| /aosp_15_r20/external/tensorflow/tensorflow/compiler/xla/service/ |
| H A D | service.cc | 171 << "Requested more replicas than there are devices."; in Service()
394 TF_ASSIGN_OR_RETURN(auto replicas, Replicas(*backend, device_handles[i])); in ExecuteParallelAndRegisterResult()
395 CHECK_EQ(replicas.size(), arguments[i].size()); in ExecuteParallelAndRegisterResult()
396 for (int64_t replica = 0, end = replicas.size(); replica < end; ++replica) { in ExecuteParallelAndRegisterResult()
397 device_assignment(replica, i) = replicas[replica]->device_ordinal(); in ExecuteParallelAndRegisterResult()
402 // Stream executors for the replicas of the current computation. in ExecuteParallelAndRegisterResult()
403 TF_ASSIGN_OR_RETURN(auto replicas, Replicas(*backend, device_handles[i])); in ExecuteParallelAndRegisterResult()
404 CHECK_EQ(replicas.size(), arguments[i].size()); in ExecuteParallelAndRegisterResult()
406 const int64_t n = replicas.size(); in ExecuteParallelAndRegisterResult()
410 backend->BorrowStream(replicas[replica])); in ExecuteParallelAndRegisterResult()
[all …]
|
| H A D | hlo_replication_analysis.h | 31 // the same value across replicas or across partitions (depending on the value
43 // replicas or partitions.
54 // the given index across all replicas or partitions.
127 // partitions (not across replicas) on an SPMD partitioned module. This means
133 // If false, HloReplicationAnalysis runs across replicas.
137 // across replicas or partitions. This is provided by the caller as additional
145 // whether the instruction outputs the same value across replicas or
|
| /aosp_15_r20/external/google-cloud-java/java-aiplatform/proto-google-cloud-aiplatform-v1/src/main/java/com/google/cloud/aiplatform/v1/ |
| H A D | AutomaticResources.java | 76 * Immutable. The minimum number of replicas this DeployedModel will be always
78 * deployed onto more replicas up to
80 * and as traffic decreases, some of these extra replicas may be freed. If the
99 * Immutable. The maximum number of replicas this DeployedModel may be
102 * the ability to scale the model to that many replicas is guaranteed (barring
104 * what its replicas at maximum may handle, a portion of the traffic will be
488 * Immutable. The minimum number of replicas this DeployedModel will be always
490 * deployed onto more replicas up to
492 * and as traffic decreases, some of these extra replicas may be freed. If the
508 * Immutable. The minimum number of replicas this DeployedModel will be always
[all …]
|
| H A D | AutomaticResourcesOrBuilder.java | 30 * Immutable. The minimum number of replicas this DeployedModel will be always
32 * deployed onto more replicas up to
34 * and as traffic decreases, some of these extra replicas may be freed. If the
48 * Immutable. The maximum number of replicas this DeployedModel may be
51 * the ability to scale the model to that many replicas is guaranteed (barring
53 * what its replicas at maximum may handle, a portion of the traffic will be
|
| /aosp_15_r20/external/google-cloud-java/java-aiplatform/proto-google-cloud-aiplatform-v1beta1/src/main/java/com/google/cloud/aiplatform/v1beta1/ |
| H A D | AutomaticResources.java | 76 * Immutable. The minimum number of replicas this DeployedModel will be always
78 * deployed onto more replicas up to
80 * and as traffic decreases, some of these extra replicas may be freed. If the
99 * Immutable. The maximum number of replicas this DeployedModel may be
102 * the ability to scale the model to that many replicas is guaranteed (barring
104 * what its replicas at maximum may handle, a portion of the traffic will be
489 * Immutable. The minimum number of replicas this DeployedModel will be always
491 * deployed onto more replicas up to
493 * and as traffic decreases, some of these extra replicas may be freed. If the
509 * Immutable. The minimum number of replicas this DeployedModel will be always
[all …]
|
| H A D | AutomaticResourcesOrBuilder.java | 30 * Immutable. The minimum number of replicas this DeployedModel will be always
32 * deployed onto more replicas up to
34 * and as traffic decreases, some of these extra replicas may be freed. If the
48 * Immutable. The maximum number of replicas this DeployedModel may be
51 * the ability to scale the model to that many replicas is guaranteed (barring
53 * what its replicas at maximum may handle, a portion of the traffic will be
|
| /aosp_15_r20/external/tensorflow/tensorflow/python/distribute/ |
| H A D | distribute_lib.py | 68 worker may contain one or more replicas, but contains at least one
72 replicas on that worker.
76 updates the model variables independently. You may also have replicas
81 `tf.distribute.experimental.ParameterServerStrategy`). All replicas that want
103 the replicas (like `reduce_to()`). By default you start in a _replica context_
121 on all replicas are the same. `Mirrored` values are kept synchronized by the
129 replicas, like `strategy.run(fn, args=[w])` with an
141 from all replicas before applying the update. _All-reduce_ is an algorithm for
182 `all_reduce` method for aggregating across all replicas.
465 Replicas will dequeue from the local Dataset on their worker.
[all …]
|
| H A D | custom_training_loop_input_test.py | 57 Due to the number of replicas in the strategy, the output may have a
372 # on the number of replicas.
506 # This assumes that there are exactly 2 replicas
530 # This assumes that there are exactly 2 replicas
559 # This assumes that there are exactly 2 replicas
582 # This assumes that there are exactly 2 replicas
604 # This assumes that there are exactly 2 replicas
657 # This assumes that there are exactly 2 replicas
682 # This assumes that there are exactly 2 replicas
726 # This assumes that there are exactly 2 replicas
[all …]
|
| H A D | central_storage_strategy.py | 131 replicas, with the provided "per replica" arguments specific to that device.
146 """Reduce `value` across replicas.
149 per-example loss, the batch will be divided across all the replicas. This
150 function allows you to aggregate across replicas and optionally also across
152 replicas, values for examples `[0, 1, 2, 3]` will be on replica 0 and
154 aggregate across replicas, returning `[0+4, 1+5, 2+6, 3+7]`. This is useful
162 that the resulting shape is consistent across replicas. So if the last
203 `None` to only reduce across replicas (e.g. if the tensor has no batch
|
| H A D | input_lib.py | 140 """Returns the next input from the iterator for all replicas.
160 the next input for all replicas.
196 """Returns a `tf.experimental.Optional` that contains the next value for all replicas.
257 number of replicas in sync) and autosharding (i.e. the
495 """Calcualates the number of replicas that have values.
621 """Returns the next input from the iterator for all replicas."""
653 replicas = []
661 # Make `replicas` a flat list of values across all replicas.
662 replicas.extend(self._iterators[i].get_next_as_list(new_name))
663 return _create_per_replica(replicas, self._strategy)
[all …]
|
| /aosp_15_r20/external/google-cloud-java/java-aiplatform/proto-google-cloud-aiplatform-v1/src/main/proto/google/cloud/aiplatform/v1/ |
| H A D | machine_resources.proto | 67 // Required. Immutable. The minimum number of machine replicas this
72 // deployed onto more replicas, and as traffic decreases, some of these extra
73 // replicas may be freed.
79 // Immutable. The maximum number of replicas this DeployedModel may be
82 // the ability to scale the model to that many replicas is guaranteed (barring
84 // what its replicas at maximum may handle, a portion of the traffic will be
126 // Immutable. The minimum number of replicas this DeployedModel will be always
128 // deployed onto more replicas up to
130 // and as traffic decreases, some of these extra replicas may be freed. If the
134 // Immutable. The maximum number of replicas this DeployedModel may be
[all …]
|
| /aosp_15_r20/external/google-cloud-java/java-aiplatform/proto-google-cloud-aiplatform-v1beta1/src/main/proto/google/cloud/aiplatform/v1beta1/ |
| H A D | machine_resources.proto | 67 // Required. Immutable. The minimum number of machine replicas this
72 // deployed onto more replicas, and as traffic decreases, some of these extra
73 // replicas may be freed.
79 // Immutable. The maximum number of replicas this DeployedModel may be
82 // the ability to scale the model to that many replicas is guaranteed (barring
84 // what its replicas at maximum may handle, a portion of the traffic will be
126 // Immutable. The minimum number of replicas this DeployedModel will be always
128 // deployed onto more replicas up to
130 // and as traffic decreases, some of these extra replicas may be freed. If the
134 // Immutable. The maximum number of replicas this DeployedModel may be
[all …]
|
| /aosp_15_r20/external/tensorflow/tensorflow/python/training/ |
| H A D | sync_replicas_optimizer.py | 16 """Synchronize replicas for training."""
35 # Please note that the gradients from replicas are averaged instead of summed
37 # rate according to the number of replicas. This change is introduced to be
52 optimizer avoids stale gradients by collecting gradients from all replicas,
54 which replicas can fetch the new variables and continue.
88 For the replicas:
104 # Wrap the optimizer with sync_replicas_optimizer with 50 replicas: at each
106 # Note that if you want to have 2 backup replicas, you can change
108 # replicas you started in your job.
161 replicas_to_aggregate: number of replicas to aggregate for each variable
[all …]
|
| /aosp_15_r20/external/googleapis/google/cloud/aiplatform/v1/ |
| H A D | machine_resources.proto | 71 // Required. Immutable. The minimum number of machine replicas this
76 // deployed onto more replicas, and as traffic decreases, some of these extra
77 // replicas may be freed.
83 // Immutable. The maximum number of replicas this DeployedModel may be
86 // the ability to scale the model to that many replicas is guaranteed (barring
88 // what its replicas at maximum may handle, a portion of the traffic will be
130 // Immutable. The minimum number of replicas this DeployedModel will be always
132 // deployed onto more replicas up to
134 // and as traffic decreases, some of these extra replicas may be freed. If the
138 // Immutable. The maximum number of replicas this DeployedModel may be
[all …]
|
| /aosp_15_r20/external/googleapis/google/cloud/aiplatform/v1beta1/ |
| H A D | machine_resources.proto | 71 // Required. Immutable. The minimum number of machine replicas this
76 // deployed onto more replicas, and as traffic decreases, some of these extra
77 // replicas may be freed.
83 // Immutable. The maximum number of replicas this DeployedModel may be
86 // the ability to scale the model to that many replicas is guaranteed (barring
88 // what its replicas at maximum may handle, a portion of the traffic will be
130 // Immutable. The minimum number of replicas this DeployedModel will be always
132 // deployed onto more replicas up to
134 // and as traffic decreases, some of these extra replicas may be freed. If the
138 // Immutable. The maximum number of replicas this DeployedModel may be
[all …]
|
| /aosp_15_r20/external/google-cloud-java/java-secretmanager/proto-google-cloud-secretmanager-v1beta1/src/main/java/com/google/cloud/secretmanager/v1beta1/ |
| H A D | Replication.java | 611 * Required. The list of Replicas for this [Secret][google.cloud.secrets.v1beta1.Secret].
616 …* repeated .google.cloud.secrets.v1beta1.Replication.UserManaged.Replica replicas = 1 [(.google.ap…
625 * Required. The list of Replicas for this [Secret][google.cloud.secrets.v1beta1.Secret].
630 …* repeated .google.cloud.secrets.v1beta1.Replication.UserManaged.Replica replicas = 1 [(.google.ap…
638 * Required. The list of Replicas for this [Secret][google.cloud.secrets.v1beta1.Secret].
643 …* repeated .google.cloud.secrets.v1beta1.Replication.UserManaged.Replica replicas = 1 [(.google.ap…
651 * Required. The list of Replicas for this [Secret][google.cloud.secrets.v1beta1.Secret].
656 …* repeated .google.cloud.secrets.v1beta1.Replication.UserManaged.Replica replicas = 1 [(.google.ap…
667 * Required. The list of Replicas for this [Secret][google.cloud.secrets.v1beta1.Secret].
672 …* repeated .google.cloud.secrets.v1beta1.Replication.UserManaged.Replica replicas = 1 [(.google.ap…
[all …]
|
| /aosp_15_r20/external/tensorflow/tensorflow/python/tpu/ |
| H A D | device_assignment.py | 32 """Computes a nested dict which maps task and logical core to replicas."""
49 for core, replicas in core_to_replicas.items():
50 core_to_sorted_replicas[core] = sorted(replicas)
114 """The number of replicas of the computation."""
139 A sorted list of the replicas that are attached to that task and
326 # Form rings for replicas and model-parallel cores.
328 # Form meshes for replicas and/or model-parallel cores.
362 num_replicas: The number of computation replicas to run. The replicas will
377 ValueError: If computation's replicas cannot fit into the TPU topology.
433 "requested {} replicas but only {} replicas with shape {} and "
[all …]
|
| /aosp_15_r20/external/google-cloud-java/java-secretmanager/proto-google-cloud-secretmanager-v1/src/main/java/com/google/cloud/secretmanager/v1/ |
| H A D | Replication.java | 973 * Required. The list of Replicas for this
979 …* repeated .google.cloud.secretmanager.v1.Replication.UserManaged.Replica replicas = 1 [(.google.a…
988 * Required. The list of Replicas for this
994 …* repeated .google.cloud.secretmanager.v1.Replication.UserManaged.Replica replicas = 1 [(.google.a…
1002 * Required. The list of Replicas for this
1008 …* repeated .google.cloud.secretmanager.v1.Replication.UserManaged.Replica replicas = 1 [(.google.a…
1016 * Required. The list of Replicas for this
1022 …* repeated .google.cloud.secretmanager.v1.Replication.UserManaged.Replica replicas = 1 [(.google.a…
1032 * Required. The list of Replicas for this
1038 …* repeated .google.cloud.secretmanager.v1.Replication.UserManaged.Replica replicas = 1 [(.google.a…
[all …]
|
| /aosp_15_r20/external/google-cloud-java/java-gke-multi-cloud/proto-google-cloud-gke-multi-cloud-v1/src/main/java/com/google/cloud/gkemulticloud/v1/ |
| H A D | AzureControlPlane.java | 93 * Required. The Kubernetes version to run on control plane replicas
120 * Required. The Kubernetes version to run on control plane replicas
738 * Optional. Configuration for where to place the control plane replicas.
741 * to the three control plane replicas as evenly as possible.
757 * Optional. Configuration for where to place the control plane replicas.
760 * to the three control plane replicas as evenly as possible.
776 * Optional. Configuration for where to place the control plane replicas.
779 * to the three control plane replicas as evenly as possible.
794 * Optional. Configuration for where to place the control plane replicas.
797 * to the three control plane replicas as evenly as possible.
[all …]
|
| H A D | AzureControlPlaneOrBuilder.java | 30 * Required. The Kubernetes version to run on control plane replicas
46 * Required. The Kubernetes version to run on control plane replicas
463 * Optional. Configuration for where to place the control plane replicas.
466 * to the three control plane replicas as evenly as possible.
478 * Optional. Configuration for where to place the control plane replicas.
481 * to the three control plane replicas as evenly as possible.
493 * Optional. Configuration for where to place the control plane replicas.
496 * to the three control plane replicas as evenly as possible.
508 * Optional. Configuration for where to place the control plane replicas.
511 * to the three control plane replicas as evenly as possible.
[all …]
|
| /aosp_15_r20/external/pytorch/torch/csrc/api/include/torch/nn/parallel/ |
| H A D | data_parallel.h | 108 const std::vector<std::shared_ptr<ModuleType>>& replicas, in replicate_grad_edges() argument
115 autograd::set_history(replicas[i]->parameters_[parameter.key()], grad_fn); in replicate_grad_edges()
125 autograd::set_history(replicas[i]->buffers_[buffer.key()], grad_fn); in replicate_grad_edges()
133 for (auto& replica : replicas) { in replicate_grad_edges()
152 std::vector<std::shared_ptr<ModuleType>> replicas; in replicate() local
153 replicas.reserve(devices.size()); in replicate()
155 replicas.push_back( in replicate()
160 replicate_grad_edges(module, replicas, devices); in replicate()
161 return replicas; in replicate()
288 auto replicas = replicate(module, *devices); variable
[all …]
|
| /aosp_15_r20/external/tensorflow/tensorflow/compiler/mlir/tensorflow/transforms/ |
| H A D | replicate_to_island.cc | 89 // for different replicas. Some ops, such as XlaHostCompute op or TPU Embedding
143 int num_replicas, llvm::SmallVectorImpl<tf_executor::IslandOp>& replicas) { in ExpandReplicateIntoReplicas() argument
144 replicas.reserve(num_replicas); in ExpandReplicateIntoReplicas()
180 replicas.push_back(replica); in ExpandReplicateIntoReplicas()
244 llvm::SmallVector<tf_executor::IslandOp, 8> replicas; in CreateIslandsFromReplicate() local
246 replicate_op, num_replicas, replicas))) in CreateIslandsFromReplicate()
252 for (auto replica_and_idx : llvm::enumerate(replicas)) in CreateIslandsFromReplicate()
263 // Add sink island to pin all replicas as a control dependency if there is a in CreateIslandsFromReplicate()
267 for (auto& replica : replicas) island_operands.push_back(replica.control()); in CreateIslandsFromReplicate()
280 // Replicas with no uses should be pinned to a graph fetch so they still in CreateIslandsFromReplicate()
[all …]
|
| /aosp_15_r20/external/pytorch/test/jit/ |
| H A D | test_data_parallel.py | 81 def check_replicas(self, module, replicas, input_shape=(2, 2)): argument
84 for i, replica in enumerate(replicas):
95 replicas = dp.replicate(module, {0, 1})
96 self.check_replicas(module, replicas)
103 replicas = dp.replicate(module, {0, 1})
104 self.check_replicas(module, replicas)
109 replicas = dp.replicate(module, {0, 1})
110 self.check_replicas(module, replicas)
|
| /aosp_15_r20/external/tensorflow/tensorflow/core/tpu/graph_rewrite/ |
| H A D | distributed_tpu_rewrite_pass.h | 30 // controls how many replicas of the computation to create. Replicas are mostly
31 // independent; replicas can only communicate using the CrossReplicaSum
52 // replicas, the first six arguments to TPUReplicate will be:
56 // If there are two distributed inputs (E, F) and two replicas, the following
60 // replicas receive a copy of each of these inputs.
64 // For example, for a computation with two replicas, three per-replica inputs
218 // replicas).
230 // Indices of mirrored variables across replicas, which should be
259 // * num_replicas: the number of replicas requested
301 // Also verifies that all replicas have identical input shapes for their
[all …]
|
| /aosp_15_r20/external/pytorch/torch/nn/parallel/ |
| H A D | data_parallel.py | 84 is done on the replicas which are destroyed after ``forward``. However,
192 replicas = self.replicate(self.module, self.device_ids[: len(inputs)])
193 outputs = self.parallel_apply(replicas, inputs, module_kwargs)
210 self, replicas: Sequence[T], inputs: Sequence[Any], kwargs: Any
213 replicas, inputs, kwargs, self.device_ids[: len(replicas)]
283 replicas = replicate(module, used_device_ids)
284 outputs = parallel_apply(replicas, inputs, module_kwargs, used_device_ids)
|