bugfix: propage weight key_mapping to peft to fix 3.52 VLM renaming (#38627)

* propage key mapping to peft

* propage key mapping to peft

* make requested changes

* revert

src/transformers/integrations/peft.py

@@ -14,6 +14,7 @@
 
 import importlib
 import inspect
+import re
 import warnings
 from typing import Any, Dict, List, Optional, Union
 
@@ -149,6 +150,7 @@ class PeftAdapterMixin:
 
         # peft only supports low_cpu_mem_usage starting from v0.13.0
         peft_load_kwargs = {}
+        key_mapping = adapter_kwargs.pop("key_mapping", None) if adapter_kwargs is not None else None
         if low_cpu_mem_usage:
             min_version_lcmu = "0.13.0"
             if version.parse(importlib.metadata.version("peft")) >= version.parse(min_version_lcmu):
@@ -233,6 +235,13 @@ class PeftAdapterMixin:
                 new_key = key[len(prefix) :]
             else:
                 new_key = key
+
+            if key_mapping:
+                for pattern, replacement in key_mapping.items():
+                    new_key, n_replace = re.subn(pattern, replacement, new_key)
+                    # Early exit of the loop
+                    if n_replace > 0:
+                        break
             processed_adapter_state_dict[new_key] = value
 
         # Load state dict

src/transformers/modeling_utils.py

@@ -4778,6 +4778,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, PushToHubMixin, PeftAdapterMi
             model.hf_quantizer = hf_quantizer
 
         if _adapter_model_path is not None:
+            adapter_kwargs["key_mapping"] = key_mapping
             model.load_adapter(
                 _adapter_model_path,
                 adapter_name=adapter_name,