Refactor(JAX): Use jax.jit's out_shardings instead of _enforce_jax_state_sharding

keras/src/backend/jax/trainer.py

@@ -153,7 +153,7 @@ def train_step(self, state, data):
             metrics_variables, unscaled_loss, x, y, y_pred, sample_weight
         )
 
-        state = self._enforce_jax_state_sharding(
+        state = (
             trainable_variables,
             non_trainable_variables,
             optimizer_variables,
@@ -185,17 +185,6 @@ def test_step(self, state, data):
             metrics_variables, unscaled_loss, x, y, y_pred, sample_weight
         )
 
-        (
-            trainable_variables,
-            non_trainable_variables,
-            _,
-            metrics_variables,
-        ) = self._enforce_jax_state_sharding(
-            trainable_variables=trainable_variables,
-            non_trainable_variables=non_trainable_variables,
-            optimizer_variables=None,
-            metrics_variables=metrics_variables,
-        )
         state = (
             trainable_variables,
             non_trainable_variables,
@@ -213,17 +202,6 @@ def predict_step(self, state, data):
         outputs, non_trainable_variables = self.stateless_call(
             trainable_variables, non_trainable_variables, x, **kwargs
         )
-        (
-            _,
-            non_trainable_variables,
-            _,
-            _,
-        ) = self._enforce_jax_state_sharding(
-            trainable_variables=None,
-            non_trainable_variables=non_trainable_variables,
-            optimizer_variables=None,
-            metrics_variables=None,
-        )
         return outputs, non_trainable_variables
 
     def _make_function(self, step_function, concatenate_outputs=False):
@@ -281,11 +259,15 @@ def make_train_function(self, force=False):
         if self.train_function is not None and not force:
             return
         if not self.run_eagerly and self.jit_compile:
-            # Note that we mark the state to be donated to jax,
-            # so that jax will reuse the memory buffer for outputs.
-            # This will reduce the memory usage of the training function by
-            # half.
-            train_step = jit(self.train_step, donate_argnums=0)
+            out_shardings = None
+            if distribution_lib.distribution() is not None:
+                state_shardings = self._record_training_state_sharding_spec()
+                out_shardings = (None, state_shardings)
+            train_step = jit(
+                self.train_step,
+                donate_argnums=0,
+                out_shardings=out_shardings,
+            )
         else:
             train_step = self.train_step
 
@@ -297,12 +279,25 @@ def make_test_function(self, force=False):
         if self.test_function is not None and not force:
             return
         if not self.run_eagerly and self.jit_compile:
-            # Note that we mark the state to be donated to jax,
-            # so that jax will reuse the memory buffer for outputs.
-            # This will reduce the memory usage of the training function by
-            # half.
-            test_step = jit(self.test_step, donate_argnums=0)
-
+            out_shardings = None
+            if distribution_lib.distribution() is not None:
+                (
+                    trainable_shardings,
+                    non_trainable_shardings,
+                    _,  # optimizer_shardings
+                    metrics_shardings,
+                ) = self._record_training_state_sharding_spec()
+                state_shardings = (
+                    trainable_shardings,
+                    non_trainable_shardings,
+                    metrics_shardings,
+                )
+                out_shardings = (None, state_shardings)
+            test_step = jit(
+                self.test_step,
+                donate_argnums=0,
+                out_shardings=out_shardings,
+            )
         else:
             test_step = self.test_step
 
@@ -319,7 +314,24 @@ def predict_step(state, data):
             return outputs, (state[0], non_trainable_variables)
 
         if not self.run_eagerly and self.jit_compile:
-            predict_step = jit(predict_step, donate_argnums=0)
+            out_shardings = None
+            if distribution_lib.distribution() is not None:
+                (
+                    trainable_shardings,
+                    non_trainable_shardings,
+                    _,  # optimizer_shardings
+                    _,  # metrics_shardings
+                ) = self._record_training_state_sharding_spec()
+                state_shardings = (
+                    trainable_shardings,
+                    non_trainable_shardings,
+                )
+                out_shardings = (None, state_shardings)
+            predict_step = jit(
+                predict_step,
+                donate_argnums=0,
+                out_shardings=out_shardings,
+            )
 
         _step_function = self._make_function(
             predict_step, concatenate_outputs=True
@@ -402,7 +414,6 @@ def fit(
                 steps=epoch_iterator.num_batches,
                 model=self,
             )
-        self._record_training_state_sharding_spec()
 
         self.make_train_function()
         self.stop_training = False
@@ -518,7 +529,6 @@ def fit(
             if training_finished:
                 callbacks.on_train_end(logs=training_logs)
             self._jax_state = None
-            self._clear_jax_state_sharding()
         return self.history
 
     @traceback_utils.filter_traceback
@@ -568,7 +578,6 @@ def evaluate(
                 steps=epoch_iterator.num_batches,
                 model=self,
             )
-        self._record_training_state_sharding_spec()
 
         self.make_test_function()
         self.stop_evaluating = False
@@ -620,9 +629,6 @@ def evaluate(
         logs = self._get_metrics_result_or_logs(logs)
         callbacks.on_test_end(logs)
         self._jax_state = None
-        if not use_cached_eval_dataset:
-            # Only clear sharding if evaluate is not called from `fit`.
-            self._clear_jax_state_sharding()
         if return_dict:
             return logs
         return self._flatten_metrics_in_order(logs)
@@ -664,7 +670,6 @@ def predict(
                 steps=epoch_iterator.num_batches,
                 model=self,
             )
-        self._record_training_state_sharding_spec()
 
         self.make_predict_function()
         self.stop_predicting = False
@@ -723,7 +728,6 @@ def append_to_outputs(batch_outputs, outputs):
         self.jax_state_sync()
         callbacks.on_predict_end()
         self._jax_state = None
-        self._clear_jax_state_sharding()
         return tree.map_structure_up_to(batch_outputs, np.concatenate, outputs)
 
     def train_on_batch(
@@ -752,7 +756,6 @@ def data():
 
         # Maybe build model
         self._symbolic_build(data_batch=next(data()))
-        self._record_training_state_sharding_spec()
         self.make_train_function()
 
         # Train step
@@ -801,7 +804,6 @@ def data():
 
         # Maybe build model
         self._symbolic_build(data_batch=next(data()))
-        self._record_training_state_sharding_spec()
         self.make_test_function()
 
         # Test step
@@ -834,7 +836,6 @@ def predict_on_batch(self, x):
             # Build model
             with backend.StatelessScope():
                 self(x)
-        self._record_training_state_sharding_spec()
         self.make_predict_function()
 
         state = self._get_jax_state(
@@ -901,60 +902,6 @@ def _record_training_state_sharding_spec(self):
             v.value.sharding for v in self.metrics_variables
         ]
 
-    def _clear_jax_state_sharding(self):
-        self._trainable_variable_shardings = None
-        self._non_trainable_variable_shardings = None
-        self._optimizer_variable_shardings = None
-        self._metrics_variable_shardings = None
-
-    def _enforce_jax_state_sharding(
-        self,
-        trainable_variables=None,
-        non_trainable_variables=None,
-        optimizer_variables=None,
-        metrics_variables=None,
-    ):
-        """Enforce the sharding spec constraint for all the training state.
-
-        Since the output of the train/eval step will be used as inputs to next
-        step, we need to ensure that they have the same sharding spec, so that
-        nnx.jit/jax.jit won't have to recompile the train/eval function.
-
-        Note that this function will also rely on the recorded sharding spec
-        for each of states.
-
-        This function is expected to be called within the jitted train/eval
-        function, especially around the end of the function.
-        """
-        trainable_variables = trainable_variables or []
-        non_trainable_variables = non_trainable_variables or []
-        optimizer_variables = optimizer_variables or []
-        metrics_variables = metrics_variables or []
-
-        for i in range(len(trainable_variables)):
-            trainable_variables[i] = jax.lax.with_sharding_constraint(
-                trainable_variables[i], self._trainable_variable_shardings[i]
-            )
-        for i in range(len(non_trainable_variables)):
-            non_trainable_variables[i] = jax.lax.with_sharding_constraint(
-                non_trainable_variables[i],
-                self._non_trainable_variable_shardings[i],
-            )
-        for i in range(len(optimizer_variables)):
-            optimizer_variables[i] = jax.lax.with_sharding_constraint(
-                optimizer_variables[i], self._optimizer_variable_shardings[i]
-            )
-        for i in range(len(metrics_variables)):
-            metrics_variables[i] = jax.lax.with_sharding_constraint(
-                metrics_variables[i], self._metrics_variable_shardings[i]
-            )
-        return (
-            trainable_variables,
-            non_trainable_variables,
-            optimizer_variables,
-            metrics_variables,
-        )
-
     def _purge_model_variables(
         self,
         trainable_variables=False,

keras/src/trainers/trainer_test.py

@@ -1,5 +1,7 @@
+import types
 from unittest import mock
 
+import jax
 import numpy as np
 import pytest
 from absl.testing import parameterized
@@ -16,13 +18,16 @@
 from keras.src import testing
 from keras.src.backend import config
 from keras.src.backend.common.symbolic_scope import in_symbolic_scope
+from keras.src.backend.jax import trainer as jax_trainer_lib
 from keras.src.callbacks.callback import Callback
 from keras.src.optimizers.rmsprop import RMSprop
 from keras.src.testing.test_utils import named_product
 from keras.src.trainers.data_adapters import py_dataset_adapter
 
 if backend.backend() == "jax":
     from keras.src.backend.jax.trainer import JAXTrainer as Trainer
+    from keras.src.distribution import DataParallel
+    from keras.src.distribution import DeviceMesh
 elif backend.backend() == "torch":
     from keras.src.backend.torch.trainer import TorchTrainer as Trainer
 elif backend.backend() == "tensorflow":
@@ -2857,3 +2862,69 @@ def predict_step(self, *args):
             verbose=0,
         )
         self.assertLessEqual(tracing_count[0], 2)
+
+    @parameterized.named_parameters(
+        ("single_device", False),
+        ("distributed", True),
+    )
+    def test_jit_fit_with_out_shardings_logic(self, distributed):
+        def patched_get_jax_state(
+            self_model,
+            trainable_variables=False,
+            non_trainable_variables=False,
+            optimizer_variables=False,
+            metrics_variables=False,
+            purge_model_variables=False,
+        ):
+            state = []
+            if trainable_variables:
+                state.append([v.value for v in self_model.trainable_variables])
+            if non_trainable_variables:
+                state.append(
+                    [v.value for v in self_model.non_trainable_variables]
+                )
+            if optimizer_variables:
+                state.append([v.value for v in self_model.optimizer.variables])
+            if metrics_variables:
+                state.append([v.value for v in self_model.metrics_variables])
+            return tuple(state)
+
+        original_train_step = jax_trainer_lib.JAXTrainer.train_step
+
+        def patched_train_step_wrapper(self, state, data):
+            logs, new_state = original_train_step(self, state, data)
+            return logs, new_state
+
+        x = np.random.rand(64, 8).astype("float32")
+        y = np.random.rand(64, 1).astype("float32")
+
+        if distributed:
+            if len(jax.local_devices()) < 2:
+                self.skipTest(
+                    "Distributed test requires at least 2 JAX devices."
+                )
+
+            devices = jax.local_devices()
+            mesh = DeviceMesh(
+                shape=(len(devices),), axis_names=("batch",), devices=devices
+            )
+            distribution = DataParallel(mesh)
+
+            with (
+                mock.patch(
+                    "keras.src.backend.jax.trainer.JAXTrainer.train_step",
+                    new=patched_train_step_wrapper,
+                ),
+                distribution.scope(),
+            ):
+                model = models.Sequential(
+                    [
+                        layers.Dense(4, activation="relu", input_shape=(8,)),
+                        layers.Dense(1),
+                    ]
+                )
+                model._get_jax_state = types.MethodType(
+                    patched_get_jax_state, model
+                )
+                model.compile(optimizer="adam", loss="mse", jit_compile=True)
+                model.fit(x, y, epochs=2, batch_size=32, verbose=0)