Patch: v4.30.1

* fix bnb config json serialization

* forward contrib credits from discussions

---------

Co-authored-by: Andrechang <Andrechang@users.noreply.github.com>

setup.py

@@ -428,7 +428,7 @@ install_requires = [
 
 setup(
     name="transformers",
-    version="4.30.0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="4.30.1",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     author="The Hugging Face team (past and future) with the help of all our contributors (https://github.com/huggingface/transformers/graphs/contributors)",
     author_email="transformers@huggingface.co",
     description="State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow",

src/transformers/__init__.py

@@ -18,7 +18,7 @@
 # to defer the actual importing for when the objects are requested. This way `import transformers` provides the names
 # in the namespace without actually importing anything (and especially none of the backends).
 
-__version__ = "4.30.0"
+__version__ = "4.30.1"
 
 from typing import TYPE_CHECKING
 

src/transformers/configuration_utils.py

@@ -784,6 +784,13 @@ class PretrainedConfig(PushToHubMixin):
             ):
                 serializable_config_dict[key] = value
 
+        if hasattr(self, "quantization_config"):
+            serializable_config_dict["quantization_config"] = (
+                self.quantization_config.to_dict()
+                if not isinstance(self.quantization_config, dict)
+                else self.quantization_config
+            )
+
         self.dict_torch_dtype_to_str(serializable_config_dict)
 
         return serializable_config_dict

src/transformers/modeling_tf_utils.py

@@ -81,7 +81,7 @@ elif parse(tf.__version__).minor >= 11:
     from keras.engine.keras_tensor import KerasTensor
 else:
     from tensorflow.python.keras import backend as K
-    from tensorflow.python.keras.engine import call_context
+    from tensorflow.python.keras.engine.base_layer_utils import call_context
     from tensorflow.python.keras.engine.keras_tensor import KerasTensor
 
 
@@ -1156,8 +1156,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         if self.built or call_context().in_call:
             self.built = True
         else:
-            self(self.dummy_inputs, training=False)
             self.built = True
+            self(self.dummy_inputs, training=False)
 
     def __init__(self, config, *inputs, **kwargs):
         super().__init__(*inputs, **kwargs)

src/transformers/trainer.py

@@ -1749,7 +1749,16 @@ class Trainer:
 
         # prepare using `accelerator` prepare
         if use_accelerator_prepare:
-            model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)
+            if hasattr(self.lr_scheduler, "step"):
+                if self.use_apex:
+                    model = self.accelerator.prepare(self.model)
+                else:
+                    model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)
+            else:
+                # to handle cases wherein we pass "DummyScheduler" such as when it is specified in DeepSpeed config.
+                model, self.optimizer, self.lr_scheduler = self.accelerator.prepare(
+                    self.model, self.optimizer, self.lr_scheduler
+                )
 
         if self.is_fsdp_enabled:
             self.model = model
@@ -2841,6 +2850,7 @@ class Trainer:
             or self.is_fsdp_enabled
         ):
             if self.is_fsdp_enabled:
+                os.makedirs(output_dir, exist_ok=True)
                 self.accelerator.state.fsdp_plugin.save_model(self.accelerator, self.model, output_dir)
             else:
                 state_dict = self.model.state_dict()

tests/bitsandbytes/test_4bit.py

@@ -111,6 +111,19 @@ class Bnb4BitTest(Base4bitTest):
         gc.collect()
         torch.cuda.empty_cache()
 
+    def test_quantization_config_json_serialization(self):
+        r"""
+        A simple test to check if the quantization config is correctly serialized and deserialized
+        """
+        config = self.model_4bit.config
+
+        self.assertTrue(hasattr(config, "quantization_config"))
+
+        _ = config.to_dict()
+        _ = config.to_diff_dict()
+
+        _ = config.to_json_string()
+
     def test_memory_footprint(self):
         r"""
         A simple test to check if the model conversion has been done correctly by checking on the

tests/bitsandbytes/test_mixed_int8.py

@@ -118,6 +118,19 @@ class MixedInt8Test(BaseMixedInt8Test):
         gc.collect()
         torch.cuda.empty_cache()
 
+    def test_quantization_config_json_serialization(self):
+        r"""
+        A simple test to check if the quantization config is correctly serialized and deserialized
+        """
+        config = self.model_8bit.config
+
+        self.assertTrue(hasattr(config, "quantization_config"))
+
+        _ = config.to_dict()
+        _ = config.to_diff_dict()
+
+        _ = config.to_json_string()
+
     def test_memory_footprint(self):
         r"""
         A simple test to check if the model conversion has been done correctly by checking on the