- Sort Score
- Result 10 results
- Languages All
Results 1 - 3 of 3 for CrossReplicaAndPartition (0.2 sec)
-
tensorflow/compiler/mlir/tf2xla/transforms/legalize_tf_collective.cc
} int64_t channel_id = channel_id_++; // FIXME(b/226139061): Mode should be set to CrossReplicaAndPartition // in order to use XLA:GPU for more than one workers. // The mode is set to use CrossReplica to keep the // behavior on the primary user of this optimized path, because // CrossReplicaAndPartition triggers a conflict with the channel_id
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 16K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/tests/legalize-tf-collective.mlir
// CHECK-SAME{LITERAL}: replica_groups = dense<[[0], [1]]> : tensor<2x1xi64> // CHECK: mhlo.add // CHECK: mhlo.return %0 = "tf.XlaAllReduce"(%input, %group_assignment) {reduce_op = "Add", mode = "CrossReplicaAndPartition"} : (tensor<f32>, tensor<2x1xi32>) -> tensor<f32> // CHECK: "mhlo.all_reduce" // CHECK-SAME: channel_handle = #mhlo.channel_handle<handle = 10000, type = 1>
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Sat Apr 06 15:32:52 UTC 2024 - 15.6K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/ir/tf_generated_ops.td
Arg<TF_Int32Tensor, [{Groups between which the reductions are performed.}]>:$group_assignment, TF_AnyStrAttrOf<["Min", "Max", "Mul", "Add", "Mean"]>:$reduce_op, TF_AnyStrAttrOf<["CrossReplica", "CrossReplicaAndPartition"]>:$mode ); let results = (outs TensorOf<[TF_Bfloat16, TF_Float16, TF_Float32, TF_Int32, TF_Uint32]>:$output ); TF_DerivedOperandTypeAttr T = TF_DerivedOperandTypeAttr<0>;
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Jun 11 23:24:08 UTC 2024 - 793K bytes - Viewed (0)