- Sort Score
- Result 10 results
- Languages All
Results 21 - 30 of 54 for tpu0 (0.15 sec)
-
tensorflow/compiler/mlir/quantization/tensorflow/passes/passes.h
// Creates a pass that converts TPU models for CPU by removing TPU related ops // such as TPUPartitionedCall, TPUReplicatedOp, etc. The TF quantizer does not // work with models specifically designed for TPU, so this pass makes the input // TPU model compatible with the TF quantizer by rewriting the TPU ops. The // output model of this pass is expected to be ready for the TF quantizer.
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri May 10 04:07:09 UTC 2024 - 12.3K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/api/v2/legalize_tf_test.cc
namespace tensorflow { namespace tf2xla { namespace v2 { using ::tensorflow::monitoring::testing::CellReader; using ::testing::Not; using ::testing::TestWithParam; using tpu::FunctionToHloArgs; using tpu::MlirToHloArgs; using tpu::ShardingAndIndex; using tpu::TPUCompileMetadataProto; static constexpr char kCompilationTimeStreamzName[] = "/tensorflow/core/tf2xla/api/v2/phase2_compilation_time";
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Jun 13 23:59:33 UTC 2024 - 16.1K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/host_runtime/tpu_variable_runtime_reformatting.cc
}); if (replicate) reshard_was_inserted |= HandleReplicateOp(while_op, replicate); }); if (reshard_was_inserted) VLOG(1) << "tf-tpu-variable-runtime-reformatting inserted at least one " "TPUReshardVariables"; else VLOG(1) << "tf-tpu-variable-runtime-reformatting inserted no " "TPUReshardVariables"; } } // namespace std::unique_ptr<OperationPass<ModuleOp>>
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 21.9K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/utils/tpu_rewrite_device_util_test.cc
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri Apr 26 09:37:10 UTC 2024 - 46.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/tpu_dynamic_layout_pass.cc
void runOnFunction( func::FuncOp func, const TF::ResourceAliasAnalysis::Info& resource_alias_analysis); StringRef getArgument() const final { return "tf-tpu-dynamic-layout-pass"; } StringRef getDescription() const final { return "Inserts TPU layout ops to determine layout at run time."; } }; // Checks if the input producer op is supported in this transform. Right now, we
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 12.7K bytes - Viewed (0) -
tensorflow/compiler/jit/xla_platform_info.cc
return absl::OkStatus(); } // TFRT-TPU is used if device type is `DEVICE_TPU` and platform_info does not // have `xla_device_metadata`. This is used for TFRT-TPU when // BuildXlaDeviceCompiler() is called in GetCompilerIr(). Currently only // lowering to HLO is needed there and xla::LocalClient doesn't support // building the executable for TFRT-TPU and hence, is set to nullptr here.
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu May 02 17:23:27 UTC 2024 - 17.4K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/tests/passes/replace_stablehlo_ops_in_main_function_with_xla_call_module_ops.mlir
// CHECK: %[[IDENTIFY:.*]] = "tf.Identity"(%[[SUBGRAPH_0]]#1) {device = ""} : (tensor<1024x3xf32>) -> tensor<1024x3xf32> // CHECK: %[[SUBGRAPH_1:.*]] = "tf.XlaCallModule"() <{Sout = [#tf_type.shape<1024x3>], {{.*}} ["CPU", "TPU"], {{.*}}}> {_entry_function = @_stablehlo_main_1
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 01:09:50 UTC 2024 - 39.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/utils/xla_sharding_util.cc
return builder->create<mlir::TF::ConcatOp>( location, output_type, concat_dimension_op.getOutput(), inputs); } // For tile sharded inputs to TPU computation, inject split op between the // input values and TPU computation so that tiled input values are passed in // as inputs to TPU computations. If more than one dimension is sharded, then // a tree of connected split ops are added before tf_device.parallel_execute op.
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed May 22 21:28:13 UTC 2024 - 34K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/tensorflow/quantize_preprocess.cc
// of dropping the attribute. This need not be correct for models not trained // on TPU. // Extracts the StableHLO module from tf.XlaCallModuleOp if the StableHLO // module is serialized in it. pm.addPass(mlir::odml::CreateLegalizeTFXlaCallModuleToStablehloPass()); // Preprocesses TPU-targeting StableHLO module for support in TF Quantizer. pm.addPass(mlir::quant::CreateConvertTpuModelToCpuPass());
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Apr 24 12:49:45 UTC 2024 - 9.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/internal/passes/tpu_cluster_formation.cc
// device. Device attribute is used to infer the appropriate sharding // within TPUs for this op. // TODO(b/183598857): Use explicit sharding ops from the front-end. // For example, dequeue ops generated by // tensorflow/python/tpu/tpu_feed.py if (!tensorflow::IsTPUReplicatedCore(attr.getValue())) { inner_op->removeAttr(kDeviceAttr); }
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu May 02 22:03:30 UTC 2024 - 39.3K bytes - Viewed (0)