monkey-patch transformers to simplify monkey-patching modeling code (#1877)

* monkey-patch transformers so that monkey-patched modeling code doesnt get overwritten

* unnecessary now

* add comment

src/axolotl/utils/models.py

@@ -43,6 +43,9 @@ from axolotl.monkeypatch.multipack import (
     SUPPORTED_MULTIPACK_MODEL_TYPES,
     patch_for_multipack,
 )
+from axolotl.monkeypatch.transformers_dynamic_module_utils import (
+    patch_transformers_dynamic_module_utils,
+)
 from axolotl.prompt_tokenizers import LLAMA_DEFAULT_EOS_TOKEN
 from axolotl.utils.bench import log_gpu_memory_usage
 from axolotl.utils.chat_templates import chat_templates
@@ -54,6 +57,8 @@ from axolotl.utils.model_shard_quant import load_sharded_model, load_sharded_mod
 
 LOG = logging.getLogger("axolotl")
 
+patch_transformers_dynamic_module_utils()
+
 
 # copied from accelerator.FullyShardedDataParallelPlugin
 def get_module_class_from_name(module, name):