading best model from {self.state.best_model_checkpoint} (score: {self.state.best_metric}).")
         best_model_path = os.path.join(self.state.best_model_checkpoint, WEIGHTS_NAME)
         best_safe_model_path = os.path.join(self.state.best_model_checkpoint, SAFE_WEIGHTS_NAME)
+        best_adapter_model_path = os.path.join(self.state.best_model_checkpoint, ADAPTER_WEIGHTS_NAME)
+        best_safe_adapter_model_path = os.path.join(self.state.best_model_checkpoint, ADAPTER_SAFE_WEIGHTS_NAME)
+
         model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model
-        if os.path.exists(best_model_path) or os.path.exists(best_safe_model_path):
+        if (
+            os.path.exists(best_model_path)
+            or os.path.exists(best_safe_model_path)
+            or os.path.exists(best_adapter_model_path)
+            or os.path.exists(best_safe_adapter_model_path)
+        ):
             if self.is_deepspeed_enabled:
                 deepspeed_load_checkpoint(self.model_wrapped, self.state.best_model_checkpoint)
             else:
+                has_been_loaded = True
                 if is_sagemaker_mp_enabled():
                     if os.path.isfile(os.path.join(self.state.best_model_checkpoint, "user_content.pt")):
                         # If the 'user_content.pt' file exists, load with the new smp api.
@@ -2207,10 +2218,10 @@ def _load_best_model(self):
                         self.accelerator, model, self.state.best_model_checkpoint
                     )
                 else:
-                    if hasattr(model, "base_model") and getattr(model.base_model, "is_8bit_serializable", False):
-                        # If train base_8_bit_models using PEFT & LoRA, assume that adapter have been saved properly.
+                    if is_peft_available() and isinstance(model, PeftModel):
+                        # If train a model using PEFT & LoRA, assume that adapter have been saved properly.
                         if hasattr(model, "active_adapter") and hasattr(model, "load_adapter"):
-                            if os.path.exists(os.path.join(self.state.best_model_checkpoint, "adapter_model.bin")):
+                            if os.path.exists(best_adapter_model_path) or os.path.exists(best_safe_adapter_model_path):
                                 model.load_adapter(self.state.best_model_checkpoint, model.active_adapter)
                                 # Load_adapter has no return value present, modify it when appropriate.
                                 from torch.nn.modules.module import _IncompatibleKeys
@@ -2219,12 +2230,13 @@ def _load_best_model(self):
                             else:
                                 logger.warning(
                                     "The intermediate checkpoints of PEFT may not be saved correctly, "
-                                    "using `TrainerCallback` to save adapter_model.bin in corresponding folders, "
+                                    f"using `TrainerCallback` to save {ADAPTER_WEIGHTS_NAME} in corresponding folders, "
                                     "here are some examples https://github.com./huggingface/peft/issues/96"
                                 )
+                                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.
                         if self.args.save_safetensors and os.path.isfile(best_safe_model_path):
@@ -2236,7 +2248,7 @@ def _load_best_model(self):
                         # workaround for FSDP bug https://github.com./pytorch/pytorch/issues/82963
                         # which takes *args instead of **kwargs
                         load_result = model.load_state_dict(state_dict, False)
-                if not is_sagemaker_mp_enabled():
+                if not is_sagemaker_mp_enabled() and has_been_loaded:
                     self._issue_warnings_after_load(load_result)
         elif os.path.exists(os.path.join(self.state.best_model_checkpoint, WEIGHTS_INDEX_NAME)):
             load_result = load_sharded_checkpoint(

src/transformers/utils/__init__.py

@@ -177,6 +177,8 @@
 
 WEIGHTS_NAME = "pytorch_model.bin"
 WEIGHTS_INDEX_NAME = "pytorch_model.bin.index.json"
+ADAPTER_WEIGHTS_NAME = "adapter_model.bin"
+ADAPTER_SAFE_WEIGHTS_NAME = "adapter_model.safetensors"
 TF2_WEIGHTS_NAME = "tf_model.h5"
 TF2_WEIGHTS_INDEX_NAME = "tf_model.h5.index.json"
 TF_WEIGHTS_NAME = "model.ckpt"