[Flax] Add remat (gradient checkpointing) (#17843)

* [Flax] Add remat (gradient checkpointing)

* fix variable naming in test

* flip: checkpoint using a method

* fix naming

* fix class naming

* apply PVP's suggestions from code review

* make fix-copies

* fix big-bird, electra, roberta

* cookie-cutter

* fix flax big-bird

* move test to common

src/transformers/modeling_flax_utils.py

@@ -235,6 +235,9 @@ class FlaxPreTrainedModel(PushToHubMixin, FlaxGenerationMixin):
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> Dict:
         raise NotImplementedError(f"init method has to be implemented for {self}")
 
+    def enable_gradient_checkpointing(self):
+        raise NotImplementedError(f"gradient checkpointing method has to be implemented for {self}")
+
     @classmethod
     def _from_config(cls, config, **kwargs):
         """

src/transformers/models/bert/modeling_flax_bert.py

@@ -23,6 +23,7 @@ import jax
 import jax.numpy as jnp
 from flax.core.frozen_dict import FrozenDict, freeze, unfreeze
 from flax.linen import combine_masks, make_causal_mask
+from flax.linen import partitioning as nn_partitioning
 from flax.linen.attention import dot_product_attention_weights
 from flax.traverse_util import flatten_dict, unflatten_dict
 from jax import lax
@@ -56,6 +57,8 @@ _CHECKPOINT_FOR_DOC = "bert-base-uncased"
 _CONFIG_FOR_DOC = "BertConfig"
 _TOKENIZER_FOR_DOC = "BertTokenizer"
 
+remat = nn_partitioning.remat
+
 
 @flax.struct.dataclass
 class FlaxBertForPreTrainingOutput(ModelOutput):
@@ -544,11 +547,19 @@ class FlaxBertLayer(nn.Module):
 class FlaxBertLayerCollection(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layers = [
-            FlaxBertLayer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
-        ]
+        if self.gradient_checkpointing:
+            FlaxBertCheckpointLayer = remat(FlaxBertLayer, static_argnums=(5, 6, 7))
+            self.layers = [
+                FlaxBertCheckpointLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
+        else:
+            self.layers = [
+                FlaxBertLayer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
+            ]
 
     def __call__(
         self,
@@ -582,12 +593,12 @@ class FlaxBertLayerCollection(nn.Module):
             layer_outputs = layer(
                 hidden_states,
                 attention_mask,
-                layer_head_mask=head_mask[i] if head_mask is not None else None,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                init_cache=init_cache,
-                deterministic=deterministic,
-                output_attentions=output_attentions,
+                head_mask[i] if head_mask is not None else None,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                init_cache,
+                deterministic,
+                output_attentions,
             )
 
             hidden_states = layer_outputs[0]
@@ -617,9 +628,14 @@ class FlaxBertLayerCollection(nn.Module):
 class FlaxBertEncoder(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layer = FlaxBertLayerCollection(self.config, dtype=self.dtype)
+        self.layer = FlaxBertLayerCollection(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
 
     def __call__(
         self,
@@ -756,11 +772,24 @@ class FlaxBertPreTrainedModel(FlaxPreTrainedModel):
         seed: int = 0,
         dtype: jnp.dtype = jnp.float32,
         _do_init: bool = True,
+        gradient_checkpointing: bool = False,
         **kwargs
     ):
-        module = self.module_class(config=config, dtype=dtype, **kwargs)
+        module = self.module_class(
+            config=config,
+            dtype=dtype,
+            gradient_checkpointing=gradient_checkpointing,
+            **kwargs,
+        )
         super().__init__(config, module, input_shape=input_shape, seed=seed, dtype=dtype, _do_init=_do_init)
 
+    def enable_gradient_checkpointing(self):
+        self._module = self.module_class(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=True,
+        )
+
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> FrozenDict:
         # init input tensors
         input_ids = jnp.zeros(input_shape, dtype="i4")
@@ -925,10 +954,15 @@ class FlaxBertModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
     add_pooling_layer: bool = True
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.embeddings = FlaxBertEmbeddings(self.config, dtype=self.dtype)
-        self.encoder = FlaxBertEncoder(self.config, dtype=self.dtype)
+        self.encoder = FlaxBertEncoder(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.pooler = FlaxBertPooler(self.config, dtype=self.dtype)
 
     def __call__(
@@ -1003,9 +1037,14 @@ append_call_sample_docstring(
 class FlaxBertForPreTrainingModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBertPreTrainingHeads(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1099,9 +1138,15 @@ append_replace_return_docstrings(
 class FlaxBertForMaskedLMModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBertOnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1161,9 +1206,14 @@ append_call_sample_docstring(
 class FlaxBertForNextSentencePredictionModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBertOnlyNSPHead(dtype=self.dtype)
 
     def __call__(
@@ -1248,9 +1298,14 @@ append_replace_return_docstrings(
 class FlaxBertForSequenceClassificationModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         classifier_dropout = (
             self.config.classifier_dropout
             if self.config.classifier_dropout is not None
@@ -1324,9 +1379,14 @@ append_call_sample_docstring(
 class FlaxBertForMultipleChoiceModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(1, dtype=self.dtype)
 
@@ -1399,9 +1459,15 @@ append_call_sample_docstring(
 class FlaxBertForTokenClassificationModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         classifier_dropout = (
             self.config.classifier_dropout
             if self.config.classifier_dropout is not None
@@ -1468,9 +1534,15 @@ append_call_sample_docstring(
 class FlaxBertForQuestionAnsweringModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.qa_outputs = nn.Dense(self.config.num_labels, dtype=self.dtype)
 
     def __call__(
@@ -1539,9 +1611,15 @@ append_call_sample_docstring(
 class FlaxBertForCausalLMModule(nn.Module):
     config: BertConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBertModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.bert = FlaxBertModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBertOnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(

src/transformers/models/big_bird/modeling_flax_big_bird.py

@@ -23,6 +23,7 @@ import jax
 import jax.numpy as jnp
 from flax.core.frozen_dict import FrozenDict, freeze, unfreeze
 from flax.linen import combine_masks, make_causal_mask
+from flax.linen import partitioning as nn_partitioning
 from flax.linen.attention import dot_product_attention_weights
 from flax.traverse_util import flatten_dict, unflatten_dict
 from jax import lax
@@ -54,6 +55,8 @@ _CHECKPOINT_FOR_DOC = "google/bigbird-roberta-base"
 _CONFIG_FOR_DOC = "BigBirdConfig"
 _TOKENIZER_FOR_DOC = "BigBirdTokenizer"
 
+remat = nn_partitioning.remat
+
 
 @flax.struct.dataclass
 class FlaxBigBirdForPreTrainingOutput(ModelOutput):
@@ -1368,12 +1371,20 @@ class FlaxBigBirdLayer(nn.Module):
 class FlaxBigBirdLayerCollection(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layers = [
-            FlaxBigBirdLayer(self.config, layer_id=i, name=str(i), dtype=self.dtype)
-            for i in range(self.config.num_hidden_layers)
-        ]
+        if self.gradient_checkpointing:
+            FlaxBigBirdCheckpointLayer = remat(FlaxBigBirdLayer, static_argnums=(5, 6, 7))
+            self.layers = [
+                FlaxBigBirdCheckpointLayer(self.config, layer_id=i, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
+        else:
+            self.layers = [
+                FlaxBigBirdLayer(self.config, layer_id=i, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
 
     # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertLayerCollection.__call__ with Bert->BigBird
     def __call__(
@@ -1408,12 +1419,12 @@ class FlaxBigBirdLayerCollection(nn.Module):
             layer_outputs = layer(
                 hidden_states,
                 attention_mask,
-                layer_head_mask=head_mask[i] if head_mask is not None else None,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                init_cache=init_cache,
-                deterministic=deterministic,
-                output_attentions=output_attentions,
+                head_mask[i] if head_mask is not None else None,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                init_cache,
+                deterministic,
+                output_attentions,
             )
 
             hidden_states = layer_outputs[0]
@@ -1444,9 +1455,14 @@ class FlaxBigBirdLayerCollection(nn.Module):
 class FlaxBigBirdEncoder(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layer = FlaxBigBirdLayerCollection(self.config, dtype=self.dtype)
+        self.layer = FlaxBigBirdLayerCollection(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
 
     def __call__(
         self,
@@ -1559,9 +1575,10 @@ class FlaxBigBirdPreTrainedModel(FlaxPreTrainedModel):
         seed: int = 0,
         dtype: jnp.dtype = jnp.float32,
         _do_init: bool = True,
+        gradient_checkpointing: bool = False,
         **kwargs
     ):
-        module = self.module_class(config=config, dtype=dtype, **kwargs)
+        module = self.module_class(config=config, dtype=dtype, gradient_checkpointing=gradient_checkpointing, **kwargs)
         if config.attention_type == "block_sparse" and input_shape is None:
             input_shape = (1, 12 * config.block_size)
         elif input_shape is None:
@@ -1569,6 +1586,14 @@ class FlaxBigBirdPreTrainedModel(FlaxPreTrainedModel):
 
         super().__init__(config, module, input_shape=input_shape, seed=seed, dtype=dtype, _do_init=_do_init)
 
+    # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.enable_gradient_checkpointing
+    def enable_gradient_checkpointing(self):
+        self._module = self.module_class(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=True,
+        )
+
     # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.init_weights
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> FrozenDict:
         # init input tensors
@@ -1735,10 +1760,13 @@ class FlaxBigBirdModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
     add_pooling_layer: bool = True
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.embeddings = FlaxBigBirdEmbeddings(self.config, dtype=self.dtype)
-        self.encoder = FlaxBigBirdEncoder(self.config, dtype=self.dtype)
+        self.encoder = FlaxBigBirdEncoder(
+            self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.pooler = nn.Dense(
             self.config.hidden_size,
             kernel_init=jax.nn.initializers.normal(self.config.initializer_range),
@@ -1812,9 +1840,14 @@ append_call_sample_docstring(
 class FlaxBigBirdForPreTrainingModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBigBirdModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBigBirdPreTrainingHeads(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1910,9 +1943,15 @@ append_replace_return_docstrings(
 class FlaxBigBirdForMaskedLMModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.bert = FlaxBigBirdModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBigBirdOnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1999,9 +2038,12 @@ class FlaxBigBirdClassificationHead(nn.Module):
 class FlaxBigBirdForSequenceClassificationModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBigBirdModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.classifier = FlaxBigBirdClassificationHead(self.config, dtype=self.dtype)
 
     def __call__(
@@ -2067,9 +2109,14 @@ append_call_sample_docstring(
 class FlaxBigBirdForMultipleChoiceModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, dtype=self.dtype)
+        self.bert = FlaxBigBirdModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(1, dtype=self.dtype)
 
@@ -2162,9 +2209,15 @@ append_call_sample_docstring(
 class FlaxBigBirdForTokenClassificationModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.bert = FlaxBigBirdModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         classifier_dropout = (
             self.config.classifier_dropout
             if self.config.classifier_dropout is not None
@@ -2255,10 +2308,16 @@ class FlaxBigBirdForQuestionAnsweringModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
     add_pooling_layer: bool = False
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.config.num_labels = 2
-        self.bert = FlaxBigBirdModule(self.config, dtype=self.dtype, add_pooling_layer=self.add_pooling_layer)
+        self.bert = FlaxBigBirdModule(
+            self.config,
+            dtype=self.dtype,
+            add_pooling_layer=self.add_pooling_layer,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.qa_classifier = FlaxBigBirdForQuestionAnsweringHead(self.config, dtype=self.dtype)
 
     def __call__(
@@ -2414,9 +2473,15 @@ append_call_sample_docstring(
 class FlaxBigBirdForCausalLMModule(nn.Module):
     config: BigBirdConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.bert = FlaxBigBirdModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.bert = FlaxBigBirdModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.cls = FlaxBigBirdOnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(

src/transformers/models/electra/modeling_flax_electra.py

@@ -23,6 +23,7 @@ import jax
 import jax.numpy as jnp
 from flax.core.frozen_dict import FrozenDict, freeze, unfreeze
 from flax.linen import combine_masks, make_causal_mask
+from flax.linen import partitioning as nn_partitioning
 from flax.linen.attention import dot_product_attention_weights
 from flax.traverse_util import flatten_dict, unflatten_dict
 from jax import lax
@@ -54,6 +55,8 @@ _CHECKPOINT_FOR_DOC = "google/electra-small-discriminator"
 _CONFIG_FOR_DOC = "ElectraConfig"
 _TOKENIZER_FOR_DOC = "ElectraTokenizer"
 
+remat = nn_partitioning.remat
+
 
 @flax.struct.dataclass
 class FlaxElectraForPreTrainingOutput(ModelOutput):
@@ -521,11 +524,20 @@ class FlaxElectraLayer(nn.Module):
 class FlaxElectraLayerCollection(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layers = [
-            FlaxElectraLayer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
-        ]
+        if self.gradient_checkpointing:
+            FlaxElectraCheckpointLayer = remat(FlaxElectraLayer, static_argnums=(5, 6, 7))
+            self.layers = [
+                FlaxElectraCheckpointLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
+        else:
+            self.layers = [
+                FlaxElectraLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
 
     def __call__(
         self,
@@ -559,12 +571,12 @@ class FlaxElectraLayerCollection(nn.Module):
             layer_outputs = layer(
                 hidden_states,
                 attention_mask,
-                layer_head_mask=head_mask[i] if head_mask is not None else None,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                init_cache=init_cache,
-                deterministic=deterministic,
-                output_attentions=output_attentions,
+                head_mask[i] if head_mask is not None else None,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                init_cache,
+                deterministic,
+                output_attentions,
             )
 
             hidden_states = layer_outputs[0]
@@ -595,9 +607,14 @@ class FlaxElectraLayerCollection(nn.Module):
 class FlaxElectraEncoder(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layer = FlaxElectraLayerCollection(self.config, dtype=self.dtype)
+        self.layer = FlaxElectraLayerCollection(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
 
     def __call__(
         self,
@@ -675,11 +692,20 @@ class FlaxElectraPreTrainedModel(FlaxPreTrainedModel):
         seed: int = 0,
         dtype: jnp.dtype = jnp.float32,
         _do_init: bool = True,
+        gradient_checkpointing: bool = False,
         **kwargs
     ):
-        module = self.module_class(config=config, dtype=dtype, **kwargs)
+        module = self.module_class(config=config, dtype=dtype, gradient_checkpointing=gradient_checkpointing, **kwargs)
         super().__init__(config, module, input_shape=input_shape, seed=seed, dtype=dtype, _do_init=_do_init)
 
+    # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.enable_gradient_checkpointing
+    def enable_gradient_checkpointing(self):
+        self._module = self.module_class(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=True,
+        )
+
     # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.init_weights
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> FrozenDict:
         # init input tensors
@@ -845,12 +871,15 @@ class FlaxElectraPreTrainedModel(FlaxPreTrainedModel):
 class FlaxElectraModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.embeddings = FlaxElectraEmbeddings(self.config, dtype=self.dtype)
         if self.config.embedding_size != self.config.hidden_size:
             self.embeddings_project = nn.Dense(self.config.hidden_size, dtype=self.dtype)
-        self.encoder = FlaxElectraEncoder(self.config, dtype=self.dtype)
+        self.encoder = FlaxElectraEncoder(
+            self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
 
     def __call__(
         self,
@@ -925,9 +954,12 @@ class FlaxElectraTiedDense(nn.Module):
 class FlaxElectraForMaskedLMModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.generator_predictions = FlaxElectraGeneratorPredictions(config=self.config, dtype=self.dtype)
         if self.config.tie_word_embeddings:
             self.generator_lm_head = FlaxElectraTiedDense(self.config.vocab_size, dtype=self.dtype)
@@ -989,9 +1021,12 @@ append_call_sample_docstring(
 class FlaxElectraForPreTrainingModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.discriminator_predictions = FlaxElectraDiscriminatorPredictions(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1074,9 +1109,12 @@ append_replace_return_docstrings(
 class FlaxElectraForTokenClassificationModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         classifier_dropout = (
             self.config.classifier_dropout
             if self.config.classifier_dropout is not None
@@ -1218,9 +1256,12 @@ class FlaxElectraSequenceSummary(nn.Module):
 class FlaxElectraForMultipleChoiceModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.sequence_summary = FlaxElectraSequenceSummary(config=self.config, dtype=self.dtype)
         self.classifier = nn.Dense(1, dtype=self.dtype)
 
@@ -1297,9 +1338,12 @@ append_call_sample_docstring(
 class FlaxElectraForQuestionAnsweringModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.qa_outputs = nn.Dense(self.config.num_labels, dtype=self.dtype)
 
     def __call__(
@@ -1392,9 +1436,12 @@ class FlaxElectraClassificationHead(nn.Module):
 class FlaxElectraForSequenceClassificationModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.classifier = FlaxElectraClassificationHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1457,9 +1504,12 @@ append_call_sample_docstring(
 class FlaxElectraForCausalLMModule(nn.Module):
     config: ElectraConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.electra = FlaxElectraModule(config=self.config, dtype=self.dtype)
+        self.electra = FlaxElectraModule(
+            config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing
+        )
         self.generator_predictions = FlaxElectraGeneratorPredictions(config=self.config, dtype=self.dtype)
         if self.config.tie_word_embeddings:
             self.generator_lm_head = FlaxElectraTiedDense(self.config.vocab_size, dtype=self.dtype)

src/transformers/models/roberta/modeling_flax_roberta.py

@@ -21,6 +21,7 @@ import jax
 import jax.numpy as jnp
 from flax.core.frozen_dict import FrozenDict, freeze, unfreeze
 from flax.linen import combine_masks, make_causal_mask
+from flax.linen import partitioning as nn_partitioning
 from flax.linen.attention import dot_product_attention_weights
 from flax.traverse_util import flatten_dict, unflatten_dict
 from jax import lax
@@ -47,6 +48,8 @@ _CHECKPOINT_FOR_DOC = "roberta-base"
 _CONFIG_FOR_DOC = "RobertaConfig"
 _TOKENIZER_FOR_DOC = "RobertaTokenizer"
 
+remat = nn_partitioning.remat
+
 
 def create_position_ids_from_input_ids(input_ids, padding_idx):
     """
@@ -511,11 +514,20 @@ class FlaxRobertaLayer(nn.Module):
 class FlaxRobertaLayerCollection(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layers = [
-            FlaxRobertaLayer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
-        ]
+        if self.gradient_checkpointing:
+            FlaxRobertaCheckpointLayer = remat(FlaxRobertaLayer, static_argnums=(5, 6, 7))
+            self.layers = [
+                FlaxRobertaCheckpointLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
+        else:
+            self.layers = [
+                FlaxRobertaLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
 
     def __call__(
         self,
@@ -549,12 +561,12 @@ class FlaxRobertaLayerCollection(nn.Module):
             layer_outputs = layer(
                 hidden_states,
                 attention_mask,
-                layer_head_mask=head_mask[i] if head_mask is not None else None,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                init_cache=init_cache,
-                deterministic=deterministic,
-                output_attentions=output_attentions,
+                head_mask[i] if head_mask is not None else None,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                init_cache,
+                deterministic,
+                output_attentions,
             )
 
             hidden_states = layer_outputs[0]
@@ -585,9 +597,14 @@ class FlaxRobertaLayerCollection(nn.Module):
 class FlaxRobertaEncoder(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layer = FlaxRobertaLayerCollection(self.config, dtype=self.dtype)
+        self.layer = FlaxRobertaLayerCollection(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
 
     def __call__(
         self,
@@ -719,11 +736,20 @@ class FlaxRobertaPreTrainedModel(FlaxPreTrainedModel):
         seed: int = 0,
         dtype: jnp.dtype = jnp.float32,
         _do_init: bool = True,
+        gradient_checkpointing: bool = False,
         **kwargs
     ):
-        module = self.module_class(config=config, dtype=dtype, **kwargs)
+        module = self.module_class(config=config, dtype=dtype, gradient_checkpointing=gradient_checkpointing, **kwargs)
         super().__init__(config, module, input_shape=input_shape, seed=seed, dtype=dtype, _do_init=_do_init)
 
+    # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.enable_gradient_checkpointing
+    def enable_gradient_checkpointing(self):
+        self._module = self.module_class(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=True,
+        )
+
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> FrozenDict:
         # init input tensors
         input_ids = jnp.zeros(input_shape, dtype="i4")
@@ -889,10 +915,15 @@ class FlaxRobertaModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
     add_pooling_layer: bool = True
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.embeddings = FlaxRobertaEmbeddings(self.config, dtype=self.dtype)
-        self.encoder = FlaxRobertaEncoder(self.config, dtype=self.dtype)
+        self.encoder = FlaxRobertaEncoder(
+            self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.pooler = FlaxRobertaPooler(self.config, dtype=self.dtype)
 
     def __call__(
@@ -967,9 +998,15 @@ append_call_sample_docstring(
 class FlaxRobertaForMaskedLMModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.lm_head = FlaxRobertaLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1034,9 +1071,15 @@ append_call_sample_docstring(
 class FlaxRobertaForSequenceClassificationModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.classifier = FlaxRobertaClassificationHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1101,9 +1144,14 @@ append_call_sample_docstring(
 class FlaxRobertaForMultipleChoiceModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, dtype=self.dtype)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(1, dtype=self.dtype)
 
@@ -1181,9 +1229,15 @@ append_call_sample_docstring(
 class FlaxRobertaForTokenClassificationModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         classifier_dropout = (
             self.config.classifier_dropout
             if self.config.classifier_dropout is not None
@@ -1255,9 +1309,15 @@ append_call_sample_docstring(
 class FlaxRobertaForQuestionAnsweringModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            dtype=self.dtype,
+            add_pooling_layer=False,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.qa_outputs = nn.Dense(self.config.num_labels, dtype=self.dtype)
 
     def __call__(
@@ -1326,9 +1386,15 @@ append_call_sample_docstring(
 class FlaxRobertaForCausalLMModule(nn.Module):
     config: RobertaConfig
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.roberta = FlaxRobertaModule(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.roberta = FlaxRobertaModule(
+            config=self.config,
+            add_pooling_layer=False,
+            dtype=self.dtype,
+            gradient_checkpointing=self.gradient_checkpointing,
+        )
         self.lm_head = FlaxRobertaLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/modeling_flax_{{cookiecutter.lowercase_modelname}}.py

@@ -25,6 +25,7 @@ import jax
 import jax.numpy as jnp
 from flax.core.frozen_dict import FrozenDict, unfreeze, freeze
 from flax.linen import combine_masks, make_causal_mask
+from flax.linen import partitioning as nn_partitioning
 from flax.traverse_util import flatten_dict, unflatten_dict
 from flax.linen.attention import dot_product_attention_weights
 from jax import lax
@@ -126,6 +127,8 @@ _TOKENIZER_FOR_DOC = "{{cookiecutter.camelcase_modelname}}Tokenizer"
 
 """
 
+remat = nn_partitioning.remat
+
 
 
 # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertEmbeddings with Bert->{{cookiecutter.camelcase_modelname}}
@@ -507,11 +510,19 @@ class Flax{{cookiecutter.camelcase_modelname}}Layer(nn.Module):
 class Flax{{cookiecutter.camelcase_modelname}}LayerCollection(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layers = [
-            Flax{{cookiecutter.camelcase_modelname}}Layer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
-        ]
+        if self.gradient_checkpointing:
+            Flax{{cookiecutter.camelcase_modelname}}CheckpointLayer = remat(Flax{{cookiecutter.camelcase_modelname}}Layer, static_argnums=(5, 6, 7))
+            self.layers = [
+                Flax{{cookiecutter.camelcase_modelname}}CheckpointLayer(self.config, name=str(i), dtype=self.dtype)
+                for i in range(self.config.num_hidden_layers)
+            ]
+        else:
+            self.layers = [
+                Flax{{cookiecutter.camelcase_modelname}}Layer(self.config, name=str(i), dtype=self.dtype) for i in range(self.config.num_hidden_layers)
+            ]
 
     def __call__(
         self,
@@ -545,12 +556,12 @@ class Flax{{cookiecutter.camelcase_modelname}}LayerCollection(nn.Module):
             layer_outputs = layer(
                 hidden_states,
                 attention_mask,
-                layer_head_mask=head_mask[i] if head_mask is not None else None,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                init_cache=init_cache,
-                deterministic=deterministic,
-                output_attentions=output_attentions,
+                head_mask[i] if head_mask is not None else None,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                init_cache,
+                deterministic,
+                output_attentions,
             )
 
             hidden_states = layer_outputs[0]
@@ -581,9 +592,10 @@ class Flax{{cookiecutter.camelcase_modelname}}LayerCollection(nn.Module):
 class Flax{{cookiecutter.camelcase_modelname}}Encoder(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.layer = Flax{{cookiecutter.camelcase_modelname}}LayerCollection(self.config, dtype=self.dtype)
+        self.layer = Flax{{cookiecutter.camelcase_modelname}}LayerCollection(self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
 
     def __call__(
         self,
@@ -725,11 +737,20 @@ class Flax{{cookiecutter.camelcase_modelname}}PreTrainedModel(FlaxPreTrainedMode
         seed: int = 0,
         dtype: jnp.dtype = jnp.float32,
         _do_init: bool = True,
+        gradient_checkpointing: bool = False,
         **kwargs
     ):
-        module = self.module_class(config=config, dtype=dtype, **kwargs)
+        module = self.module_class(config=config, dtype=dtype, gradient_checkpointing=gradient_checkpointing, **kwargs)
         super().__init__(config, module, input_shape=input_shape, seed=seed, dtype=dtype, _do_init=_do_init)
 
+    # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.enable_gradient_checkpointing
+    def enable_gradient_checkpointing(self):
+        self._module = self.module_class(
+            config=self.config,
+            dtype=self.dtype,
+            gradient_checkpointing=True,
+        )
+
     # Copied from transformers.models.bert.modeling_flax_bert.FlaxBertPreTrainedModel.init_weights with Bert->{{cookiecutter.camelcase_modelname}}
     def init_weights(self, rng: jax.random.PRNGKey, input_shape: Tuple, params: FrozenDict = None) -> FrozenDict:
         # init input tensors
@@ -897,10 +918,11 @@ class Flax{{cookiecutter.camelcase_modelname}}Module(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32  # the dtype of the computation
     add_pooling_layer: bool = True
+    gradient_checkpointing: bool = False
 
     def setup(self):
         self.embeddings = Flax{{cookiecutter.camelcase_modelname}}Embeddings(self.config, dtype=self.dtype)
-        self.encoder = Flax{{cookiecutter.camelcase_modelname}}Encoder(self.config, dtype=self.dtype)
+        self.encoder = Flax{{cookiecutter.camelcase_modelname}}Encoder(self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.pooler = Flax{{cookiecutter.camelcase_modelname}}Pooler(self.config, dtype=self.dtype)
 
     def __call__(
@@ -969,9 +991,10 @@ class Flax{{cookiecutter.camelcase_modelname}}Model(Flax{{cookiecutter.camelcase
 class Flax{{cookiecutter.camelcase_modelname}}ForMaskedLMModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.cls = Flax{{cookiecutter.camelcase_modelname}}OnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1030,9 +1053,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForCausalLMModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.cls = Flax{{cookiecutter.camelcase_modelname}}OnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(
@@ -1092,9 +1116,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForSequenceClassificationModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(
             self.config.num_labels,
@@ -1163,9 +1188,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForMultipleChoiceModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(1, dtype=self.dtype)
 
@@ -1238,9 +1264,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForTokenClassificationModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, add_pooling_layer=False, gradient_checkpointing=self.gradient_checkpointing)
         self.dropout = nn.Dropout(rate=self.config.hidden_dropout_prob)
         self.classifier = nn.Dense(self.config.num_labels, dtype=self.dtype)
 
@@ -1302,9 +1329,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForQuestionAnsweringModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, add_pooling_layer=False)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, dtype=self.dtype, add_pooling_layer=False, gradient_checkpointing=self.gradient_checkpointing)
         self.qa_outputs = nn.Dense(self.config.num_labels, dtype=self.dtype)
 
     def __call__(
@@ -1373,9 +1401,10 @@ append_call_sample_docstring(
 class Flax{{cookiecutter.camelcase_modelname}}ForCausalLMModule(nn.Module):
     config: {{cookiecutter.camelcase_modelname}}Config
     dtype: jnp.dtype = jnp.float32
+    gradient_checkpointing: bool = False
 
     def setup(self):
-        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype)
+        self.{{cookiecutter.lowercase_modelname}} = Flax{{cookiecutter.camelcase_modelname}}Module(config=self.config, add_pooling_layer=False, dtype=self.dtype, gradient_checkpointing=self.gradient_checkpointing)
         self.cls = Flax{{cookiecutter.camelcase_modelname}}OnlyMLMHead(config=self.config, dtype=self.dtype)
 
     def __call__(

tests/test_modeling_flax_common.py

@@ -1099,6 +1099,33 @@ class FlaxModelTesterMixin:
                 for p1, p2 in zip(flatten_dict(model.params).values(), flatten_dict(new_model.params).values()):
                     self.assertTrue(np.allclose(np.array(p1), np.array(p2)))
 
+    def test_gradient_checkpointing(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            # prepare inputs
+            prepared_inputs_dict = self._prepare_for_class(inputs_dict, model_class)
+            model = model_class(config)
+            remat_model = model_class(config)
+
+            try:
+                remat_model.enable_gradient_checkpointing()
+            except NotImplementedError:
+                continue
+
+            outputs = model(**prepared_inputs_dict)
+            remat_outputs = remat_model(**prepared_inputs_dict)
+
+            # ensure that the dicts of outputs contain the same keys
+            self.assertEqual(outputs.keys(), remat_outputs.keys())
+
+            outputs = outputs.to_tuple()
+            remat_outputs = remat_outputs.to_tuple()
+
+            # ensure that the outputs remain precisely equal
+            for output, remat_output in zip(outputs, remat_outputs):
+                self.assertTrue((output == remat_output).all())
+
 
 @require_flax
 @is_staging_test