basedpyright 1.36.2 → 1.37.0-961d59bacff1958014fe4e9ed2ef3116fe2266e2

package/dist/typeshed-fallback/stubs/tensorflow/tensorflow/compiler/xla/xla_pb2.pyi

@@ -0,0 +1,2558 @@
+"""
+@generated by mypy-protobuf.  Do not edit manually!
+isort:skip_file
+Copyright 2017 The OpenXLA Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================
+"""
+
+import builtins
+import collections.abc
+import sys
+import typing
+
+import google.protobuf.any_pb2
+import google.protobuf.descriptor
+import google.protobuf.internal.containers
+import google.protobuf.internal.enum_type_wrapper
+import google.protobuf.message
+import tensorflow.compiler.xla.service.hlo_pb2
+import tensorflow.compiler.xla.xla_data_pb2
+
+if sys.version_info >= (3, 10):
+    import typing as typing_extensions
+else:
+    import typing_extensions
+
+DESCRIPTOR: google.protobuf.descriptor.FileDescriptor
+
+@typing.final
+class CompilationEnvironmentsProto(google.protobuf.message.Message):
+    """Proto version of `xla::CompilationEnvironments`."""
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    ENVIRONMENTS_FIELD_NUMBER: builtins.int
+    @property
+    def environments(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[google.protobuf.any_pb2.Any]: ...
+    def __init__(self, *, environments: collections.abc.Iterable[google.protobuf.any_pb2.Any] | None = ...) -> None: ...
+    def ClearField(self, field_name: typing.Literal["environments", b"environments"]) -> None: ...
+
+global___CompilationEnvironmentsProto = CompilationEnvironmentsProto
+
+@typing.final
+class DebugOptions(google.protobuf.message.Message):
+    """Debugging options for XLA. These options may change at any time - there are
+    no guarantees about backward or forward compatibility for these fields.
+
+    Debug options naming and organization:
+
+    1. Backend-agnostic options: `xla_$flag_name` - go first, and sorted
+       alphabetically by the flag name.
+
+    2. Backend-specific options: `xla_$backend_$flag_name` - must be in the
+       corresponding backend section, and sorted alphabetically by the flag name.
+    --------------------------------------------------------------------------//
+    XLA backend-agnostic options.
+    --------------------------------------------------------------------------//
+    go/keep-sorted start
+    """
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    class _ShapeChecks:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _ShapeChecksEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._ShapeChecks.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        IGNORE: DebugOptions._ShapeChecks.ValueType  # 0
+        """Do not insert any shape checks for dynamically shaped operations; output
+        buffers might contain garbage data if shapes don't match.
+        """
+        RUNTIME: DebugOptions._ShapeChecks.ValueType  # 1
+        """Check shapes at runtime, will insert an extra synchronization if shapes
+        cannot be proven correct at compile time.
+        """
+        COMPILE_TIME: DebugOptions._ShapeChecks.ValueType  # 2
+        """Will refuse to compile any program where shape correctness can not be
+        established at compile time.
+        """
+
+    class ShapeChecks(_ShapeChecks, metaclass=_ShapeChecksEnumTypeWrapper): ...
+    IGNORE: DebugOptions.ShapeChecks.ValueType  # 0
+    """Do not insert any shape checks for dynamically shaped operations; output
+    buffers might contain garbage data if shapes don't match.
+    """
+    RUNTIME: DebugOptions.ShapeChecks.ValueType  # 1
+    """Check shapes at runtime, will insert an extra synchronization if shapes
+    cannot be proven correct at compile time.
+    """
+    COMPILE_TIME: DebugOptions.ShapeChecks.ValueType  # 2
+    """Will refuse to compile any program where shape correctness can not be
+    established at compile time.
+    """
+
+    class _StepMarkerLocation:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _StepMarkerLocationEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._StepMarkerLocation.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        STEP_MARK_AT_ENTRY: DebugOptions._StepMarkerLocation.ValueType  # 0
+        """Generate a step marker at the program entry. This handles the case where
+        each step is done by one or multiple program execution(s). Only the first
+        program will be tagged for generating a step marker at the program entry.
+        This is the default.
+        """
+        STEP_MARK_AT_TOP_LEVEL_WHILE_LOOP: DebugOptions._StepMarkerLocation.ValueType  # 1
+        """Generate a step marker at each iteration of the top level while loop,
+        which is assumed to be a training loop.
+        """
+        STEP_MARK_AT_SECOND_LEVEL_WHILE_LOOP: DebugOptions._StepMarkerLocation.ValueType  # 3
+        """Generate a step marker at each iteration of the second level while loops,
+        which is assumed to be a training or eval loop.
+        """
+        STEP_MARK_NONE: DebugOptions._StepMarkerLocation.ValueType  # 2
+        """No step marker generated."""
+
+    class StepMarkerLocation(_StepMarkerLocation, metaclass=_StepMarkerLocationEnumTypeWrapper): ...
+    STEP_MARK_AT_ENTRY: DebugOptions.StepMarkerLocation.ValueType  # 0
+    """Generate a step marker at the program entry. This handles the case where
+    each step is done by one or multiple program execution(s). Only the first
+    program will be tagged for generating a step marker at the program entry.
+    This is the default.
+    """
+    STEP_MARK_AT_TOP_LEVEL_WHILE_LOOP: DebugOptions.StepMarkerLocation.ValueType  # 1
+    """Generate a step marker at each iteration of the top level while loop,
+    which is assumed to be a training loop.
+    """
+    STEP_MARK_AT_SECOND_LEVEL_WHILE_LOOP: DebugOptions.StepMarkerLocation.ValueType  # 3
+    """Generate a step marker at each iteration of the second level while loops,
+    which is assumed to be a training or eval loop.
+    """
+    STEP_MARK_NONE: DebugOptions.StepMarkerLocation.ValueType  # 2
+    """No step marker generated."""
+
+    class _CollectiveOpType:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _CollectiveOpTypeEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._CollectiveOpType.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        NOOP: DebugOptions._CollectiveOpType.ValueType  # 0
+        ALLREDUCE: DebugOptions._CollectiveOpType.ValueType  # 1
+        ALLGATHER: DebugOptions._CollectiveOpType.ValueType  # 2
+        REDUCESCATTER: DebugOptions._CollectiveOpType.ValueType  # 3
+        COLLECTIVEBROADCAST: DebugOptions._CollectiveOpType.ValueType  # 4
+        ALLTOALL: DebugOptions._CollectiveOpType.ValueType  # 5
+        COLLECTIVEPERMUTE: DebugOptions._CollectiveOpType.ValueType  # 6
+
+    class CollectiveOpType(_CollectiveOpType, metaclass=_CollectiveOpTypeEnumTypeWrapper):
+        """Enum to define all collective ops
+        that xla supports.
+        """
+
+    NOOP: DebugOptions.CollectiveOpType.ValueType  # 0
+    ALLREDUCE: DebugOptions.CollectiveOpType.ValueType  # 1
+    ALLGATHER: DebugOptions.CollectiveOpType.ValueType  # 2
+    REDUCESCATTER: DebugOptions.CollectiveOpType.ValueType  # 3
+    COLLECTIVEBROADCAST: DebugOptions.CollectiveOpType.ValueType  # 4
+    ALLTOALL: DebugOptions.CollectiveOpType.ValueType  # 5
+    COLLECTIVEPERMUTE: DebugOptions.CollectiveOpType.ValueType  # 6
+
+    class _CommandBufferCmdType:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _CommandBufferCmdTypeEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._CommandBufferCmdType.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        INVALID: DebugOptions._CommandBufferCmdType.ValueType  # 0
+        FUSION: DebugOptions._CommandBufferCmdType.ValueType  # 1
+        CUBLAS: DebugOptions._CommandBufferCmdType.ValueType  # 2
+        CUDNN: DebugOptions._CommandBufferCmdType.ValueType  # 3
+        COLLECTIVES: DebugOptions._CommandBufferCmdType.ValueType  # 4
+        CONDITIONALS: DebugOptions._CommandBufferCmdType.ValueType  # 5
+        CUSTOM_CALL: DebugOptions._CommandBufferCmdType.ValueType  # 6
+        CUBLASLT: DebugOptions._CommandBufferCmdType.ValueType  # 7
+
+    class CommandBufferCmdType(_CommandBufferCmdType, metaclass=_CommandBufferCmdTypeEnumTypeWrapper):
+        """Commands are categorized into 5 types:
+        FUSION represents regular fusion kernels.
+        CUBLAS/CUBLASLT, CUDNN, and COLLECTIVES represent library calls.
+        CONDITIONALS represents control flow.
+        """
+
+    INVALID: DebugOptions.CommandBufferCmdType.ValueType  # 0
+    FUSION: DebugOptions.CommandBufferCmdType.ValueType  # 1
+    CUBLAS: DebugOptions.CommandBufferCmdType.ValueType  # 2
+    CUDNN: DebugOptions.CommandBufferCmdType.ValueType  # 3
+    COLLECTIVES: DebugOptions.CommandBufferCmdType.ValueType  # 4
+    CONDITIONALS: DebugOptions.CommandBufferCmdType.ValueType  # 5
+    CUSTOM_CALL: DebugOptions.CommandBufferCmdType.ValueType  # 6
+    CUBLASLT: DebugOptions.CommandBufferCmdType.ValueType  # 7
+
+    class _PartitioningAlgorithm:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _PartitioningAlgorithmEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._PartitioningAlgorithm.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        PARTITIONING_ALGORITHM_NOOP: DebugOptions._PartitioningAlgorithm.ValueType  # 0
+        PARTITIONING_ALGORITHM_EXP0: DebugOptions._PartitioningAlgorithm.ValueType  # 1
+        PARTITIONING_ALGORITHM_EXP1: DebugOptions._PartitioningAlgorithm.ValueType  # 2
+        PARTITIONING_ALGORITHM_EXP2: DebugOptions._PartitioningAlgorithm.ValueType  # 3
+
+    class PartitioningAlgorithm(_PartitioningAlgorithm, metaclass=_PartitioningAlgorithmEnumTypeWrapper): ...
+    PARTITIONING_ALGORITHM_NOOP: DebugOptions.PartitioningAlgorithm.ValueType  # 0
+    PARTITIONING_ALGORITHM_EXP0: DebugOptions.PartitioningAlgorithm.ValueType  # 1
+    PARTITIONING_ALGORITHM_EXP1: DebugOptions.PartitioningAlgorithm.ValueType  # 2
+    PARTITIONING_ALGORITHM_EXP2: DebugOptions.PartitioningAlgorithm.ValueType  # 3
+
+    class _WhileLoopUnrolling:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _WhileLoopUnrollingEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._WhileLoopUnrolling.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        WHILE_LOOP_UNROLLING_NO_UNROLL: DebugOptions._WhileLoopUnrolling.ValueType  # 0
+        WHILE_LOOP_UNROLLING_DOUBLE_BUFFER: DebugOptions._WhileLoopUnrolling.ValueType  # 1
+        """Has the same effect as setting
+        `xla_gpu_enable_while_loop_double_buffering`.
+        """
+        WHILE_LOOP_UNROLLING_FULL_UNROLL: DebugOptions._WhileLoopUnrolling.ValueType  # 2
+        """Enables full loop unrolling using the same strategy as `DOUBLE_BUFFER`."""
+
+    class WhileLoopUnrolling(_WhileLoopUnrolling, metaclass=_WhileLoopUnrollingEnumTypeWrapper): ...
+    WHILE_LOOP_UNROLLING_NO_UNROLL: DebugOptions.WhileLoopUnrolling.ValueType  # 0
+    WHILE_LOOP_UNROLLING_DOUBLE_BUFFER: DebugOptions.WhileLoopUnrolling.ValueType  # 1
+    """Has the same effect as setting
+    `xla_gpu_enable_while_loop_double_buffering`.
+    """
+    WHILE_LOOP_UNROLLING_FULL_UNROLL: DebugOptions.WhileLoopUnrolling.ValueType  # 2
+    """Enables full loop unrolling using the same strategy as `DOUBLE_BUFFER`."""
+
+    class _AutotuneCacheMode:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _AutotuneCacheModeEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[DebugOptions._AutotuneCacheMode.ValueType], builtins.type
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        AUTOTUNE_CACHE_MODE_UNSPECIFIED: DebugOptions._AutotuneCacheMode.ValueType  # 0
+        AUTOTUNE_CACHE_MODE_UPDATE: DebugOptions._AutotuneCacheMode.ValueType  # 1
+        """If the cache exists per fusion autotuner loads it and terminates,
+        otherwise runs autotuner and dumps the result.
+        """
+        AUTOTUNE_CACHE_MODE_READ: DebugOptions._AutotuneCacheMode.ValueType  # 2
+        """Sets readonly access to the cache for the per fusion autotuner. Same as
+        above, but doesn't dump anything.
+        """
+
+    class AutotuneCacheMode(_AutotuneCacheMode, metaclass=_AutotuneCacheModeEnumTypeWrapper): ...
+    AUTOTUNE_CACHE_MODE_UNSPECIFIED: DebugOptions.AutotuneCacheMode.ValueType  # 0
+    AUTOTUNE_CACHE_MODE_UPDATE: DebugOptions.AutotuneCacheMode.ValueType  # 1
+    """If the cache exists per fusion autotuner loads it and terminates,
+    otherwise runs autotuner and dumps the result.
+    """
+    AUTOTUNE_CACHE_MODE_READ: DebugOptions.AutotuneCacheMode.ValueType  # 2
+    """Sets readonly access to the cache for the per fusion autotuner. Same as
+    above, but doesn't dump anything.
+    """
+
+    @typing.final
+    class XlaBackendExtraOptionsEntry(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        KEY_FIELD_NUMBER: builtins.int
+        VALUE_FIELD_NUMBER: builtins.int
+        key: builtins.str
+        value: builtins.str
+        def __init__(self, *, key: builtins.str | None = ..., value: builtins.str | None = ...) -> None: ...
+        def ClearField(self, field_name: typing.Literal["key", b"key", "value", b"value"]) -> None: ...
+
+    XLA_CPU_ENABLE_CONCURRENCY_OPTIMIZED_SCHEDULER_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_FAST_MATH_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_FAST_MIN_MAX_FIELD_NUMBER: builtins.int
+    XLA_CPU_FAST_MATH_HONOR_DIVISION_FIELD_NUMBER: builtins.int
+    XLA_CPU_FAST_MATH_HONOR_FUNCTIONS_FIELD_NUMBER: builtins.int
+    XLA_CPU_FAST_MATH_HONOR_INFS_FIELD_NUMBER: builtins.int
+    XLA_CPU_FAST_MATH_HONOR_NANS_FIELD_NUMBER: builtins.int
+    XLA_CPU_USE_THUNK_RUNTIME_FIELD_NUMBER: builtins.int
+    XLA_CPU_PARALLEL_CODEGEN_SPLIT_COUNT_FIELD_NUMBER: builtins.int
+    XLA_CPU_PREFER_VECTOR_WIDTH_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXPERIMENTAL_AUTOTUNE_CACHE_MODE_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXPERIMENTAL_DISABLE_BINARY_LIBRARIES_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXPERIMENTAL_ENABLE_TRITON_SOFTMAX_PRIORITY_FUSION_FIELD_NUMBER: builtins.int
+    XLA_GPU_UNSUPPORTED_ENABLE_TRITON_GEMM_FIELD_NUMBER: builtins.int
+    XLA_HLO_GRAPH_ADDRESSES_FIELD_NUMBER: builtins.int
+    XLA_HLO_PROFILE_FIELD_NUMBER: builtins.int
+    XLA_DISABLE_HLO_PASSES_FIELD_NUMBER: builtins.int
+    XLA_ENABLE_HLO_PASSES_ONLY_FIELD_NUMBER: builtins.int
+    XLA_DISABLE_ALL_HLO_PASSES_FIELD_NUMBER: builtins.int
+    XLA_BACKEND_OPTIMIZATION_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_EMBED_IR_IN_EXECUTABLE_FIELD_NUMBER: builtins.int
+    XLA_ELIMINATE_HLO_IMPLICIT_BROADCAST_FIELD_NUMBER: builtins.int
+    XLA_CPU_MULTI_THREAD_EIGEN_FIELD_NUMBER: builtins.int
+    XLA_GPU_CUDA_DATA_DIR_FIELD_NUMBER: builtins.int
+    XLA_GPU_FTZ_FIELD_NUMBER: builtins.int
+    XLA_LLVM_ENABLE_ALIAS_SCOPE_METADATA_FIELD_NUMBER: builtins.int
+    XLA_LLVM_ENABLE_NOALIAS_METADATA_FIELD_NUMBER: builtins.int
+    XLA_LLVM_ENABLE_INVARIANT_LOAD_METADATA_FIELD_NUMBER: builtins.int
+    XLA_LLVM_DISABLE_EXPENSIVE_PASSES_FIELD_NUMBER: builtins.int
+    XLA_TEST_ALL_OUTPUT_LAYOUTS_FIELD_NUMBER: builtins.int
+    XLA_TEST_ALL_INPUT_LAYOUTS_FIELD_NUMBER: builtins.int
+    XLA_HLO_GRAPH_SHARDING_COLOR_FIELD_NUMBER: builtins.int
+    XLA_CPU_USE_MKL_DNN_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_FAST_MIN_MAX_FIELD_NUMBER: builtins.int
+    XLA_ALLOW_EXCESS_PRECISION_FIELD_NUMBER: builtins.int
+    XLA_GPU_CRASH_ON_VERIFICATION_FAILURES_FIELD_NUMBER: builtins.int
+    XLA_GPU_AUTOTUNE_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_FORCE_HOST_PLATFORM_DEVICE_COUNT_FIELD_NUMBER: builtins.int
+    XLA_GPU_DISABLE_GPUASM_OPTIMIZATIONS_FIELD_NUMBER: builtins.int
+    XLA_GPU_SHAPE_CHECKS_FIELD_NUMBER: builtins.int
+    XLA_HLO_EVALUATOR_USE_FAST_PATH_FIELD_NUMBER: builtins.int
+    XLA_ALLOW_SCALAR_INDEX_DYNAMIC_OPS_FIELD_NUMBER: builtins.int
+    XLA_STEP_MARKER_LOCATION_FIELD_NUMBER: builtins.int
+    XLA_DUMP_TO_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_MODULE_RE_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_PASS_RE_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_TEXT_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_PROTO_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_DOT_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_URL_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_HTML_FIELD_NUMBER: builtins.int
+    XLA_DUMP_FUSION_VISUALIZATION_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_SNAPSHOTS_FIELD_NUMBER: builtins.int
+    XLA_DUMP_INCLUDE_TIMESTAMP_FIELD_NUMBER: builtins.int
+    XLA_DUMP_MAX_HLO_MODULES_FIELD_NUMBER: builtins.int
+    XLA_DUMP_MODULE_METADATA_FIELD_NUMBER: builtins.int
+    XLA_DUMP_COMPRESS_PROTOS_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_AS_LONG_TEXT_FIELD_NUMBER: builtins.int
+    XLA_GPU_FORCE_CONV_NCHW_FIELD_NUMBER: builtins.int
+    XLA_GPU_FORCE_CONV_NHWC_FIELD_NUMBER: builtins.int
+    XLA_GPU_PTX_FILE_FIELD_NUMBER: builtins.int
+    XLA_GPU_DUMP_LLVMIR_FIELD_NUMBER: builtins.int
+    XLA_DUMP_ENABLE_MLIR_PRETTY_FORM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ALGORITHM_DENYLIST_PATH_FIELD_NUMBER: builtins.int
+    XLA_TPU_DETECT_NAN_FIELD_NUMBER: builtins.int
+    XLA_TPU_DETECT_INF_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_XPROF_TRACEME_FIELD_NUMBER: builtins.int
+    XLA_GPU_UNSAFE_FALLBACK_TO_DRIVER_ON_PTXAS_NOT_FOUND_FIELD_NUMBER: builtins.int
+    XLA_GPU_ASM_EXTRA_FLAGS_FIELD_NUMBER: builtins.int
+    XLA_MULTIHEAP_SIZE_CONSTRAINT_PER_HEAP_FIELD_NUMBER: builtins.int
+    XLA_DETAILED_LOGGING_FIELD_NUMBER: builtins.int
+    XLA_ENABLE_DUMPING_FIELD_NUMBER: builtins.int
+    XLA_GPU_FORCE_COMPILATION_PARALLELISM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_LLVM_MODULE_COMPILATION_PARALLELISM_FIELD_NUMBER: builtins.int
+    XLA_GPU_DETERMINISTIC_OPS_FIELD_NUMBER: builtins.int
+    XLA_GPU_LLVM_IR_FILE_FIELD_NUMBER: builtins.int
+    XLA_GPU_DISABLE_ASYNC_COLLECTIVES_FIELD_NUMBER: builtins.int
+    XLA_GPU_ALL_REDUCE_COMBINE_THRESHOLD_BYTES_FIELD_NUMBER: builtins.int
+    XLA_GPU_ALL_GATHER_COMBINE_THRESHOLD_BYTES_FIELD_NUMBER: builtins.int
+    XLA_GPU_REDUCE_SCATTER_COMBINE_THRESHOLD_BYTES_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_ALL_GATHER_COMBINE_BY_DIM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_REDUCE_SCATTER_COMBINE_BY_DIM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_REASSOCIATION_FOR_CONVERTED_AR_FIELD_NUMBER: builtins.int
+    XLA_GPU_ALL_REDUCE_BLUECONNECT_NUM_DEVICES_PER_HOST_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_WHILE_LOOP_REDUCE_SCATTER_CODE_MOTION_FIELD_NUMBER: builtins.int
+    XLA_GPU_COLLECTIVE_INFLATION_FACTOR_FIELD_NUMBER: builtins.int
+    XLA_LLVM_FORCE_INLINE_BEFORE_SPLIT_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUDNN_FRONTEND_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUDNN_FMHA_FIELD_NUMBER: builtins.int
+    XLA_GPU_FUSED_ATTENTION_USE_CUDNN_RNG_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUDNN_LAYER_NORM_FIELD_NUMBER: builtins.int
+    XLA_DUMP_DISABLE_METADATA_FIELD_NUMBER: builtins.int
+    XLA_DUMP_HLO_PIPELINE_RE_FIELD_NUMBER: builtins.int
+    XLA_GPU_STRICT_CONV_ALGORITHM_PICKER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUSTOM_FUSIONS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUSTOM_FUSIONS_RE_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_DYNAMIC_SLICE_FUSION_FIELD_NUMBER: builtins.int
+    XLA_GPU_NCCL_TERMINATION_TIMEOUT_SECONDS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_SHARED_CONSTANTS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUBLASLT_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_COMMAND_BUFFER_FIELD_NUMBER: builtins.int
+    XLA_GPU_GRAPH_MIN_GRAPH_SIZE_FIELD_NUMBER: builtins.int
+    XLA_GPU_GRAPH_ENABLE_CONCURRENT_REGION_FIELD_NUMBER: builtins.int
+    XLA_GPU_REDZONE_SCRATCH_MAX_MEGABYTES_FIELD_NUMBER: builtins.int
+    XLA_GPU_REDZONE_PADDING_BYTES_FIELD_NUMBER: builtins.int
+    XLA_CPU_USE_ACL_FIELD_NUMBER: builtins.int
+    XLA_CPU_STRICT_DOT_CONV_MATH_FIELD_NUMBER: builtins.int
+    XLA_GPU_USE_RUNTIME_FUSION_FIELD_NUMBER: builtins.int
+    XLA_DUMP_LATENCY_HIDING_SCHEDULE_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_MLIR_TILING_AND_FUSION_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_CUSTOM_MATMUL_TILING_FIELD_NUMBER: builtins.int
+    XLA_CPU_MATMUL_TILING_M_DIM_FIELD_NUMBER: builtins.int
+    XLA_CPU_MATMUL_TILING_N_DIM_FIELD_NUMBER: builtins.int
+    XLA_CPU_MATMUL_TILING_K_DIM_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_MLIR_FUSION_OUTLINING_FIELD_NUMBER: builtins.int
+    XLA_CPU_ENABLE_EXPERIMENTAL_DEALLOCATION_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_LATENCY_HIDING_SCHEDULER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_HIGHEST_PRIORITY_ASYNC_STREAM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_ANALYTICAL_LATENCY_ESTIMATOR_FIELD_NUMBER: builtins.int
+    XLA_GPU_LHS_ENABLE_GPU_ASYNC_TRACKER_FIELD_NUMBER: builtins.int
+    XLA_GPU_PGLE_PROFILE_FILE_OR_DIRECTORY_PATH_FIELD_NUMBER: builtins.int
+    XLA_GPU_MEMORY_LIMIT_SLOP_FACTOR_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PIPELINED_COLLECTIVES_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PIPELINED_ALL_REDUCE_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PIPELINED_ALL_GATHER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PIPELINED_REDUCE_SCATTER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PIPELINED_P2P_FIELD_NUMBER: builtins.int
+    XLA_GPU_RUN_POST_LAYOUT_COLLECTIVE_PIPELINER_FIELD_NUMBER: builtins.int
+    XLA_GPU_COLLECTIVE_PERMUTE_DECOMPOSER_THRESHOLD_FIELD_NUMBER: builtins.int
+    XLA_PARTITIONING_ALGORITHM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_TRITON_GEMM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUDNN_INT8X32_CONVOLUTION_REORDERING_FIELD_NUMBER: builtins.int
+    XLA_GPU_TRITON_GEMM_ANY_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXHAUSTIVE_TILING_SEARCH_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PRIORITY_FUSION_FIELD_NUMBER: builtins.int
+    XLA_GPU_DUMP_AUTOTUNE_RESULTS_TO_FIELD_NUMBER: builtins.int
+    XLA_GPU_LOAD_AUTOTUNE_RESULTS_FROM_FIELD_NUMBER: builtins.int
+    XLA_GPU_TARGET_CONFIG_FILENAME_FIELD_NUMBER: builtins.int
+    XLA_GPU_AUTO_SPMD_PARTITIONING_MEMORY_BUDGET_GB_FIELD_NUMBER: builtins.int
+    XLA_GPU_AUTO_SPMD_PARTITIONING_MEMORY_BUDGET_RATIO_FIELD_NUMBER: builtins.int
+    XLA_GPU_TRITON_GEMM_DISABLE_REDUCED_PRECISION_REDUCTION_FIELD_NUMBER: builtins.int
+    XLA_GPU_TRITON_FUSION_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_GPU_DUMP_AUTOTUNED_GEMM_FUSIONS_FIELD_NUMBER: builtins.int
+    XLA_GPU_OVERRIDE_GEMM_AUTOTUNER_FIELD_NUMBER: builtins.int
+    XLA_GPU_COPY_INSERTION_USE_REGION_ANALYSIS_FIELD_NUMBER: builtins.int
+    XLA_GPU_COLLECT_COST_MODEL_STATS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_SPLIT_K_AUTOTUNING_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_REDUCTION_EPILOGUE_FUSION_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_NCCL_CLIQUE_OPTIMIZATION_FIELD_NUMBER: builtins.int
+    XLA_GPU_MOCK_CUSTOM_CALLS_FIELD_NUMBER: builtins.int
+    XLA_GPU_CUBLAS_FALLBACK_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_WHILE_LOOP_DOUBLE_BUFFERING_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_WHILE_LOOP_UNROLLING_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENSURE_MINOR_DOT_CONTRACTION_DIMS_FIELD_NUMBER: builtins.int
+    XLA_GPU_FILTER_KERNELS_SPILLING_REGISTERS_ON_AUTOTUNING_FIELD_NUMBER: builtins.int
+    XLA_DEBUG_BUFFER_ASSIGNMENT_SHOW_MAX_FIELD_NUMBER: builtins.int
+    XLA_GPU_LLVM_VERIFICATION_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_CUB_RADIX_SORT_FIELD_NUMBER: builtins.int
+    XLA_GPU_THRESHOLD_FOR_WINDOWED_EINSUM_MIB_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_TRITON_HOPPER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_NCCL_USER_BUFFERS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_NCCL_COMM_SPLITTING_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_NCCL_PER_STREAM_COMMS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_LIBNVPTXCOMPILER_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_DOT_STRENGTH_REDUCTION_FIELD_NUMBER: builtins.int
+    XLA_GPU_MULTI_STREAMED_WINDOWED_EINSUM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_BF16_6WAY_GEMM_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_BF16_3WAY_GEMM_FIELD_NUMBER: builtins.int
+    XLA_GPU_NCCL_COLLECTIVE_MAX_NCHANNELS_FIELD_NUMBER: builtins.int
+    XLA_GPU_NCCL_P2P_MAX_NCHANNELS_FIELD_NUMBER: builtins.int
+    XLA_GPU_MLIR_EMITTER_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_GPU_GEMM_REWRITE_SIZE_THRESHOLD_FIELD_NUMBER: builtins.int
+    XLA_GPU_REQUIRE_COMPLETE_AOT_AUTOTUNE_RESULTS_FIELD_NUMBER: builtins.int
+    XLA_GPU_CUDNN_GEMM_FUSION_LEVEL_FIELD_NUMBER: builtins.int
+    XLA_GPU_USE_MEMCPY_LOCAL_P2P_FIELD_NUMBER: builtins.int
+    XLA_GPU_AUTOTUNE_MAX_SOLUTIONS_FIELD_NUMBER: builtins.int
+    XLA_DUMP_LARGE_CONSTANTS_FIELD_NUMBER: builtins.int
+    XLA_GPU_VERIFY_TRITON_FUSION_NUMERICS_FIELD_NUMBER: builtins.int
+    XLA_GPU_DUMP_AUTOTUNE_LOGS_TO_FIELD_NUMBER: builtins.int
+    XLA_REDUCE_WINDOW_REWRITE_BASE_LENGTH_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_HOST_MEMORY_OFFLOADING_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXCLUDE_NONDETERMINISTIC_OPS_FIELD_NUMBER: builtins.int
+    XLA_GPU_NCCL_TERMINATE_ON_ERROR_FIELD_NUMBER: builtins.int
+    XLA_GPU_SHARD_AUTOTUNING_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_APPROX_COSTLY_COLLECTIVES_FIELD_NUMBER: builtins.int
+    XLA_GPU_KERNEL_CACHE_FILE_FIELD_NUMBER: builtins.int
+    XLA_GPU_UNSAFE_PIPELINED_LOOP_ANNOTATOR_FIELD_NUMBER: builtins.int
+    XLA_GPU_PER_FUSION_AUTOTUNE_CACHE_DIR_FIELD_NUMBER: builtins.int
+    XLA_CMD_BUFFER_TRACE_CACHE_SIZE_FIELD_NUMBER: builtins.int
+    XLA_GPU_TEMP_BUFFER_USE_SEPARATE_COLOR_FIELD_NUMBER: builtins.int
+    LEGACY_COMMAND_BUFFER_CUSTOM_CALL_TARGETS_FIELD_NUMBER: builtins.int
+    XLA_SYNTAX_SUGAR_ASYNC_OPS_FIELD_NUMBER: builtins.int
+    XLA_GPU_AUTOTUNE_GEMM_RTOL_FIELD_NUMBER: builtins.int
+    XLA_ENABLE_COMMAND_BUFFERS_DURING_PROFILING_FIELD_NUMBER: builtins.int
+    XLA_GPU_CUDNN_GEMM_MAX_PLANS_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_LIBNVJITLINK_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_TRITON_GEMM_INT4_FIELD_NUMBER: builtins.int
+    XLA_GPU_ASYNC_DOT_FIELD_NUMBER: builtins.int
+    XLA_GPU_ENABLE_PGLE_ACCURACY_CHECKER_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXECUTABLE_WARN_STUCK_TIMEOUT_SECONDS_FIELD_NUMBER: builtins.int
+    XLA_GPU_EXECUTABLE_TERMINATE_TIMEOUT_SECONDS_FIELD_NUMBER: builtins.int
+    XLA_EXPERIMENTAL_IGNORE_CHANNEL_ID_FIELD_NUMBER: builtins.int
+    XLA_BACKEND_EXTRA_OPTIONS_FIELD_NUMBER: builtins.int
+    xla_cpu_enable_concurrency_optimized_scheduler: builtins.bool
+    """--------------------------------------------------------------------------//
+    XLA:CPU options.
+    --------------------------------------------------------------------------//
+
+    go/keep-sorted start newline_separated=yes
+
+    When true, XLA:CPU uses HLO module scheduler that is optimized for
+    extracting concurrency at the cost of extra memory: we extend the live
+    ranges of temporaries to allow XLA runtime to schedule independent
+    operations in parallel on separate threads.
+    """
+    xla_cpu_enable_fast_math: builtins.bool
+    """When true, "unsafe" mathematical optimizations are enabled. These
+    transformations include but are not limited to:
+
+     - Reducing the precision of operations (e.g. using an approximate sin
+       function, or transforming x/y into x * (1/y)).
+     - Assuming that operations never produce or consume NaN or +/- Inf (this
+       behavior can be adjusted using xla_cpu_fast_math_allow_{nans|infs}).
+     - Assuming that +0 and -0 are indistinguishable.
+    """
+    xla_cpu_enable_fast_min_max: builtins.bool
+    """When false we lower the Minimum and Maximum hlos in the CPU backend such
+    that Min(NotNaN, NaN) = Min(NaN, NotNaN) = NaN.  In other words, if flag
+    this is false we always propagate NaNs through Min and Max.
+
+    Note, this does not correspond to the exact same behavior as the gpu flag
+    below!
+    """
+    xla_cpu_fast_math_honor_division: builtins.bool
+    """When xla_cpu_enable_fast_math is true then this controls whether we forbid
+    to use the reciprocal of an argument instead of division. Ignored when
+    xla_cpu_enable_fast_math is false.
+    """
+    xla_cpu_fast_math_honor_functions: builtins.bool
+    """When xla_cpu_enable_fast_math is true then this controls whether we forbid
+    to approximate calculations for functions. Ignored when
+    xla_cpu_enable_fast_math is false.
+    """
+    xla_cpu_fast_math_honor_infs: builtins.bool
+    """When xla_cpu_enable_fast_math is true then this controls whether we allow
+    operations to produce infinites. Ignored when xla_cpu_enable_fast_math is
+    false.
+    """
+    xla_cpu_fast_math_honor_nans: builtins.bool
+    """When xla_cpu_enable_fast_math is true then this controls whether we allow
+    operations to produce NaNs.  Ignored when xla_cpu_enable_fast_math is
+    false.
+    """
+    xla_cpu_use_thunk_runtime: builtins.bool
+    """When true, XLA:CPU uses the thunk runtime to execute compiled program."""
+    xla_cpu_parallel_codegen_split_count: builtins.int
+    """The number of parts to split the LLVM module into before codegen. This
+    allows XLA to compile all parts in parallel, and resolve kernel symbols
+    from different dynamic libraries.
+    """
+    xla_cpu_prefer_vector_width: builtins.int
+    """A `prefer-vector-width` value that is passed to the LLVM backend. Default
+    value is `256` (AVX2 on x86 platforms).
+    """
+    xla_gpu_experimental_autotune_cache_mode: global___DebugOptions.AutotuneCacheMode.ValueType
+    """--------------------------------------------------------------------------//
+    XLA:GPU options.
+    --------------------------------------------------------------------------//
+    go/keep-sorted start newline_separated=yes skip_lines=1
+
+    Specifies the behavior of per kernel autotuning cache.
+    """
+    xla_gpu_experimental_disable_binary_libraries: builtins.bool
+    """Experimentally disables binary libraries in GPU compiler passes."""
+    xla_gpu_experimental_enable_triton_softmax_priority_fusion: builtins.bool
+    """Gates the experimental feature coupling the Triton Softmax pattern matcher
+    with priority fusion.
+    """
+    xla_gpu_unsupported_enable_triton_gemm: builtins.bool
+    """Internal debug/testing flag to switch Triton GEMM fusions on or off."""
+    xla_hlo_graph_addresses: builtins.bool
+    """--------------------------------------------------------------------------//
+    A bag of XLA options that have to be categorized.
+    --------------------------------------------------------------------------//
+
+    Show addresses of HLO ops in graph dump.
+    """
+    xla_hlo_profile: builtins.bool
+    """Instrument the computation to collect per-HLO cycle counts."""
+    xla_disable_all_hlo_passes: builtins.bool
+    """Disables all HLO passes.  Notes that some passes are necessary for
+    correctness and the invariants that must be satisfied by "fully optimized"
+    HLO are different for different devices and may change over time.  The only
+    "guarantee", such as it is, is that if you compile XLA and dump the
+    optimized HLO for some graph, you should be able to run it again on the
+    same device with the same build of XLA.
+    """
+    xla_backend_optimization_level: builtins.int
+    """Numerical optimization level for the XLA compiler backend; the specific
+    interpretation of this value is left to the backends.
+    """
+    xla_embed_ir_in_executable: builtins.bool
+    """Embed the compiler IR as a string in the executable."""
+    xla_eliminate_hlo_implicit_broadcast: builtins.bool
+    """Eliminate implicit broadcasts when lowering user computations to HLO
+    instructions; use explicit broadcast instead.
+    """
+    xla_cpu_multi_thread_eigen: builtins.bool
+    """When generating calls to Eigen in the CPU backend, use multi-threaded Eigen
+    mode.
+    """
+    xla_gpu_cuda_data_dir: builtins.str
+    """Path to directory with cuda/ptx tools and libraries."""
+    xla_gpu_ftz: builtins.bool
+    """Enable flush-to-zero semantics in the GPU backend."""
+    xla_llvm_enable_alias_scope_metadata: builtins.bool
+    """If true, in LLVM-based backends, emit !alias.scope metadata in
+    generated IR.
+    """
+    xla_llvm_enable_noalias_metadata: builtins.bool
+    """If true, in LLVM-based backends, emit !noalias metadata in the
+    generated IR.
+    """
+    xla_llvm_enable_invariant_load_metadata: builtins.bool
+    """If true, in LLVM-based backends, emit !invariant.load metadata in
+    the generated IR.
+    """
+    xla_llvm_disable_expensive_passes: builtins.bool
+    """If true, a set of expensive LLVM optimization passes will not be run."""
+    xla_test_all_output_layouts: builtins.bool
+    """This is used by ClientLibraryTestBase::ComputeAndCompare*. If true, the
+    computation will run n! times with all permunations of layouts for the
+    output shape in rank n. For example, with a 3D shape, all permutations of
+    the set {0, 1, 2} are tried.
+    """
+    xla_test_all_input_layouts: builtins.bool
+    """This is used by ClientLibraryTestBase::ComputeAndCompare*. If true, the
+    computation will run for all permunations of layouts of all input
+    arguments. For example, with 2 input arguments in 2D and 4D shapes, the
+    computation will run 2! * 4! times.
+    """
+    xla_hlo_graph_sharding_color: builtins.bool
+    """Assign colors based on sharding information when generating the Graphviz
+    HLO graph.
+    """
+    xla_cpu_use_mkl_dnn: builtins.bool
+    """Generate calls to MKL-DNN in the CPU backend."""
+    xla_gpu_enable_fast_min_max: builtins.bool
+    """When true we lower the Minimum and Maximum hlos in the GPU backend such
+    that Min(NotNaN, NaN) = Min(NaN, NotNaN) = NotNaN.  In other words, if flag
+    this is true we don't propagate NaNs through Min and Max.
+
+    Note, this does not correspond to the exact same behavior as the cpu flag
+    above!
+    """
+    xla_allow_excess_precision: builtins.bool
+    """Allows xla to increase the output precision of floating point operations
+    and all floating-point conversions to be simplified, including those
+    that affect the numerics. The `FloatNormalization` pass inserts many
+    `f32 -> bf16 -> f32` conversion pairs. These are not removed by the
+    `AlgebraicSimplifier`, as that will only simplify conversions that are
+    no-ops, e.g. `bf16 -> f32 -> bf16`. Removing these improves accuracy.
+    """
+    xla_gpu_crash_on_verification_failures: builtins.bool
+    """Crashes the program when any kind of verification fails, instead of just
+    logging the failures. One example is cross checking of convolution results
+    among different algorithms.
+    """
+    xla_gpu_autotune_level: builtins.int
+    """0:   Disable gemm and convolution autotuning.
+    1:   Enable autotuning, but disable correctness checking.
+    2:   Also set output buffers to random numbers during autotuning.
+    3:   Also reset output buffers to random numbers after autotuning each
+         algorithm.
+    4+:  Also check for correct outputs and for out-of-bounds reads/writes.
+
+    Default: 4.
+    """
+    xla_force_host_platform_device_count: builtins.int
+    """Force the host platform to pretend that there are these many host
+    "devices".  All these devices are backed by the same threadpool.  Defaults
+    to 1.
+
+    Setting this to anything other than 1 can increase overhead from context
+    switching but we let the user override this behavior to help run tests on
+    the host that run models in parallel across multiple devices.
+    """
+    xla_gpu_disable_gpuasm_optimizations: builtins.bool
+    """If set to true XLA:GPU invokes `ptxas` with -O0 (default is -O3)."""
+    xla_gpu_shape_checks: global___DebugOptions.ShapeChecks.ValueType
+    xla_hlo_evaluator_use_fast_path: builtins.bool
+    """Enable fast math with eigen in the HLO evaluator."""
+    xla_allow_scalar_index_dynamic_ops: builtins.bool
+    """Temporary option to allow support for both the R1 and the scalar index
+    versions of DynamicSlice and DynamicUpdateSlice. Only used for testing.
+    """
+    xla_step_marker_location: global___DebugOptions.StepMarkerLocation.ValueType
+    """Option to emit a target-specific marker to indicate the start of a training
+    step. The location of the marker (if any) is determined by the option
+    value.
+    """
+    xla_dump_to: builtins.str
+    """
+    BEGIN flags controlling dumping HLO modules for debugging.
+
+    When dumping is enabled, HLO modules dumped at the very beginning and end
+    of compilation, and optionally also during the pass pipeline.
+
+    In general, if you set one of these flags, we will try to infer reasonable
+    defaults for the others.  For example:
+
+     * Setting --xla_dump_to=/tmp/foo without specifying a format
+       with --xla_dump_hlo_as_* will turn on --xla_dump_hlo_as_text.
+
+     * Setting --xla_dump_hlo_as_text without specifying --xla_dump_to will
+       dump to stdout.
+
+    Directory to dump into.
+    """
+    xla_dump_hlo_module_re: builtins.str
+    """If specified, will only dump modules which match this regexp."""
+    xla_dump_hlo_pass_re: builtins.str
+    """If this flag is specified, will also dump HLO before and after passes that
+    match this regular expression.  Set to .* to dump before/after all passes.
+    """
+    xla_dump_hlo_as_text: builtins.bool
+    """Specifies the format that HLO is dumped in.  Multiple of these may be
+    specified.
+    """
+    xla_dump_hlo_as_proto: builtins.bool
+    xla_dump_hlo_as_dot: builtins.bool
+    xla_dump_hlo_as_url: builtins.bool
+    xla_dump_hlo_as_html: builtins.bool
+    """Dump HLO graphs as an HTML (DOT -> SVG inlined in HTML)"""
+    xla_dump_fusion_visualization: builtins.bool
+    """Dump the visualization of the fusion progress."""
+    xla_dump_hlo_snapshots: builtins.bool
+    """If true, every time an HLO module is run, we will dump an HloSnapshot
+    (essentially, a serialized module plus its inputs) to the --xla_dump_to
+    directory.
+    """
+    xla_dump_include_timestamp: builtins.bool
+    """Include a timestamp in the dumped filenames."""
+    xla_dump_max_hlo_modules: builtins.int
+    """Max number of hlo module dumps in a directory. Set to < 0 for unbounded."""
+    xla_dump_module_metadata: builtins.bool
+    """Dump HloModuleMetadata as a text proto for each HLO module."""
+    xla_dump_compress_protos: builtins.bool
+    """GZip-compress protos dumped via --xla_dump_hlo_as_proto."""
+    xla_dump_hlo_as_long_text: builtins.bool
+    """Dump HLO in long text format. Ignored unless xla_dump_hlo_as_text is true."""
+    xla_gpu_force_conv_nchw: builtins.bool
+    """
+    END flags controlling dumping HLO modules.
+
+    Overrides for XLA GPU's convolution layout heuristic.
+    """
+    xla_gpu_force_conv_nhwc: builtins.bool
+    xla_gpu_dump_llvmir: builtins.bool
+    """Whether to dump llvm ir when compiling to ptx."""
+    xla_dump_enable_mlir_pretty_form: builtins.bool
+    """Whether to dump mlir using pretty print form."""
+    xla_gpu_algorithm_denylist_path: builtins.str
+    """Denylist for cuDNN convolutions."""
+    xla_tpu_detect_nan: builtins.bool
+    """Debug options that trigger execution errors when NaN or Inf are detected."""
+    xla_tpu_detect_inf: builtins.bool
+    xla_cpu_enable_xprof_traceme: builtins.bool
+    """True if TraceMe annotations are enabled for XLA:CPU."""
+    xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found: builtins.bool
+    """It is usually preferable to not fallback to the driver; it can consume more
+    memory, or have bugs.
+    """
+    xla_gpu_asm_extra_flags: builtins.str
+    """Extra parameters to pass the GPU assembler."""
+    xla_multiheap_size_constraint_per_heap: builtins.int
+    """Per-heap size constraint. New heaps will be created if per-heap max size is
+    reached.
+    """
+    xla_detailed_logging: builtins.bool
+    """Enable detailed logging into vlog. If this is disabled, no
+    compilation summary will be printed in the end of computation.
+    """
+    xla_enable_dumping: builtins.bool
+    """Enable HLO dumping. If this is disabled, no HLO modules will be dumped."""
+    xla_gpu_force_compilation_parallelism: builtins.int
+    """Overrides normal multi-threaded compilation setting to use this many
+    threads. Setting to 0 (the default value) means no enforcement.
+    """
+    xla_gpu_enable_llvm_module_compilation_parallelism: builtins.bool
+    xla_gpu_deterministic_ops: builtins.bool
+    """Guarantees run-to-run determinism.
+    This flag implies --xla_gpu_exclude_nondeterministic_ops and in addition
+    disables autotuning.
+    """
+    xla_gpu_all_reduce_combine_threshold_bytes: builtins.int
+    """Size threshold (in bytes) for the GPU collective combiners."""
+    xla_gpu_all_gather_combine_threshold_bytes: builtins.int
+    xla_gpu_reduce_scatter_combine_threshold_bytes: builtins.int
+    xla_gpu_enable_all_gather_combine_by_dim: builtins.bool
+    """Combine all-gather/scatter-reduce ops with the same dimension or
+    irrespective of their dimension.
+    """
+    xla_gpu_enable_reduce_scatter_combine_by_dim: builtins.bool
+    xla_gpu_enable_reassociation_for_converted_ar: builtins.bool
+    """Enable allreduce reassociation on allreduces that are converted to a wider
+    type. The resulting allreduce will be promoted to a wider-typed allreduce.
+    """
+    xla_gpu_all_reduce_blueconnect_num_devices_per_host: builtins.int
+    """Number of devices per host for first stage of BlueConnect decomposition
+    pass. The pass will attempt to decompose all-reduces ops into a
+    ReduceScatter-AllReduce-AllGather sequence, with the initial ReduceScatter
+    being performed over all of the devices in the same host. Set to < 1 to
+    disable all-reduce decomposition.
+    """
+    xla_gpu_enable_while_loop_reduce_scatter_code_motion: builtins.bool
+    """Enable hoisting of reduce-scatter out of while loops."""
+    xla_gpu_collective_inflation_factor: builtins.int
+    """Inflate collective cost by running each collective multiple times."""
+    xla_llvm_force_inline_before_split: builtins.bool
+    """Whether to force inline before llvm module split to get a more balanced
+    splits for parallel compilation.
+    """
+    xla_gpu_enable_cudnn_frontend: builtins.bool
+    """Whether to use the cuDNN frontend API for convolutions when possible."""
+    xla_gpu_enable_cudnn_fmha: builtins.bool
+    xla_gpu_fused_attention_use_cudnn_rng: builtins.bool
+    xla_gpu_enable_cudnn_layer_norm: builtins.bool
+    """Rewrite layer norm patterns into cuDNN library calls."""
+    xla_dump_disable_metadata: builtins.bool
+    """Disable dumping metadata in HLO dumps."""
+    xla_dump_hlo_pipeline_re: builtins.str
+    """If this flag is specified, will only dump HLO before and after passes in
+    the pass pipeline that matches this regular expression. Default empty value
+    enables dumping in all pipelines.
+    """
+    xla_gpu_strict_conv_algorithm_picker: builtins.bool
+    """If true, abort immediately when conv algorithm picker fails, rather than
+    logging a warning and proceeding with fallback.
+    """
+    xla_gpu_enable_custom_fusions: builtins.bool
+    """If true, XLA will try to pattern match subgraphs of HLO operations into
+    custom fusions registered in the current process (pre-compiled hand written
+    kernels, e.g. various GEMM fusions writtent in CUTLASS).
+    """
+    xla_gpu_enable_custom_fusions_re: builtins.str
+    """A regular expression enabling only a subset of custom fusions. Enabled only
+    if `xla_gpu_enable_custom_fusion` set to true.
+    """
+    xla_gpu_enable_dynamic_slice_fusion: builtins.bool
+    """Enables address computation fusion to optimize dynamic-slice and
+    dynamic-update-slice operations around library calls.
+    """
+    xla_gpu_nccl_termination_timeout_seconds: builtins.int
+    """Timeout in seconds before terminating jobs that are stuck in a NCCL
+    Rendezvous. Negative value disables the timeout and will not terminate.
+    """
+    xla_gpu_enable_shared_constants: builtins.bool
+    """Enables shared constants for XLA/GPU. This allows large constants to be
+    shared among multiple GPU executables.
+    """
+    xla_gpu_enable_cublaslt: builtins.bool
+    """Whether to use cuBLASLt for GEMMs on GPUs."""
+    xla_gpu_graph_min_graph_size: builtins.int
+    """This number determines how many moved instructions like fusion kernels are
+    required for a region to be captured as a function to be launched as a GPU
+    graph.
+    """
+    xla_gpu_graph_enable_concurrent_region: builtins.bool
+    """Identify concurrent regions in GPU graphs and execute them concurrently."""
+    xla_gpu_redzone_scratch_max_megabytes: builtins.int
+    """Size threshold (in megabytes) for the GPU redzone scratch allocator."""
+    xla_gpu_redzone_padding_bytes: builtins.int
+    """Amount of padding the redzone allocator will put on one side of each buffer
+    it allocates.  (So the buffer's total size will be increased by 2x this
+    value.)
+
+    Higher values make it more likely that we'll catch an out-of-bounds read or
+    write.  Smaller values consume less memory during autotuning.  Note that a
+    fused cudnn conv has up to 6 total buffers (4 inputs, 1 output, and 1
+    scratch), so this can be multiplied by quite a lot.
+    """
+    xla_cpu_use_acl: builtins.bool
+    """Generate calls to Arm Compute Library in the CPU backend."""
+    xla_cpu_strict_dot_conv_math: builtins.bool
+    """By default, XLA:CPU will run fp16 dot/conv as fp32, as this is generally
+    (much) faster on our hardware.  Set this flag to disable this behavior.
+    """
+    xla_gpu_use_runtime_fusion: builtins.bool
+    """An option to enable using cuDNN runtime compiled fusion kernels which is
+    available and recommended for Ampere+ GPUs.
+    """
+    xla_dump_latency_hiding_schedule: builtins.bool
+    xla_cpu_enable_mlir_tiling_and_fusion: builtins.bool
+    """By default, MLIR lowering will use Linalg elementwise fusion. If this flag
+    is enabled, the pipeline will use tiling, fusion, peeling, vectorization
+    instead.
+    """
+    xla_cpu_enable_custom_matmul_tiling: builtins.bool
+    """XLA:CPU-Next tiling parameters for matmul."""
+    xla_cpu_matmul_tiling_m_dim: builtins.int
+    xla_cpu_matmul_tiling_n_dim: builtins.int
+    xla_cpu_matmul_tiling_k_dim: builtins.int
+    xla_cpu_enable_mlir_fusion_outlining: builtins.bool
+    xla_cpu_enable_experimental_deallocation: builtins.bool
+    """If set, use the experimental deallocation pass from mlir-hlo."""
+    xla_gpu_enable_latency_hiding_scheduler: builtins.bool
+    xla_gpu_enable_highest_priority_async_stream: builtins.bool
+    xla_gpu_enable_analytical_latency_estimator: builtins.bool
+    xla_gpu_lhs_enable_gpu_async_tracker: builtins.bool
+    xla_gpu_pgle_profile_file_or_directory_path: builtins.str
+    xla_gpu_memory_limit_slop_factor: builtins.int
+    xla_gpu_enable_pipelined_collectives: builtins.bool
+    xla_gpu_enable_pipelined_all_reduce: builtins.bool
+    xla_gpu_enable_pipelined_all_gather: builtins.bool
+    xla_gpu_enable_pipelined_reduce_scatter: builtins.bool
+    xla_gpu_enable_pipelined_p2p: builtins.bool
+    xla_gpu_run_post_layout_collective_pipeliner: builtins.bool
+    xla_gpu_collective_permute_decomposer_threshold: builtins.int
+    """The minimum data size in bytes to trigger collective-permute-decomposer
+    transformation.
+    """
+    xla_partitioning_algorithm: global___DebugOptions.PartitioningAlgorithm.ValueType
+    """The partitioning algorithm to be used in the PartitionAssignment pass."""
+    xla_gpu_enable_triton_gemm: builtins.bool
+    xla_gpu_enable_cudnn_int8x32_convolution_reordering: builtins.bool
+    xla_gpu_triton_gemm_any: builtins.bool
+    """Creates triton fusion for all supported gemms.
+    To make sure only triton gemm is chosen by the autotuner run with
+    `xla_gpu_cublas_fallback` set to false.
+    """
+    xla_gpu_exhaustive_tiling_search: builtins.bool
+    xla_gpu_enable_priority_fusion: builtins.bool
+    xla_gpu_dump_autotune_results_to: builtins.str
+    """File to write autotune results to. It will be a binary file unless the name
+    ends with .txt or .textproto. Warning: The results are written at every
+    compilation, possibly multiple times per process. This only works on CUDA.
+    """
+    xla_gpu_load_autotune_results_from: builtins.str
+    """File to load autotune results from. It will be considered a binary file
+    unless the name ends with .txt or .textproto. At most one loading will
+    happen during the lifetime of one process, even if the first one is
+    unsuccessful or different file paths are passed here. This only works on
+    CUDA.
+    """
+    xla_gpu_target_config_filename: builtins.str
+    """Description of the target platform in GpuTargetConfigProto format; if
+    provided, deviceless compilation is assumed, and the current device is
+    ignored.
+    """
+    xla_gpu_auto_spmd_partitioning_memory_budget_gb: builtins.int
+    """Memory budget in GB per device for AutoSharding."""
+    xla_gpu_auto_spmd_partitioning_memory_budget_ratio: builtins.float
+    """See the definition of the
+    xla_gpu_auto_spmd_partitioning_memory_budget_ratio flag for the meaning of
+    this field.
+    """
+    xla_gpu_triton_gemm_disable_reduced_precision_reduction: builtins.bool
+    xla_gpu_triton_fusion_level: builtins.int
+    xla_gpu_dump_autotuned_gemm_fusions: builtins.bool
+    xla_gpu_override_gemm_autotuner: builtins.str
+    xla_gpu_copy_insertion_use_region_analysis: builtins.bool
+    xla_gpu_collect_cost_model_stats: builtins.bool
+    """If true, each fusion instruction will have a cost model runtime estimate in
+    backend config after compilation.
+    """
+    xla_gpu_enable_split_k_autotuning: builtins.bool
+    xla_gpu_enable_reduction_epilogue_fusion: builtins.bool
+    """Whether reduction epilogue fusion is enabled in fusion passes."""
+    xla_gpu_enable_nccl_clique_optimization: builtins.bool
+    """Allow early return when acquiring NCCL cliques."""
+    xla_gpu_mock_custom_calls: builtins.bool
+    """Replace custom calls with noop operations."""
+    xla_gpu_cublas_fallback: builtins.bool
+    """Allow Triton GEMM autotuning to fall back to cuBLAS when that is
+    faster.
+    """
+    xla_gpu_enable_while_loop_double_buffering: builtins.bool
+    """Enable double buffering for loops."""
+    xla_gpu_enable_while_loop_unrolling: global___DebugOptions.WhileLoopUnrolling.ValueType
+    """Determine the while loop unrolling scheme."""
+    xla_gpu_ensure_minor_dot_contraction_dims: builtins.bool
+    """Change the layout of the second triton dot operand to be column major.
+    Only works for (bf16 x bf16) -> bf16.
+    """
+    xla_gpu_filter_kernels_spilling_registers_on_autotuning: builtins.bool
+    """Filter out kernels that spill registers during autotuning."""
+    xla_debug_buffer_assignment_show_max: builtins.int
+    """Maximum number of buffers to print when debugging buffer assignment."""
+    xla_gpu_llvm_verification_level: builtins.int
+    xla_gpu_enable_cub_radix_sort: builtins.bool
+    """Enable radix sort using CUB."""
+    xla_gpu_threshold_for_windowed_einsum_mib: builtins.int
+    """Threshold to enable windowed einsum (collective matmul) in MB."""
+    xla_gpu_enable_triton_hopper: builtins.bool
+    """Enables currently disabled features within Triton for Hopper."""
+    xla_gpu_enable_nccl_user_buffers: builtins.bool
+    """Enable NCCL user buffers."""
+    xla_gpu_enable_nccl_comm_splitting: builtins.bool
+    """Enable NCCL communicator splitting."""
+    xla_gpu_enable_nccl_per_stream_comms: builtins.bool
+    """Enable NCCL per stream communicators."""
+    xla_gpu_enable_libnvptxcompiler: builtins.bool
+    """If enabled, uses the libnvptxcompiler library to compile PTX to cuBIN."""
+    xla_gpu_enable_dot_strength_reduction: builtins.bool
+    xla_gpu_multi_streamed_windowed_einsum: builtins.bool
+    """Whether to use multiple compute streams to run windowed einsum."""
+    xla_gpu_enable_bf16_6way_gemm: builtins.bool
+    """If enabled, uses bf16_6way gemm to compute F32 gemm."""
+    xla_gpu_enable_bf16_3way_gemm: builtins.bool
+    """If enabled, uses bf16_3way gemm to compute F32 gemm."""
+    xla_gpu_nccl_collective_max_nchannels: builtins.int
+    """Specify the maximum number of channels(SMs) NCCL
+    will use for collective operations.
+    """
+    xla_gpu_nccl_p2p_max_nchannels: builtins.int
+    """Specify the maximum number of channels(SMs) NCCL
+    will use for p2p operations.
+    """
+    xla_gpu_mlir_emitter_level: builtins.int
+    """Choose the level of mlir emitters that are enabled.
+    Current levels:
+    0: Disabled.
+    1: Loop emitter
+    2: + Loop-like emitters
+    3: + Transpose
+    4: + Reduce
+    """
+    xla_gpu_gemm_rewrite_size_threshold: builtins.int
+    """Threshold to rewrite matmul to cuBLAS or Triton (minimum combined number of
+    elements of both matrices in non-batch dimensions to be considered for a
+    rewrite).
+    """
+    xla_gpu_require_complete_aot_autotune_results: builtins.bool
+    """If true, will require complete AOT autotuning results; in the case of
+    missing AOT result, the model will not be compiled or executed, a
+    `NotFound` error will be returned.
+    """
+    xla_gpu_cudnn_gemm_fusion_level: builtins.int
+    """Let GEMM fusion autotuning probe cuDNN as a backend.
+    Current levels:
+    0: Disabled.
+    1: Fusions of GEMM, elementwise, transpose/reshape operations.
+    2: + Broadcasts, slicing.
+    3: + Nontrivial noncontracting dimension reshapes/transposes.
+    """
+    xla_gpu_use_memcpy_local_p2p: builtins.bool
+    """This instructs the runtime whether to use
+    memcpy for p2p communication when source and
+    target are located within a node(nvlink).
+    """
+    xla_gpu_autotune_max_solutions: builtins.int
+    """If non-zero, limits the number of solutions to be used by GEMM autotuner.
+    This might be useful if underlying math library returns too many GEMM
+    solutions.
+    """
+    xla_dump_large_constants: builtins.bool
+    """If true, large constants will be printed out when dumping HLOs."""
+    xla_gpu_verify_triton_fusion_numerics: builtins.bool
+    """If true, will verify that the numerical results of Triton fusions match
+    the results of regular emitters.
+    """
+    xla_gpu_dump_autotune_logs_to: builtins.str
+    """File to write autotune logs to. It will stored in txt format."""
+    xla_reduce_window_rewrite_base_length: builtins.int
+    """Base length to rewrite the reduce window to, no rewrite if set to 0."""
+    xla_gpu_enable_host_memory_offloading: builtins.bool
+    """If true, will enable host memory offloading on a device."""
+    xla_gpu_exclude_nondeterministic_ops: builtins.bool
+    """Excludes non-deterministic ops from compiled executables.
+    Unlike --xla_gpu_deterministic_ops does not disable autotuning - the
+    compilation itself can be non-deterministic.
+    At present, the HLO op SelectAndScatter does not have a
+    deterministic XLA:GPU implementation.
+    Compilation errors out if SelectAndScatter is encountered.
+    Scatter ops can non-deterministic by default; these get converted to
+    a deterministic implementation.
+    """
+    xla_gpu_nccl_terminate_on_error: builtins.bool
+    """If true, Nccl errors will terminate the process."""
+    xla_gpu_shard_autotuning: builtins.bool
+    xla_gpu_enable_approx_costly_collectives: builtins.bool
+    xla_gpu_kernel_cache_file: builtins.str
+    xla_gpu_unsafe_pipelined_loop_annotator: builtins.bool
+    """Recognises rotate-right patterns (slice, slice, concat) within a while
+    loop and labels the while loop as a pipelined while loop. This is an
+    unsafe flag.
+    """
+    xla_gpu_per_fusion_autotune_cache_dir: builtins.str
+    xla_cmd_buffer_trace_cache_size: builtins.int
+    """The command buffer trace cache size, increasing the cache size may
+    sometimes reduces the chances of doing command buffer tracing for
+    updating command buffer instance.
+    """
+    xla_gpu_temp_buffer_use_separate_color: builtins.bool
+    """Enable this flag will use a separate memory space color for
+    temp buffer, and then will use separate memory allocator to allocate it,
+    as there is no other memory allocation interference,
+    it will allocate temp buffer to some fix address on every iteration,
+    which is good for cuda-graph perf.
+    """
+    xla_syntax_sugar_async_ops: builtins.bool
+    """This flag is used for controlling HLO dumping and NVTX marker. If turned
+    on, both HLO dumping and NVTX marker will use syntactic sugar wrappers
+    as op names, while the actual op names will be shown if turned off.
+
+    Here is an example HLO excerpt with the flag off:
+
+      async_computation {
+       param_0 = f32[1,4,8]{1,0,2} parameter(0)
+       ROOT all-to-all.3.1 = f32[1,4,8]{1,0,2} all-to-all(param_0),
+                             replica_groups={{0,1,2,3,4,5,6,7}}, dimensions={2}
+      }
+     ...
+
+     all-to-all-start =
+       ((f32[1,4,8]{1,0,2}), f32[1,4,8]{1,0,2}) async-start(bitcast.24.0),
+       calls=async_computation, backend_config={...}
+     all-to-all-done = f32[1,4,8]{1,0,2} async-done(all-to-all-start)
+
+    and with the flag on:
+
+     all-to-all-start = ((f32[1,4,8]{1,0,2}), f32[1,4,8]{1,0,2})
+                        all-to-all-start(bitcast.24.0),
+                        replica_groups={{0,1,2,3,4,5,6,7}}, dimensions={2},
+                        backend_config={...}
+     all-to-all-done = f32[1,4,8]{1,0,2} all-to-all-done(all-to-all-start)
+    """
+    xla_gpu_autotune_gemm_rtol: builtins.float
+    """Relative precision for comparing different GEMM solutions"""
+    xla_enable_command_buffers_during_profiling: builtins.bool
+    """Allow launching command buffers while profiling active.
+    When disabled, execute in op-by-op mode.
+    TODO(b/355487968): Remove this option when validation complete.
+    """
+    xla_gpu_cudnn_gemm_max_plans: builtins.int
+    """Limit for the number of kernel configurations (plans) to use during
+    autotuning of cuDNN GEMM fusions. The more - the slower the autotuning
+    but potentially higher the performance.
+    """
+    xla_gpu_enable_libnvjitlink: builtins.bool
+    """If enabled, uses the libnvjitlink library for PTX compilation and linking"""
+    xla_gpu_enable_triton_gemm_int4: builtins.bool
+    """If enabled, generates triton gemm kernels for int4 inputs."""
+    xla_gpu_async_dot: builtins.bool
+    """If true, XLA will wrap `dot` operations into async computations in an
+    effort to parallelize matrix operations.
+    """
+    xla_gpu_enable_pgle_accuracy_checker: builtins.bool
+    """Enables strict PGLE checking. If an FDO profile is specified and latency
+    hiding scheduler encounters missing instructions in the profile
+    compilation will halt.
+    """
+    xla_gpu_executable_warn_stuck_timeout_seconds: builtins.int
+    """Timeouts for RendezvousSingle stuck warning and termination."""
+    xla_gpu_executable_terminate_timeout_seconds: builtins.int
+    xla_experimental_ignore_channel_id: builtins.bool
+    """Whether to ignore channel ids(including verifier channel id checks)
+    for collectives in the given HLO.
+    """
+    @property
+    def xla_disable_hlo_passes(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
+        """List of HLO passes to disable/enable. These names must exactly match the
+        pass names as specified by the HloPassInterface::name() method.
+
+        At least one of xla_disable_hlo_passes and xla_enable_hlo_passes_only must
+        be empty.
+        """
+
+    @property
+    def xla_enable_hlo_passes_only(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]: ...
+    @property
+    def xla_gpu_ptx_file(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
+        """Paths to files with ptx code."""
+
+    @property
+    def xla_gpu_llvm_ir_file(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
+        """Paths to files with LLVM code."""
+
+    @property
+    def xla_gpu_disable_async_collectives(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[global___DebugOptions.CollectiveOpType.ValueType]: ...
+    @property
+    def xla_gpu_enable_command_buffer(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[global___DebugOptions.CommandBufferCmdType.ValueType]:
+        """Determine the types of commands that are recorded into command buffers."""
+
+    @property
+    def legacy_command_buffer_custom_call_targets(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
+        """Custom call targets with legacy registry API (non FFI API),
+        that support recording to command buffer custom command,
+        i.e., custom call target supports cuda-graph capturing for CUDA devices.
+        This flag is read if CUSTOM_CALL command type is recorded into
+        command buffer.
+        """
+
+    @property
+    def xla_backend_extra_options(self) -> google.protobuf.internal.containers.ScalarMap[builtins.str, builtins.str]:
+        """Next id: 331
+
+        Extra options to pass to the compilation backend (e.g. LLVM); specific
+        interpretation of these values is left to the backend.
+        """
+
+    def __init__(
+        self,
+        *,
+        xla_cpu_enable_concurrency_optimized_scheduler: builtins.bool | None = ...,
+        xla_cpu_enable_fast_math: builtins.bool | None = ...,
+        xla_cpu_enable_fast_min_max: builtins.bool | None = ...,
+        xla_cpu_fast_math_honor_division: builtins.bool | None = ...,
+        xla_cpu_fast_math_honor_functions: builtins.bool | None = ...,
+        xla_cpu_fast_math_honor_infs: builtins.bool | None = ...,
+        xla_cpu_fast_math_honor_nans: builtins.bool | None = ...,
+        xla_cpu_use_thunk_runtime: builtins.bool | None = ...,
+        xla_cpu_parallel_codegen_split_count: builtins.int | None = ...,
+        xla_cpu_prefer_vector_width: builtins.int | None = ...,
+        xla_gpu_experimental_autotune_cache_mode: global___DebugOptions.AutotuneCacheMode.ValueType | None = ...,
+        xla_gpu_experimental_disable_binary_libraries: builtins.bool | None = ...,
+        xla_gpu_experimental_enable_triton_softmax_priority_fusion: builtins.bool | None = ...,
+        xla_gpu_unsupported_enable_triton_gemm: builtins.bool | None = ...,
+        xla_hlo_graph_addresses: builtins.bool | None = ...,
+        xla_hlo_profile: builtins.bool | None = ...,
+        xla_disable_hlo_passes: collections.abc.Iterable[builtins.str] | None = ...,
+        xla_enable_hlo_passes_only: collections.abc.Iterable[builtins.str] | None = ...,
+        xla_disable_all_hlo_passes: builtins.bool | None = ...,
+        xla_backend_optimization_level: builtins.int | None = ...,
+        xla_embed_ir_in_executable: builtins.bool | None = ...,
+        xla_eliminate_hlo_implicit_broadcast: builtins.bool | None = ...,
+        xla_cpu_multi_thread_eigen: builtins.bool | None = ...,
+        xla_gpu_cuda_data_dir: builtins.str | None = ...,
+        xla_gpu_ftz: builtins.bool | None = ...,
+        xla_llvm_enable_alias_scope_metadata: builtins.bool | None = ...,
+        xla_llvm_enable_noalias_metadata: builtins.bool | None = ...,
+        xla_llvm_enable_invariant_load_metadata: builtins.bool | None = ...,
+        xla_llvm_disable_expensive_passes: builtins.bool | None = ...,
+        xla_test_all_output_layouts: builtins.bool | None = ...,
+        xla_test_all_input_layouts: builtins.bool | None = ...,
+        xla_hlo_graph_sharding_color: builtins.bool | None = ...,
+        xla_cpu_use_mkl_dnn: builtins.bool | None = ...,
+        xla_gpu_enable_fast_min_max: builtins.bool | None = ...,
+        xla_allow_excess_precision: builtins.bool | None = ...,
+        xla_gpu_crash_on_verification_failures: builtins.bool | None = ...,
+        xla_gpu_autotune_level: builtins.int | None = ...,
+        xla_force_host_platform_device_count: builtins.int | None = ...,
+        xla_gpu_disable_gpuasm_optimizations: builtins.bool | None = ...,
+        xla_gpu_shape_checks: global___DebugOptions.ShapeChecks.ValueType | None = ...,
+        xla_hlo_evaluator_use_fast_path: builtins.bool | None = ...,
+        xla_allow_scalar_index_dynamic_ops: builtins.bool | None = ...,
+        xla_step_marker_location: global___DebugOptions.StepMarkerLocation.ValueType | None = ...,
+        xla_dump_to: builtins.str | None = ...,
+        xla_dump_hlo_module_re: builtins.str | None = ...,
+        xla_dump_hlo_pass_re: builtins.str | None = ...,
+        xla_dump_hlo_as_text: builtins.bool | None = ...,
+        xla_dump_hlo_as_proto: builtins.bool | None = ...,
+        xla_dump_hlo_as_dot: builtins.bool | None = ...,
+        xla_dump_hlo_as_url: builtins.bool | None = ...,
+        xla_dump_hlo_as_html: builtins.bool | None = ...,
+        xla_dump_fusion_visualization: builtins.bool | None = ...,
+        xla_dump_hlo_snapshots: builtins.bool | None = ...,
+        xla_dump_include_timestamp: builtins.bool | None = ...,
+        xla_dump_max_hlo_modules: builtins.int | None = ...,
+        xla_dump_module_metadata: builtins.bool | None = ...,
+        xla_dump_compress_protos: builtins.bool | None = ...,
+        xla_dump_hlo_as_long_text: builtins.bool | None = ...,
+        xla_gpu_force_conv_nchw: builtins.bool | None = ...,
+        xla_gpu_force_conv_nhwc: builtins.bool | None = ...,
+        xla_gpu_ptx_file: collections.abc.Iterable[builtins.str] | None = ...,
+        xla_gpu_dump_llvmir: builtins.bool | None = ...,
+        xla_dump_enable_mlir_pretty_form: builtins.bool | None = ...,
+        xla_gpu_algorithm_denylist_path: builtins.str | None = ...,
+        xla_tpu_detect_nan: builtins.bool | None = ...,
+        xla_tpu_detect_inf: builtins.bool | None = ...,
+        xla_cpu_enable_xprof_traceme: builtins.bool | None = ...,
+        xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found: builtins.bool | None = ...,
+        xla_gpu_asm_extra_flags: builtins.str | None = ...,
+        xla_multiheap_size_constraint_per_heap: builtins.int | None = ...,
+        xla_detailed_logging: builtins.bool | None = ...,
+        xla_enable_dumping: builtins.bool | None = ...,
+        xla_gpu_force_compilation_parallelism: builtins.int | None = ...,
+        xla_gpu_enable_llvm_module_compilation_parallelism: builtins.bool | None = ...,
+        xla_gpu_deterministic_ops: builtins.bool | None = ...,
+        xla_gpu_llvm_ir_file: collections.abc.Iterable[builtins.str] | None = ...,
+        xla_gpu_disable_async_collectives: (
+            collections.abc.Iterable[global___DebugOptions.CollectiveOpType.ValueType] | None
+        ) = ...,
+        xla_gpu_all_reduce_combine_threshold_bytes: builtins.int | None = ...,
+        xla_gpu_all_gather_combine_threshold_bytes: builtins.int | None = ...,
+        xla_gpu_reduce_scatter_combine_threshold_bytes: builtins.int | None = ...,
+        xla_gpu_enable_all_gather_combine_by_dim: builtins.bool | None = ...,
+        xla_gpu_enable_reduce_scatter_combine_by_dim: builtins.bool | None = ...,
+        xla_gpu_enable_reassociation_for_converted_ar: builtins.bool | None = ...,
+        xla_gpu_all_reduce_blueconnect_num_devices_per_host: builtins.int | None = ...,
+        xla_gpu_enable_while_loop_reduce_scatter_code_motion: builtins.bool | None = ...,
+        xla_gpu_collective_inflation_factor: builtins.int | None = ...,
+        xla_llvm_force_inline_before_split: builtins.bool | None = ...,
+        xla_gpu_enable_cudnn_frontend: builtins.bool | None = ...,
+        xla_gpu_enable_cudnn_fmha: builtins.bool | None = ...,
+        xla_gpu_fused_attention_use_cudnn_rng: builtins.bool | None = ...,
+        xla_gpu_enable_cudnn_layer_norm: builtins.bool | None = ...,
+        xla_dump_disable_metadata: builtins.bool | None = ...,
+        xla_dump_hlo_pipeline_re: builtins.str | None = ...,
+        xla_gpu_strict_conv_algorithm_picker: builtins.bool | None = ...,
+        xla_gpu_enable_custom_fusions: builtins.bool | None = ...,
+        xla_gpu_enable_custom_fusions_re: builtins.str | None = ...,
+        xla_gpu_enable_dynamic_slice_fusion: builtins.bool | None = ...,
+        xla_gpu_nccl_termination_timeout_seconds: builtins.int | None = ...,
+        xla_gpu_enable_shared_constants: builtins.bool | None = ...,
+        xla_gpu_enable_cublaslt: builtins.bool | None = ...,
+        xla_gpu_enable_command_buffer: (
+            collections.abc.Iterable[global___DebugOptions.CommandBufferCmdType.ValueType] | None
+        ) = ...,
+        xla_gpu_graph_min_graph_size: builtins.int | None = ...,
+        xla_gpu_graph_enable_concurrent_region: builtins.bool | None = ...,
+        xla_gpu_redzone_scratch_max_megabytes: builtins.int | None = ...,
+        xla_gpu_redzone_padding_bytes: builtins.int | None = ...,
+        xla_cpu_use_acl: builtins.bool | None = ...,
+        xla_cpu_strict_dot_conv_math: builtins.bool | None = ...,
+        xla_gpu_use_runtime_fusion: builtins.bool | None = ...,
+        xla_dump_latency_hiding_schedule: builtins.bool | None = ...,
+        xla_cpu_enable_mlir_tiling_and_fusion: builtins.bool | None = ...,
+        xla_cpu_enable_custom_matmul_tiling: builtins.bool | None = ...,
+        xla_cpu_matmul_tiling_m_dim: builtins.int | None = ...,
+        xla_cpu_matmul_tiling_n_dim: builtins.int | None = ...,
+        xla_cpu_matmul_tiling_k_dim: builtins.int | None = ...,
+        xla_cpu_enable_mlir_fusion_outlining: builtins.bool | None = ...,
+        xla_cpu_enable_experimental_deallocation: builtins.bool | None = ...,
+        xla_gpu_enable_latency_hiding_scheduler: builtins.bool | None = ...,
+        xla_gpu_enable_highest_priority_async_stream: builtins.bool | None = ...,
+        xla_gpu_enable_analytical_latency_estimator: builtins.bool | None = ...,
+        xla_gpu_lhs_enable_gpu_async_tracker: builtins.bool | None = ...,
+        xla_gpu_pgle_profile_file_or_directory_path: builtins.str | None = ...,
+        xla_gpu_memory_limit_slop_factor: builtins.int | None = ...,
+        xla_gpu_enable_pipelined_collectives: builtins.bool | None = ...,
+        xla_gpu_enable_pipelined_all_reduce: builtins.bool | None = ...,
+        xla_gpu_enable_pipelined_all_gather: builtins.bool | None = ...,
+        xla_gpu_enable_pipelined_reduce_scatter: builtins.bool | None = ...,
+        xla_gpu_enable_pipelined_p2p: builtins.bool | None = ...,
+        xla_gpu_run_post_layout_collective_pipeliner: builtins.bool | None = ...,
+        xla_gpu_collective_permute_decomposer_threshold: builtins.int | None = ...,
+        xla_partitioning_algorithm: global___DebugOptions.PartitioningAlgorithm.ValueType | None = ...,
+        xla_gpu_enable_triton_gemm: builtins.bool | None = ...,
+        xla_gpu_enable_cudnn_int8x32_convolution_reordering: builtins.bool | None = ...,
+        xla_gpu_triton_gemm_any: builtins.bool | None = ...,
+        xla_gpu_exhaustive_tiling_search: builtins.bool | None = ...,
+        xla_gpu_enable_priority_fusion: builtins.bool | None = ...,
+        xla_gpu_dump_autotune_results_to: builtins.str | None = ...,
+        xla_gpu_load_autotune_results_from: builtins.str | None = ...,
+        xla_gpu_target_config_filename: builtins.str | None = ...,
+        xla_gpu_auto_spmd_partitioning_memory_budget_gb: builtins.int | None = ...,
+        xla_gpu_auto_spmd_partitioning_memory_budget_ratio: builtins.float | None = ...,
+        xla_gpu_triton_gemm_disable_reduced_precision_reduction: builtins.bool | None = ...,
+        xla_gpu_triton_fusion_level: builtins.int | None = ...,
+        xla_gpu_dump_autotuned_gemm_fusions: builtins.bool | None = ...,
+        xla_gpu_override_gemm_autotuner: builtins.str | None = ...,
+        xla_gpu_copy_insertion_use_region_analysis: builtins.bool | None = ...,
+        xla_gpu_collect_cost_model_stats: builtins.bool | None = ...,
+        xla_gpu_enable_split_k_autotuning: builtins.bool | None = ...,
+        xla_gpu_enable_reduction_epilogue_fusion: builtins.bool | None = ...,
+        xla_gpu_enable_nccl_clique_optimization: builtins.bool | None = ...,
+        xla_gpu_mock_custom_calls: builtins.bool | None = ...,
+        xla_gpu_cublas_fallback: builtins.bool | None = ...,
+        xla_gpu_enable_while_loop_double_buffering: builtins.bool | None = ...,
+        xla_gpu_enable_while_loop_unrolling: global___DebugOptions.WhileLoopUnrolling.ValueType | None = ...,
+        xla_gpu_ensure_minor_dot_contraction_dims: builtins.bool | None = ...,
+        xla_gpu_filter_kernels_spilling_registers_on_autotuning: builtins.bool | None = ...,
+        xla_debug_buffer_assignment_show_max: builtins.int | None = ...,
+        xla_gpu_llvm_verification_level: builtins.int | None = ...,
+        xla_gpu_enable_cub_radix_sort: builtins.bool | None = ...,
+        xla_gpu_threshold_for_windowed_einsum_mib: builtins.int | None = ...,
+        xla_gpu_enable_triton_hopper: builtins.bool | None = ...,
+        xla_gpu_enable_nccl_user_buffers: builtins.bool | None = ...,
+        xla_gpu_enable_nccl_comm_splitting: builtins.bool | None = ...,
+        xla_gpu_enable_nccl_per_stream_comms: builtins.bool | None = ...,
+        xla_gpu_enable_libnvptxcompiler: builtins.bool | None = ...,
+        xla_gpu_enable_dot_strength_reduction: builtins.bool | None = ...,
+        xla_gpu_multi_streamed_windowed_einsum: builtins.bool | None = ...,
+        xla_gpu_enable_bf16_6way_gemm: builtins.bool | None = ...,
+        xla_gpu_enable_bf16_3way_gemm: builtins.bool | None = ...,
+        xla_gpu_nccl_collective_max_nchannels: builtins.int | None = ...,
+        xla_gpu_nccl_p2p_max_nchannels: builtins.int | None = ...,
+        xla_gpu_mlir_emitter_level: builtins.int | None = ...,
+        xla_gpu_gemm_rewrite_size_threshold: builtins.int | None = ...,
+        xla_gpu_require_complete_aot_autotune_results: builtins.bool | None = ...,
+        xla_gpu_cudnn_gemm_fusion_level: builtins.int | None = ...,
+        xla_gpu_use_memcpy_local_p2p: builtins.bool | None = ...,
+        xla_gpu_autotune_max_solutions: builtins.int | None = ...,
+        xla_dump_large_constants: builtins.bool | None = ...,
+        xla_gpu_verify_triton_fusion_numerics: builtins.bool | None = ...,
+        xla_gpu_dump_autotune_logs_to: builtins.str | None = ...,
+        xla_reduce_window_rewrite_base_length: builtins.int | None = ...,
+        xla_gpu_enable_host_memory_offloading: builtins.bool | None = ...,
+        xla_gpu_exclude_nondeterministic_ops: builtins.bool | None = ...,
+        xla_gpu_nccl_terminate_on_error: builtins.bool | None = ...,
+        xla_gpu_shard_autotuning: builtins.bool | None = ...,
+        xla_gpu_enable_approx_costly_collectives: builtins.bool | None = ...,
+        xla_gpu_kernel_cache_file: builtins.str | None = ...,
+        xla_gpu_unsafe_pipelined_loop_annotator: builtins.bool | None = ...,
+        xla_gpu_per_fusion_autotune_cache_dir: builtins.str | None = ...,
+        xla_cmd_buffer_trace_cache_size: builtins.int | None = ...,
+        xla_gpu_temp_buffer_use_separate_color: builtins.bool | None = ...,
+        legacy_command_buffer_custom_call_targets: collections.abc.Iterable[builtins.str] | None = ...,
+        xla_syntax_sugar_async_ops: builtins.bool | None = ...,
+        xla_gpu_autotune_gemm_rtol: builtins.float | None = ...,
+        xla_enable_command_buffers_during_profiling: builtins.bool | None = ...,
+        xla_gpu_cudnn_gemm_max_plans: builtins.int | None = ...,
+        xla_gpu_enable_libnvjitlink: builtins.bool | None = ...,
+        xla_gpu_enable_triton_gemm_int4: builtins.bool | None = ...,
+        xla_gpu_async_dot: builtins.bool | None = ...,
+        xla_gpu_enable_pgle_accuracy_checker: builtins.bool | None = ...,
+        xla_gpu_executable_warn_stuck_timeout_seconds: builtins.int | None = ...,
+        xla_gpu_executable_terminate_timeout_seconds: builtins.int | None = ...,
+        xla_experimental_ignore_channel_id: builtins.bool | None = ...,
+        xla_backend_extra_options: collections.abc.Mapping[builtins.str, builtins.str] | None = ...,
+    ) -> None: ...
+    def ClearField(
+        self,
+        field_name: typing.Literal[
+            "legacy_command_buffer_custom_call_targets",
+            b"legacy_command_buffer_custom_call_targets",
+            "xla_allow_excess_precision",
+            b"xla_allow_excess_precision",
+            "xla_allow_scalar_index_dynamic_ops",
+            b"xla_allow_scalar_index_dynamic_ops",
+            "xla_backend_extra_options",
+            b"xla_backend_extra_options",
+            "xla_backend_optimization_level",
+            b"xla_backend_optimization_level",
+            "xla_cmd_buffer_trace_cache_size",
+            b"xla_cmd_buffer_trace_cache_size",
+            "xla_cpu_enable_concurrency_optimized_scheduler",
+            b"xla_cpu_enable_concurrency_optimized_scheduler",
+            "xla_cpu_enable_custom_matmul_tiling",
+            b"xla_cpu_enable_custom_matmul_tiling",
+            "xla_cpu_enable_experimental_deallocation",
+            b"xla_cpu_enable_experimental_deallocation",
+            "xla_cpu_enable_fast_math",
+            b"xla_cpu_enable_fast_math",
+            "xla_cpu_enable_fast_min_max",
+            b"xla_cpu_enable_fast_min_max",
+            "xla_cpu_enable_mlir_fusion_outlining",
+            b"xla_cpu_enable_mlir_fusion_outlining",
+            "xla_cpu_enable_mlir_tiling_and_fusion",
+            b"xla_cpu_enable_mlir_tiling_and_fusion",
+            "xla_cpu_enable_xprof_traceme",
+            b"xla_cpu_enable_xprof_traceme",
+            "xla_cpu_fast_math_honor_division",
+            b"xla_cpu_fast_math_honor_division",
+            "xla_cpu_fast_math_honor_functions",
+            b"xla_cpu_fast_math_honor_functions",
+            "xla_cpu_fast_math_honor_infs",
+            b"xla_cpu_fast_math_honor_infs",
+            "xla_cpu_fast_math_honor_nans",
+            b"xla_cpu_fast_math_honor_nans",
+            "xla_cpu_matmul_tiling_k_dim",
+            b"xla_cpu_matmul_tiling_k_dim",
+            "xla_cpu_matmul_tiling_m_dim",
+            b"xla_cpu_matmul_tiling_m_dim",
+            "xla_cpu_matmul_tiling_n_dim",
+            b"xla_cpu_matmul_tiling_n_dim",
+            "xla_cpu_multi_thread_eigen",
+            b"xla_cpu_multi_thread_eigen",
+            "xla_cpu_parallel_codegen_split_count",
+            b"xla_cpu_parallel_codegen_split_count",
+            "xla_cpu_prefer_vector_width",
+            b"xla_cpu_prefer_vector_width",
+            "xla_cpu_strict_dot_conv_math",
+            b"xla_cpu_strict_dot_conv_math",
+            "xla_cpu_use_acl",
+            b"xla_cpu_use_acl",
+            "xla_cpu_use_mkl_dnn",
+            b"xla_cpu_use_mkl_dnn",
+            "xla_cpu_use_thunk_runtime",
+            b"xla_cpu_use_thunk_runtime",
+            "xla_debug_buffer_assignment_show_max",
+            b"xla_debug_buffer_assignment_show_max",
+            "xla_detailed_logging",
+            b"xla_detailed_logging",
+            "xla_disable_all_hlo_passes",
+            b"xla_disable_all_hlo_passes",
+            "xla_disable_hlo_passes",
+            b"xla_disable_hlo_passes",
+            "xla_dump_compress_protos",
+            b"xla_dump_compress_protos",
+            "xla_dump_disable_metadata",
+            b"xla_dump_disable_metadata",
+            "xla_dump_enable_mlir_pretty_form",
+            b"xla_dump_enable_mlir_pretty_form",
+            "xla_dump_fusion_visualization",
+            b"xla_dump_fusion_visualization",
+            "xla_dump_hlo_as_dot",
+            b"xla_dump_hlo_as_dot",
+            "xla_dump_hlo_as_html",
+            b"xla_dump_hlo_as_html",
+            "xla_dump_hlo_as_long_text",
+            b"xla_dump_hlo_as_long_text",
+            "xla_dump_hlo_as_proto",
+            b"xla_dump_hlo_as_proto",
+            "xla_dump_hlo_as_text",
+            b"xla_dump_hlo_as_text",
+            "xla_dump_hlo_as_url",
+            b"xla_dump_hlo_as_url",
+            "xla_dump_hlo_module_re",
+            b"xla_dump_hlo_module_re",
+            "xla_dump_hlo_pass_re",
+            b"xla_dump_hlo_pass_re",
+            "xla_dump_hlo_pipeline_re",
+            b"xla_dump_hlo_pipeline_re",
+            "xla_dump_hlo_snapshots",
+            b"xla_dump_hlo_snapshots",
+            "xla_dump_include_timestamp",
+            b"xla_dump_include_timestamp",
+            "xla_dump_large_constants",
+            b"xla_dump_large_constants",
+            "xla_dump_latency_hiding_schedule",
+            b"xla_dump_latency_hiding_schedule",
+            "xla_dump_max_hlo_modules",
+            b"xla_dump_max_hlo_modules",
+            "xla_dump_module_metadata",
+            b"xla_dump_module_metadata",
+            "xla_dump_to",
+            b"xla_dump_to",
+            "xla_eliminate_hlo_implicit_broadcast",
+            b"xla_eliminate_hlo_implicit_broadcast",
+            "xla_embed_ir_in_executable",
+            b"xla_embed_ir_in_executable",
+            "xla_enable_command_buffers_during_profiling",
+            b"xla_enable_command_buffers_during_profiling",
+            "xla_enable_dumping",
+            b"xla_enable_dumping",
+            "xla_enable_hlo_passes_only",
+            b"xla_enable_hlo_passes_only",
+            "xla_experimental_ignore_channel_id",
+            b"xla_experimental_ignore_channel_id",
+            "xla_force_host_platform_device_count",
+            b"xla_force_host_platform_device_count",
+            "xla_gpu_algorithm_denylist_path",
+            b"xla_gpu_algorithm_denylist_path",
+            "xla_gpu_all_gather_combine_threshold_bytes",
+            b"xla_gpu_all_gather_combine_threshold_bytes",
+            "xla_gpu_all_reduce_blueconnect_num_devices_per_host",
+            b"xla_gpu_all_reduce_blueconnect_num_devices_per_host",
+            "xla_gpu_all_reduce_combine_threshold_bytes",
+            b"xla_gpu_all_reduce_combine_threshold_bytes",
+            "xla_gpu_asm_extra_flags",
+            b"xla_gpu_asm_extra_flags",
+            "xla_gpu_async_dot",
+            b"xla_gpu_async_dot",
+            "xla_gpu_auto_spmd_partitioning_memory_budget_gb",
+            b"xla_gpu_auto_spmd_partitioning_memory_budget_gb",
+            "xla_gpu_auto_spmd_partitioning_memory_budget_ratio",
+            b"xla_gpu_auto_spmd_partitioning_memory_budget_ratio",
+            "xla_gpu_autotune_gemm_rtol",
+            b"xla_gpu_autotune_gemm_rtol",
+            "xla_gpu_autotune_level",
+            b"xla_gpu_autotune_level",
+            "xla_gpu_autotune_max_solutions",
+            b"xla_gpu_autotune_max_solutions",
+            "xla_gpu_collect_cost_model_stats",
+            b"xla_gpu_collect_cost_model_stats",
+            "xla_gpu_collective_inflation_factor",
+            b"xla_gpu_collective_inflation_factor",
+            "xla_gpu_collective_permute_decomposer_threshold",
+            b"xla_gpu_collective_permute_decomposer_threshold",
+            "xla_gpu_copy_insertion_use_region_analysis",
+            b"xla_gpu_copy_insertion_use_region_analysis",
+            "xla_gpu_crash_on_verification_failures",
+            b"xla_gpu_crash_on_verification_failures",
+            "xla_gpu_cublas_fallback",
+            b"xla_gpu_cublas_fallback",
+            "xla_gpu_cuda_data_dir",
+            b"xla_gpu_cuda_data_dir",
+            "xla_gpu_cudnn_gemm_fusion_level",
+            b"xla_gpu_cudnn_gemm_fusion_level",
+            "xla_gpu_cudnn_gemm_max_plans",
+            b"xla_gpu_cudnn_gemm_max_plans",
+            "xla_gpu_deterministic_ops",
+            b"xla_gpu_deterministic_ops",
+            "xla_gpu_disable_async_collectives",
+            b"xla_gpu_disable_async_collectives",
+            "xla_gpu_disable_gpuasm_optimizations",
+            b"xla_gpu_disable_gpuasm_optimizations",
+            "xla_gpu_dump_autotune_logs_to",
+            b"xla_gpu_dump_autotune_logs_to",
+            "xla_gpu_dump_autotune_results_to",
+            b"xla_gpu_dump_autotune_results_to",
+            "xla_gpu_dump_autotuned_gemm_fusions",
+            b"xla_gpu_dump_autotuned_gemm_fusions",
+            "xla_gpu_dump_llvmir",
+            b"xla_gpu_dump_llvmir",
+            "xla_gpu_enable_all_gather_combine_by_dim",
+            b"xla_gpu_enable_all_gather_combine_by_dim",
+            "xla_gpu_enable_analytical_latency_estimator",
+            b"xla_gpu_enable_analytical_latency_estimator",
+            "xla_gpu_enable_approx_costly_collectives",
+            b"xla_gpu_enable_approx_costly_collectives",
+            "xla_gpu_enable_bf16_3way_gemm",
+            b"xla_gpu_enable_bf16_3way_gemm",
+            "xla_gpu_enable_bf16_6way_gemm",
+            b"xla_gpu_enable_bf16_6way_gemm",
+            "xla_gpu_enable_command_buffer",
+            b"xla_gpu_enable_command_buffer",
+            "xla_gpu_enable_cub_radix_sort",
+            b"xla_gpu_enable_cub_radix_sort",
+            "xla_gpu_enable_cublaslt",
+            b"xla_gpu_enable_cublaslt",
+            "xla_gpu_enable_cudnn_fmha",
+            b"xla_gpu_enable_cudnn_fmha",
+            "xla_gpu_enable_cudnn_frontend",
+            b"xla_gpu_enable_cudnn_frontend",
+            "xla_gpu_enable_cudnn_int8x32_convolution_reordering",
+            b"xla_gpu_enable_cudnn_int8x32_convolution_reordering",
+            "xla_gpu_enable_cudnn_layer_norm",
+            b"xla_gpu_enable_cudnn_layer_norm",
+            "xla_gpu_enable_custom_fusions",
+            b"xla_gpu_enable_custom_fusions",
+            "xla_gpu_enable_custom_fusions_re",
+            b"xla_gpu_enable_custom_fusions_re",
+            "xla_gpu_enable_dot_strength_reduction",
+            b"xla_gpu_enable_dot_strength_reduction",
+            "xla_gpu_enable_dynamic_slice_fusion",
+            b"xla_gpu_enable_dynamic_slice_fusion",
+            "xla_gpu_enable_fast_min_max",
+            b"xla_gpu_enable_fast_min_max",
+            "xla_gpu_enable_highest_priority_async_stream",
+            b"xla_gpu_enable_highest_priority_async_stream",
+            "xla_gpu_enable_host_memory_offloading",
+            b"xla_gpu_enable_host_memory_offloading",
+            "xla_gpu_enable_latency_hiding_scheduler",
+            b"xla_gpu_enable_latency_hiding_scheduler",
+            "xla_gpu_enable_libnvjitlink",
+            b"xla_gpu_enable_libnvjitlink",
+            "xla_gpu_enable_libnvptxcompiler",
+            b"xla_gpu_enable_libnvptxcompiler",
+            "xla_gpu_enable_llvm_module_compilation_parallelism",
+            b"xla_gpu_enable_llvm_module_compilation_parallelism",
+            "xla_gpu_enable_nccl_clique_optimization",
+            b"xla_gpu_enable_nccl_clique_optimization",
+            "xla_gpu_enable_nccl_comm_splitting",
+            b"xla_gpu_enable_nccl_comm_splitting",
+            "xla_gpu_enable_nccl_per_stream_comms",
+            b"xla_gpu_enable_nccl_per_stream_comms",
+            "xla_gpu_enable_nccl_user_buffers",
+            b"xla_gpu_enable_nccl_user_buffers",
+            "xla_gpu_enable_pgle_accuracy_checker",
+            b"xla_gpu_enable_pgle_accuracy_checker",
+            "xla_gpu_enable_pipelined_all_gather",
+            b"xla_gpu_enable_pipelined_all_gather",
+            "xla_gpu_enable_pipelined_all_reduce",
+            b"xla_gpu_enable_pipelined_all_reduce",
+            "xla_gpu_enable_pipelined_collectives",
+            b"xla_gpu_enable_pipelined_collectives",
+            "xla_gpu_enable_pipelined_p2p",
+            b"xla_gpu_enable_pipelined_p2p",
+            "xla_gpu_enable_pipelined_reduce_scatter",
+            b"xla_gpu_enable_pipelined_reduce_scatter",
+            "xla_gpu_enable_priority_fusion",
+            b"xla_gpu_enable_priority_fusion",
+            "xla_gpu_enable_reassociation_for_converted_ar",
+            b"xla_gpu_enable_reassociation_for_converted_ar",
+            "xla_gpu_enable_reduce_scatter_combine_by_dim",
+            b"xla_gpu_enable_reduce_scatter_combine_by_dim",
+            "xla_gpu_enable_reduction_epilogue_fusion",
+            b"xla_gpu_enable_reduction_epilogue_fusion",
+            "xla_gpu_enable_shared_constants",
+            b"xla_gpu_enable_shared_constants",
+            "xla_gpu_enable_split_k_autotuning",
+            b"xla_gpu_enable_split_k_autotuning",
+            "xla_gpu_enable_triton_gemm",
+            b"xla_gpu_enable_triton_gemm",
+            "xla_gpu_enable_triton_gemm_int4",
+            b"xla_gpu_enable_triton_gemm_int4",
+            "xla_gpu_enable_triton_hopper",
+            b"xla_gpu_enable_triton_hopper",
+            "xla_gpu_enable_while_loop_double_buffering",
+            b"xla_gpu_enable_while_loop_double_buffering",
+            "xla_gpu_enable_while_loop_reduce_scatter_code_motion",
+            b"xla_gpu_enable_while_loop_reduce_scatter_code_motion",
+            "xla_gpu_enable_while_loop_unrolling",
+            b"xla_gpu_enable_while_loop_unrolling",
+            "xla_gpu_ensure_minor_dot_contraction_dims",
+            b"xla_gpu_ensure_minor_dot_contraction_dims",
+            "xla_gpu_exclude_nondeterministic_ops",
+            b"xla_gpu_exclude_nondeterministic_ops",
+            "xla_gpu_executable_terminate_timeout_seconds",
+            b"xla_gpu_executable_terminate_timeout_seconds",
+            "xla_gpu_executable_warn_stuck_timeout_seconds",
+            b"xla_gpu_executable_warn_stuck_timeout_seconds",
+            "xla_gpu_exhaustive_tiling_search",
+            b"xla_gpu_exhaustive_tiling_search",
+            "xla_gpu_experimental_autotune_cache_mode",
+            b"xla_gpu_experimental_autotune_cache_mode",
+            "xla_gpu_experimental_disable_binary_libraries",
+            b"xla_gpu_experimental_disable_binary_libraries",
+            "xla_gpu_experimental_enable_triton_softmax_priority_fusion",
+            b"xla_gpu_experimental_enable_triton_softmax_priority_fusion",
+            "xla_gpu_filter_kernels_spilling_registers_on_autotuning",
+            b"xla_gpu_filter_kernels_spilling_registers_on_autotuning",
+            "xla_gpu_force_compilation_parallelism",
+            b"xla_gpu_force_compilation_parallelism",
+            "xla_gpu_force_conv_nchw",
+            b"xla_gpu_force_conv_nchw",
+            "xla_gpu_force_conv_nhwc",
+            b"xla_gpu_force_conv_nhwc",
+            "xla_gpu_ftz",
+            b"xla_gpu_ftz",
+            "xla_gpu_fused_attention_use_cudnn_rng",
+            b"xla_gpu_fused_attention_use_cudnn_rng",
+            "xla_gpu_gemm_rewrite_size_threshold",
+            b"xla_gpu_gemm_rewrite_size_threshold",
+            "xla_gpu_graph_enable_concurrent_region",
+            b"xla_gpu_graph_enable_concurrent_region",
+            "xla_gpu_graph_min_graph_size",
+            b"xla_gpu_graph_min_graph_size",
+            "xla_gpu_kernel_cache_file",
+            b"xla_gpu_kernel_cache_file",
+            "xla_gpu_lhs_enable_gpu_async_tracker",
+            b"xla_gpu_lhs_enable_gpu_async_tracker",
+            "xla_gpu_llvm_ir_file",
+            b"xla_gpu_llvm_ir_file",
+            "xla_gpu_llvm_verification_level",
+            b"xla_gpu_llvm_verification_level",
+            "xla_gpu_load_autotune_results_from",
+            b"xla_gpu_load_autotune_results_from",
+            "xla_gpu_memory_limit_slop_factor",
+            b"xla_gpu_memory_limit_slop_factor",
+            "xla_gpu_mlir_emitter_level",
+            b"xla_gpu_mlir_emitter_level",
+            "xla_gpu_mock_custom_calls",
+            b"xla_gpu_mock_custom_calls",
+            "xla_gpu_multi_streamed_windowed_einsum",
+            b"xla_gpu_multi_streamed_windowed_einsum",
+            "xla_gpu_nccl_collective_max_nchannels",
+            b"xla_gpu_nccl_collective_max_nchannels",
+            "xla_gpu_nccl_p2p_max_nchannels",
+            b"xla_gpu_nccl_p2p_max_nchannels",
+            "xla_gpu_nccl_terminate_on_error",
+            b"xla_gpu_nccl_terminate_on_error",
+            "xla_gpu_nccl_termination_timeout_seconds",
+            b"xla_gpu_nccl_termination_timeout_seconds",
+            "xla_gpu_override_gemm_autotuner",
+            b"xla_gpu_override_gemm_autotuner",
+            "xla_gpu_per_fusion_autotune_cache_dir",
+            b"xla_gpu_per_fusion_autotune_cache_dir",
+            "xla_gpu_pgle_profile_file_or_directory_path",
+            b"xla_gpu_pgle_profile_file_or_directory_path",
+            "xla_gpu_ptx_file",
+            b"xla_gpu_ptx_file",
+            "xla_gpu_reduce_scatter_combine_threshold_bytes",
+            b"xla_gpu_reduce_scatter_combine_threshold_bytes",
+            "xla_gpu_redzone_padding_bytes",
+            b"xla_gpu_redzone_padding_bytes",
+            "xla_gpu_redzone_scratch_max_megabytes",
+            b"xla_gpu_redzone_scratch_max_megabytes",
+            "xla_gpu_require_complete_aot_autotune_results",
+            b"xla_gpu_require_complete_aot_autotune_results",
+            "xla_gpu_run_post_layout_collective_pipeliner",
+            b"xla_gpu_run_post_layout_collective_pipeliner",
+            "xla_gpu_shape_checks",
+            b"xla_gpu_shape_checks",
+            "xla_gpu_shard_autotuning",
+            b"xla_gpu_shard_autotuning",
+            "xla_gpu_strict_conv_algorithm_picker",
+            b"xla_gpu_strict_conv_algorithm_picker",
+            "xla_gpu_target_config_filename",
+            b"xla_gpu_target_config_filename",
+            "xla_gpu_temp_buffer_use_separate_color",
+            b"xla_gpu_temp_buffer_use_separate_color",
+            "xla_gpu_threshold_for_windowed_einsum_mib",
+            b"xla_gpu_threshold_for_windowed_einsum_mib",
+            "xla_gpu_triton_fusion_level",
+            b"xla_gpu_triton_fusion_level",
+            "xla_gpu_triton_gemm_any",
+            b"xla_gpu_triton_gemm_any",
+            "xla_gpu_triton_gemm_disable_reduced_precision_reduction",
+            b"xla_gpu_triton_gemm_disable_reduced_precision_reduction",
+            "xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found",
+            b"xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found",
+            "xla_gpu_unsafe_pipelined_loop_annotator",
+            b"xla_gpu_unsafe_pipelined_loop_annotator",
+            "xla_gpu_unsupported_enable_triton_gemm",
+            b"xla_gpu_unsupported_enable_triton_gemm",
+            "xla_gpu_use_memcpy_local_p2p",
+            b"xla_gpu_use_memcpy_local_p2p",
+            "xla_gpu_use_runtime_fusion",
+            b"xla_gpu_use_runtime_fusion",
+            "xla_gpu_verify_triton_fusion_numerics",
+            b"xla_gpu_verify_triton_fusion_numerics",
+            "xla_hlo_evaluator_use_fast_path",
+            b"xla_hlo_evaluator_use_fast_path",
+            "xla_hlo_graph_addresses",
+            b"xla_hlo_graph_addresses",
+            "xla_hlo_graph_sharding_color",
+            b"xla_hlo_graph_sharding_color",
+            "xla_hlo_profile",
+            b"xla_hlo_profile",
+            "xla_llvm_disable_expensive_passes",
+            b"xla_llvm_disable_expensive_passes",
+            "xla_llvm_enable_alias_scope_metadata",
+            b"xla_llvm_enable_alias_scope_metadata",
+            "xla_llvm_enable_invariant_load_metadata",
+            b"xla_llvm_enable_invariant_load_metadata",
+            "xla_llvm_enable_noalias_metadata",
+            b"xla_llvm_enable_noalias_metadata",
+            "xla_llvm_force_inline_before_split",
+            b"xla_llvm_force_inline_before_split",
+            "xla_multiheap_size_constraint_per_heap",
+            b"xla_multiheap_size_constraint_per_heap",
+            "xla_partitioning_algorithm",
+            b"xla_partitioning_algorithm",
+            "xla_reduce_window_rewrite_base_length",
+            b"xla_reduce_window_rewrite_base_length",
+            "xla_step_marker_location",
+            b"xla_step_marker_location",
+            "xla_syntax_sugar_async_ops",
+            b"xla_syntax_sugar_async_ops",
+            "xla_test_all_input_layouts",
+            b"xla_test_all_input_layouts",
+            "xla_test_all_output_layouts",
+            b"xla_test_all_output_layouts",
+            "xla_tpu_detect_inf",
+            b"xla_tpu_detect_inf",
+            "xla_tpu_detect_nan",
+            b"xla_tpu_detect_nan",
+        ],
+    ) -> None: ...
+
+global___DebugOptions = DebugOptions
+
+@typing.final
+class GpuCompilationEnvironment(google.protobuf.message.Message):
+    """Contains flags which affects the GPU compilation result.
+    These flags are part of Debug Options as of now, and will be migrated to
+    this proto.
+    """
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    DUMMY_FLAG_FIELD_NUMBER: builtins.int
+    dummy_flag: builtins.int
+    """Temporary dummy flag is added to test the flow.
+    To be removed when we add flags here.
+    """
+    def __init__(self, *, dummy_flag: builtins.int | None = ...) -> None: ...
+    def ClearField(self, field_name: typing.Literal["dummy_flag", b"dummy_flag"]) -> None: ...
+
+global___GpuCompilationEnvironment = GpuCompilationEnvironment
+
+@typing.final
+class ShardableValueUpdatePairProto(google.protobuf.message.Message):
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    INPUT_PARAMETER_NUMBER_FIELD_NUMBER: builtins.int
+    PARAMETER_SHAPE_INDEX_FIELD_NUMBER: builtins.int
+    OUTPUT_SHAPE_INDEX_FIELD_NUMBER: builtins.int
+    input_parameter_number: builtins.int
+    @property
+    def parameter_shape_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+    @property
+    def output_shape_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+    def __init__(
+        self,
+        *,
+        input_parameter_number: builtins.int | None = ...,
+        parameter_shape_index: collections.abc.Iterable[builtins.int] | None = ...,
+        output_shape_index: collections.abc.Iterable[builtins.int] | None = ...,
+    ) -> None: ...
+    def ClearField(
+        self,
+        field_name: typing.Literal[
+            "input_parameter_number",
+            b"input_parameter_number",
+            "output_shape_index",
+            b"output_shape_index",
+            "parameter_shape_index",
+            b"parameter_shape_index",
+        ],
+    ) -> None: ...
+
+global___ShardableValueUpdatePairProto = ShardableValueUpdatePairProto
+
+@typing.final
+class ExecutionOptions(google.protobuf.message.Message):
+    """These settings control how XLA compiles and/or runs code.  Not all settings
+    will have an effect on every platform.
+
+    When adding new fields, keep in mind that boolean fields default to false.
+    Next id: 25.
+    """
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    SHAPE_WITH_OUTPUT_LAYOUT_FIELD_NUMBER: builtins.int
+    SEED_FIELD_NUMBER: builtins.int
+    DEBUG_OPTIONS_FIELD_NUMBER: builtins.int
+    DEVICE_HANDLES_FIELD_NUMBER: builtins.int
+    NUM_REPLICAS_FIELD_NUMBER: builtins.int
+    DEVICE_ASSIGNMENT_FIELD_NUMBER: builtins.int
+    ALIAS_PASSTHROUGH_PARAMS_FIELD_NUMBER: builtins.int
+    NUM_PARTITIONS_FIELD_NUMBER: builtins.int
+    LAUNCH_ID_FIELD_NUMBER: builtins.int
+    USE_SPMD_PARTITIONING_FIELD_NUMBER: builtins.int
+    USE_AUTO_SPMD_PARTITIONING_FIELD_NUMBER: builtins.int
+    AUTO_SPMD_PARTITIONING_MESH_SHAPE_FIELD_NUMBER: builtins.int
+    AUTO_SPMD_PARTITIONING_MESH_IDS_FIELD_NUMBER: builtins.int
+    DEDUPLICATE_HLO_FIELD_NUMBER: builtins.int
+    ALLOW_SPMD_SHARDING_PROPAGATION_TO_PARAMETERS_FIELD_NUMBER: builtins.int
+    ALLOW_SPMD_SHARDING_PROPAGATION_TO_OUTPUT_FIELD_NUMBER: builtins.int
+    PARAM_REQUIRES_BROADCAST_VIA_COLLECTIVES_FIELD_NUMBER: builtins.int
+    ALLOW_SEPARATE_SHARDING_PROGRAMS_FIELD_NUMBER: builtins.int
+    SHARDABLE_VALUE_UPDATE_PAIRS_FIELD_NUMBER: builtins.int
+    FDO_PROFILE_FIELD_NUMBER: builtins.int
+    DEVICE_MEMORY_SIZE_FIELD_NUMBER: builtins.int
+    USE_SHARDY_PARTITIONER_FIELD_NUMBER: builtins.int
+    seed: builtins.int
+    """Used to seed random-number generators used in this computation.  If this is
+    0, we generate a seed ourselves.
+
+    TODO(b/32083678): Changing the seed unnecessarily forces a recompilation.
+    """
+    num_replicas: builtins.int
+    """Number of replicas of the computation to run. If zero, uses the default
+    number of replicas for the XLA service.
+    """
+    alias_passthrough_params: builtins.bool
+    """Alias input and output buffers for parameters that are passed-through XLA
+    modules without being changed.
+    """
+    num_partitions: builtins.int
+    """Number of partitions of the computation to run (model parallelism).
+    If zero, uses the default number of partitions for the XLA service.
+    """
+    launch_id: builtins.int
+    """Used to identify a set of programs that should be launch together."""
+    use_spmd_partitioning: builtins.bool
+    """Indicates whether to use SPMD (true) or MPMD (false) partitioning when
+    num_partitions > 1 and XLA is requested to partition the input program.
+    """
+    use_auto_spmd_partitioning: builtins.bool
+    """Whether to automatically generate XLA shardings for SPMD partitioner."""
+    deduplicate_hlo: builtins.bool
+    """If set, deduplicate hlo into function calls to reduce binary size. Only
+    works on TPU.
+    """
+    allow_separate_sharding_programs: builtins.bool
+    """If enabled, the compiler may generate sharding and unsharding programs as
+    separate HLO modules, and modify the main program's input and output to
+    be sharded.
+    """
+    fdo_profile: builtins.bytes
+    """Profiling data for feedback directed optimizations. Note that this is not
+    the only way to feed FDO data into the compiler and individual backends
+    may choose to get FDO data by other means.
+    """
+    device_memory_size: builtins.int
+    """Amount of device memory available for the executable to use."""
+    use_shardy_partitioner: builtins.bool
+    """Use Shardy, a new partitioner, to replace the existing
+    ShardingPropagation and SpmdPartitioner. See go/xla-sdy-pipeline for
+    details.
+    """
+    @property
+    def shape_with_output_layout(self) -> tensorflow.compiler.xla.xla_data_pb2.ShapeProto:
+        """This optional field's layout is used as a hint when storing the output of
+        this computation.  Subsequent transfers of this output array to the client
+        may be faster when using this layout.
+
+        We use a Shape here to accommodate computations that return a tuple.
+        """
+
+    @property
+    def debug_options(self) -> global___DebugOptions: ...
+    @property
+    def device_handles(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[tensorflow.compiler.xla.xla_data_pb2.DeviceHandle]:
+        """This optional field specifies a particular set of devices to run the
+        computation on. The computation will be partitioned across these devices.
+        If not provided, the default device will be chosen.
+        """
+
+    @property
+    def device_assignment(self) -> tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto:
+        """This optional field specifies the device assignment if known at compile
+        time.
+        """
+
+    @property
+    def auto_spmd_partitioning_mesh_shape(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+        """Device mesh shape used to create the sharding search space when
+        use_auto_spmd_partitioning=true.
+        """
+
+    @property
+    def auto_spmd_partitioning_mesh_ids(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+        """Device mesh ids compatible with the above mesh_shape used when
+        use_auto_spmd_partitioning=true.
+        """
+
+    @property
+    def allow_spmd_sharding_propagation_to_parameters(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]:
+        """Allows sharding propagation to propagate to the parameters. This changes
+        the input shape of the computation (which is undesirable), but it can be
+        used to allow to run partial compilation to determine what would be the
+        input sharding of a computation if XLA would be allowed to propagate the
+        sharding which can be used by higher level framework as a way to query
+        intermediate sharding of operations when multiple computation would be
+        chained and merged together.
+        This is a vector of bool, because the user can control which parameters can
+        have the sharding substituted. If only one boolean value is passed in the
+        vector that is interpreted as the value to be applied for every parameter.
+        """
+
+    @property
+    def allow_spmd_sharding_propagation_to_output(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]:
+        """Allows sharding propagation to propagate to the outputs. This changes the
+        output shape of the computation (which is undesirable), but it can be used
+        to allow to run partial compilation to determine what would be the output
+        sharding of a computation if XLA would be allowed to propagate the sharding
+        which can be used by higher level framework as a way to query intermediate
+        sharding of operations when multiple computation would be chained and
+        merged together.
+        This is a vector of bool, because the user can control (if the output of
+        the computation is a tuple) which elements of the tuple can have the
+        sharding substituted and which don't. If only one boolean value is passed
+        in the vector that's interpreted as the value to be applied for every
+        single element of the output tuple. One value per element of the tuple
+        means that each value is attached to one of the output elements.
+        """
+
+    @property
+    def param_requires_broadcast_via_collectives(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]:
+        """Whether to broadcast args across all replicas. One entry per arg."""
+
+    @property
+    def shardable_value_update_pairs(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___ShardableValueUpdatePairProto]:
+        """The list of input/output pairs in the main program that could be sharded."""
+
+    def __init__(
+        self,
+        *,
+        shape_with_output_layout: tensorflow.compiler.xla.xla_data_pb2.ShapeProto | None = ...,
+        seed: builtins.int | None = ...,
+        debug_options: global___DebugOptions | None = ...,
+        device_handles: collections.abc.Iterable[tensorflow.compiler.xla.xla_data_pb2.DeviceHandle] | None = ...,
+        num_replicas: builtins.int | None = ...,
+        device_assignment: tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto | None = ...,
+        alias_passthrough_params: builtins.bool | None = ...,
+        num_partitions: builtins.int | None = ...,
+        launch_id: builtins.int | None = ...,
+        use_spmd_partitioning: builtins.bool | None = ...,
+        use_auto_spmd_partitioning: builtins.bool | None = ...,
+        auto_spmd_partitioning_mesh_shape: collections.abc.Iterable[builtins.int] | None = ...,
+        auto_spmd_partitioning_mesh_ids: collections.abc.Iterable[builtins.int] | None = ...,
+        deduplicate_hlo: builtins.bool | None = ...,
+        allow_spmd_sharding_propagation_to_parameters: collections.abc.Iterable[builtins.bool] | None = ...,
+        allow_spmd_sharding_propagation_to_output: collections.abc.Iterable[builtins.bool] | None = ...,
+        param_requires_broadcast_via_collectives: collections.abc.Iterable[builtins.bool] | None = ...,
+        allow_separate_sharding_programs: builtins.bool | None = ...,
+        shardable_value_update_pairs: collections.abc.Iterable[global___ShardableValueUpdatePairProto] | None = ...,
+        fdo_profile: builtins.bytes | None = ...,
+        device_memory_size: builtins.int | None = ...,
+        use_shardy_partitioner: builtins.bool | None = ...,
+    ) -> None: ...
+    def HasField(
+        self,
+        field_name: typing.Literal[
+            "debug_options",
+            b"debug_options",
+            "device_assignment",
+            b"device_assignment",
+            "shape_with_output_layout",
+            b"shape_with_output_layout",
+        ],
+    ) -> builtins.bool: ...
+    def ClearField(
+        self,
+        field_name: typing.Literal[
+            "alias_passthrough_params",
+            b"alias_passthrough_params",
+            "allow_separate_sharding_programs",
+            b"allow_separate_sharding_programs",
+            "allow_spmd_sharding_propagation_to_output",
+            b"allow_spmd_sharding_propagation_to_output",
+            "allow_spmd_sharding_propagation_to_parameters",
+            b"allow_spmd_sharding_propagation_to_parameters",
+            "auto_spmd_partitioning_mesh_ids",
+            b"auto_spmd_partitioning_mesh_ids",
+            "auto_spmd_partitioning_mesh_shape",
+            b"auto_spmd_partitioning_mesh_shape",
+            "debug_options",
+            b"debug_options",
+            "deduplicate_hlo",
+            b"deduplicate_hlo",
+            "device_assignment",
+            b"device_assignment",
+            "device_handles",
+            b"device_handles",
+            "device_memory_size",
+            b"device_memory_size",
+            "fdo_profile",
+            b"fdo_profile",
+            "launch_id",
+            b"launch_id",
+            "num_partitions",
+            b"num_partitions",
+            "num_replicas",
+            b"num_replicas",
+            "param_requires_broadcast_via_collectives",
+            b"param_requires_broadcast_via_collectives",
+            "seed",
+            b"seed",
+            "shape_with_output_layout",
+            b"shape_with_output_layout",
+            "shardable_value_update_pairs",
+            b"shardable_value_update_pairs",
+            "use_auto_spmd_partitioning",
+            b"use_auto_spmd_partitioning",
+            "use_shardy_partitioner",
+            b"use_shardy_partitioner",
+            "use_spmd_partitioning",
+            b"use_spmd_partitioning",
+        ],
+    ) -> None: ...
+
+global___ExecutionOptions = ExecutionOptions
+
+@typing.final
+class HloModuleConfigProto(google.protobuf.message.Message):
+    """Serialization of HloModuleConfig. See the C++ class definition for
+    descriptions of each field.
+    There are no guarantees of backwards or forwards compatibility.
+    Next id: 36.
+    """
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    class _FusionConfigCollection:
+        ValueType = typing.NewType("ValueType", builtins.int)
+        V: typing_extensions.TypeAlias = ValueType
+
+    class _FusionConfigCollectionEnumTypeWrapper(
+        google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[HloModuleConfigProto._FusionConfigCollection.ValueType],
+        builtins.type,
+    ):
+        DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+        OFF: HloModuleConfigProto._FusionConfigCollection.ValueType  # 0
+        """Do not collect configuration."""
+        PER_EDGE: HloModuleConfigProto._FusionConfigCollection.ValueType  # 1
+        """Collect per-edge configuration."""
+        PER_NODE: HloModuleConfigProto._FusionConfigCollection.ValueType  # 2
+        """Collect per-node configuration."""
+
+    class FusionConfigCollection(_FusionConfigCollection, metaclass=_FusionConfigCollectionEnumTypeWrapper): ...
+    OFF: HloModuleConfigProto.FusionConfigCollection.ValueType  # 0
+    """Do not collect configuration."""
+    PER_EDGE: HloModuleConfigProto.FusionConfigCollection.ValueType  # 1
+    """Collect per-edge configuration."""
+    PER_NODE: HloModuleConfigProto.FusionConfigCollection.ValueType  # 2
+    """Collect per-node configuration."""
+
+    @typing.final
+    class BoolList(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        VALS_FIELD_NUMBER: builtins.int
+        @property
+        def vals(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]: ...
+        def __init__(self, *, vals: collections.abc.Iterable[builtins.bool] | None = ...) -> None: ...
+        def ClearField(self, field_name: typing.Literal["vals", b"vals"]) -> None: ...
+
+    @typing.final
+    class Int64List(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        VALS_FIELD_NUMBER: builtins.int
+        @property
+        def vals(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+        def __init__(self, *, vals: collections.abc.Iterable[builtins.int] | None = ...) -> None: ...
+        def ClearField(self, field_name: typing.Literal["vals", b"vals"]) -> None: ...
+
+    @typing.final
+    class Int64ListList(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        LISTS_FIELD_NUMBER: builtins.int
+        @property
+        def lists(
+            self,
+        ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___HloModuleConfigProto.Int64List]: ...
+        def __init__(self, *, lists: collections.abc.Iterable[global___HloModuleConfigProto.Int64List] | None = ...) -> None: ...
+        def ClearField(self, field_name: typing.Literal["lists", b"lists"]) -> None: ...
+
+    @typing.final
+    class DotConfigEntry(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        KEY_FIELD_NUMBER: builtins.int
+        VALUE_FIELD_NUMBER: builtins.int
+        key: builtins.str
+        @property
+        def value(self) -> global___HloModuleConfigProto.Int64List: ...
+        def __init__(
+            self, *, key: builtins.str | None = ..., value: global___HloModuleConfigProto.Int64List | None = ...
+        ) -> None: ...
+        def HasField(self, field_name: typing.Literal["value", b"value"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing.Literal["key", b"key", "value", b"value"]) -> None: ...
+
+    @typing.final
+    class AnalysisAllowanceMapEntry(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        KEY_FIELD_NUMBER: builtins.int
+        VALUE_FIELD_NUMBER: builtins.int
+        key: builtins.str
+        value: builtins.int
+        def __init__(self, *, key: builtins.str | None = ..., value: builtins.int | None = ...) -> None: ...
+        def ClearField(self, field_name: typing.Literal["key", b"key", "value", b"value"]) -> None: ...
+
+    ENTRY_COMPUTATION_LAYOUT_FIELD_NUMBER: builtins.int
+    SEED_FIELD_NUMBER: builtins.int
+    LAUNCH_ID_FIELD_NUMBER: builtins.int
+    REPLICA_COUNT_FIELD_NUMBER: builtins.int
+    NUM_PARTITIONS_FIELD_NUMBER: builtins.int
+    PARAM_REQUIRES_BROADCAST_VIA_COLLECTIVES_FIELD_NUMBER: builtins.int
+    USE_SPMD_PARTITIONING_FIELD_NUMBER: builtins.int
+    USE_AUTO_SPMD_PARTITIONING_FIELD_NUMBER: builtins.int
+    AUTO_SPMD_PARTITIONING_MESH_SHAPE_FIELD_NUMBER: builtins.int
+    AUTO_SPMD_PARTITIONING_MESH_IDS_FIELD_NUMBER: builtins.int
+    DEDUPLICATE_HLO_FIELD_NUMBER: builtins.int
+    INTRA_OP_PARALLELISM_THREADS_FIELD_NUMBER: builtins.int
+    DEVICE_TYPE_FIELD_NUMBER: builtins.int
+    DEBUG_OPTIONS_FIELD_NUMBER: builtins.int
+    STATIC_DEVICE_ASSIGNMENT_FIELD_NUMBER: builtins.int
+    PRE_SIMULATION_DEVICE_ASSIGNMENT_FIELD_NUMBER: builtins.int
+    ALLOW_SEPARATE_SHARDING_PROGRAMS_FIELD_NUMBER: builtins.int
+    SHARDABLE_VALUE_UPDATE_PAIRS_FIELD_NUMBER: builtins.int
+    ALIAS_PASSTHROUGH_PARAMS_FIELD_NUMBER: builtins.int
+    CONTENT_AWARE_COMPUTATION_SORTING_FIELD_NUMBER: builtins.int
+    FUSION_CONFIG_COLLECTION_FIELD_NUMBER: builtins.int
+    FUSION_CONFIG_FIELD_NUMBER: builtins.int
+    DOT_CONFIG_FIELD_NUMBER: builtins.int
+    LAYOUT_CONFIG_FIELD_NUMBER: builtins.int
+    MEMORY_SPACE_ASSIGNMENT_CONFIG_FIELD_NUMBER: builtins.int
+    PHASE_ORDERING_CONFIG_FIELD_NUMBER: builtins.int
+    PHASE_INDEX_FIELD_NUMBER: builtins.int
+    ALLOW_SPMD_SHARDING_PROPAGATION_TO_PARAMETERS_FIELD_NUMBER: builtins.int
+    ALLOW_SPMD_SHARDING_PROPAGATION_TO_OUTPUT_FIELD_NUMBER: builtins.int
+    ANALYSIS_ALLOWANCE_MAP_FIELD_NUMBER: builtins.int
+    MATRIX_UNIT_OPERAND_PRECISION_FIELD_NUMBER: builtins.int
+    FDO_PROFILE_FIELD_NUMBER: builtins.int
+    DEVICE_MEMORY_SIZE_FIELD_NUMBER: builtins.int
+    USE_SHARDY_PARTITIONER_FIELD_NUMBER: builtins.int
+    seed: builtins.int
+    launch_id: builtins.int
+    replica_count: builtins.int
+    num_partitions: builtins.int
+    use_spmd_partitioning: builtins.bool
+    use_auto_spmd_partitioning: builtins.bool
+    deduplicate_hlo: builtins.bool
+    intra_op_parallelism_threads: builtins.int
+    device_type: builtins.str
+    allow_separate_sharding_programs: builtins.bool
+    alias_passthrough_params: builtins.bool
+    content_aware_computation_sorting: builtins.bool
+    fusion_config_collection: global___HloModuleConfigProto.FusionConfigCollection.ValueType
+    phase_index: builtins.int
+    matrix_unit_operand_precision: tensorflow.compiler.xla.xla_data_pb2.PrecisionConfig.Precision.ValueType
+    fdo_profile: builtins.bytes
+    device_memory_size: builtins.int
+    use_shardy_partitioner: builtins.bool
+    @property
+    def entry_computation_layout(self) -> tensorflow.compiler.xla.xla_data_pb2.ProgramShapeProto: ...
+    @property
+    def param_requires_broadcast_via_collectives(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]: ...
+    @property
+    def auto_spmd_partitioning_mesh_shape(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+    @property
+    def auto_spmd_partitioning_mesh_ids(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+    @property
+    def debug_options(self) -> global___DebugOptions: ...
+    @property
+    def static_device_assignment(self) -> tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto: ...
+    @property
+    def pre_simulation_device_assignment(self) -> tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto:
+        """The original device assignment before being changed by a simulator.
+        Simulators, like HybridSim, may change the device assignment to a smaller
+        topology, to make simulation easier.
+        """
+
+    @property
+    def shardable_value_update_pairs(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___ShardableValueUpdatePairProto]: ...
+    @property
+    def fusion_config(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___HloModuleConfigProto.BoolList]: ...
+    @property
+    def dot_config(
+        self,
+    ) -> google.protobuf.internal.containers.MessageMap[builtins.str, global___HloModuleConfigProto.Int64List]: ...
+    @property
+    def layout_config(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___HloModuleConfigProto.Int64ListList]: ...
+    @property
+    def memory_space_assignment_config(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]: ...
+    @property
+    def phase_ordering_config(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___HloModuleConfigProto.BoolList]: ...
+    @property
+    def allow_spmd_sharding_propagation_to_parameters(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]: ...
+    @property
+    def allow_spmd_sharding_propagation_to_output(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.bool]: ...
+    @property
+    def analysis_allowance_map(self) -> google.protobuf.internal.containers.ScalarMap[builtins.str, builtins.int]: ...
+    def __init__(
+        self,
+        *,
+        entry_computation_layout: tensorflow.compiler.xla.xla_data_pb2.ProgramShapeProto | None = ...,
+        seed: builtins.int | None = ...,
+        launch_id: builtins.int | None = ...,
+        replica_count: builtins.int | None = ...,
+        num_partitions: builtins.int | None = ...,
+        param_requires_broadcast_via_collectives: collections.abc.Iterable[builtins.bool] | None = ...,
+        use_spmd_partitioning: builtins.bool | None = ...,
+        use_auto_spmd_partitioning: builtins.bool | None = ...,
+        auto_spmd_partitioning_mesh_shape: collections.abc.Iterable[builtins.int] | None = ...,
+        auto_spmd_partitioning_mesh_ids: collections.abc.Iterable[builtins.int] | None = ...,
+        deduplicate_hlo: builtins.bool | None = ...,
+        intra_op_parallelism_threads: builtins.int | None = ...,
+        device_type: builtins.str | None = ...,
+        debug_options: global___DebugOptions | None = ...,
+        static_device_assignment: tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto | None = ...,
+        pre_simulation_device_assignment: tensorflow.compiler.xla.xla_data_pb2.DeviceAssignmentProto | None = ...,
+        allow_separate_sharding_programs: builtins.bool | None = ...,
+        shardable_value_update_pairs: collections.abc.Iterable[global___ShardableValueUpdatePairProto] | None = ...,
+        alias_passthrough_params: builtins.bool | None = ...,
+        content_aware_computation_sorting: builtins.bool | None = ...,
+        fusion_config_collection: global___HloModuleConfigProto.FusionConfigCollection.ValueType | None = ...,
+        fusion_config: collections.abc.Iterable[global___HloModuleConfigProto.BoolList] | None = ...,
+        dot_config: collections.abc.Mapping[builtins.str, global___HloModuleConfigProto.Int64List] | None = ...,
+        layout_config: collections.abc.Iterable[global___HloModuleConfigProto.Int64ListList] | None = ...,
+        memory_space_assignment_config: collections.abc.Iterable[builtins.int] | None = ...,
+        phase_ordering_config: collections.abc.Iterable[global___HloModuleConfigProto.BoolList] | None = ...,
+        phase_index: builtins.int | None = ...,
+        allow_spmd_sharding_propagation_to_parameters: collections.abc.Iterable[builtins.bool] | None = ...,
+        allow_spmd_sharding_propagation_to_output: collections.abc.Iterable[builtins.bool] | None = ...,
+        analysis_allowance_map: collections.abc.Mapping[builtins.str, builtins.int] | None = ...,
+        matrix_unit_operand_precision: tensorflow.compiler.xla.xla_data_pb2.PrecisionConfig.Precision.ValueType | None = ...,
+        fdo_profile: builtins.bytes | None = ...,
+        device_memory_size: builtins.int | None = ...,
+        use_shardy_partitioner: builtins.bool | None = ...,
+    ) -> None: ...
+    def HasField(
+        self,
+        field_name: typing.Literal[
+            "debug_options",
+            b"debug_options",
+            "entry_computation_layout",
+            b"entry_computation_layout",
+            "pre_simulation_device_assignment",
+            b"pre_simulation_device_assignment",
+            "static_device_assignment",
+            b"static_device_assignment",
+        ],
+    ) -> builtins.bool: ...
+    def ClearField(
+        self,
+        field_name: typing.Literal[
+            "alias_passthrough_params",
+            b"alias_passthrough_params",
+            "allow_separate_sharding_programs",
+            b"allow_separate_sharding_programs",
+            "allow_spmd_sharding_propagation_to_output",
+            b"allow_spmd_sharding_propagation_to_output",
+            "allow_spmd_sharding_propagation_to_parameters",
+            b"allow_spmd_sharding_propagation_to_parameters",
+            "analysis_allowance_map",
+            b"analysis_allowance_map",
+            "auto_spmd_partitioning_mesh_ids",
+            b"auto_spmd_partitioning_mesh_ids",
+            "auto_spmd_partitioning_mesh_shape",
+            b"auto_spmd_partitioning_mesh_shape",
+            "content_aware_computation_sorting",
+            b"content_aware_computation_sorting",
+            "debug_options",
+            b"debug_options",
+            "deduplicate_hlo",
+            b"deduplicate_hlo",
+            "device_memory_size",
+            b"device_memory_size",
+            "device_type",
+            b"device_type",
+            "dot_config",
+            b"dot_config",
+            "entry_computation_layout",
+            b"entry_computation_layout",
+            "fdo_profile",
+            b"fdo_profile",
+            "fusion_config",
+            b"fusion_config",
+            "fusion_config_collection",
+            b"fusion_config_collection",
+            "intra_op_parallelism_threads",
+            b"intra_op_parallelism_threads",
+            "launch_id",
+            b"launch_id",
+            "layout_config",
+            b"layout_config",
+            "matrix_unit_operand_precision",
+            b"matrix_unit_operand_precision",
+            "memory_space_assignment_config",
+            b"memory_space_assignment_config",
+            "num_partitions",
+            b"num_partitions",
+            "param_requires_broadcast_via_collectives",
+            b"param_requires_broadcast_via_collectives",
+            "phase_index",
+            b"phase_index",
+            "phase_ordering_config",
+            b"phase_ordering_config",
+            "pre_simulation_device_assignment",
+            b"pre_simulation_device_assignment",
+            "replica_count",
+            b"replica_count",
+            "seed",
+            b"seed",
+            "shardable_value_update_pairs",
+            b"shardable_value_update_pairs",
+            "static_device_assignment",
+            b"static_device_assignment",
+            "use_auto_spmd_partitioning",
+            b"use_auto_spmd_partitioning",
+            "use_shardy_partitioner",
+            b"use_shardy_partitioner",
+            "use_spmd_partitioning",
+            b"use_spmd_partitioning",
+        ],
+    ) -> None: ...
+
+global___HloModuleConfigProto = HloModuleConfigProto
+
+@typing.final
+class HloModuleProtoWithConfig(google.protobuf.message.Message):
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    HLO_MODULE_FIELD_NUMBER: builtins.int
+    CONFIG_FIELD_NUMBER: builtins.int
+    @property
+    def hlo_module(self) -> tensorflow.compiler.xla.service.hlo_pb2.HloModuleProto: ...
+    @property
+    def config(self) -> global___HloModuleConfigProto: ...
+    def __init__(
+        self,
+        *,
+        hlo_module: tensorflow.compiler.xla.service.hlo_pb2.HloModuleProto | None = ...,
+        config: global___HloModuleConfigProto | None = ...,
+    ) -> None: ...
+    def HasField(self, field_name: typing.Literal["config", b"config", "hlo_module", b"hlo_module"]) -> builtins.bool: ...
+    def ClearField(self, field_name: typing.Literal["config", b"config", "hlo_module", b"hlo_module"]) -> None: ...
+
+global___HloModuleProtoWithConfig = HloModuleProtoWithConfig
+
+@typing.final
+class ScheduleProto(google.protobuf.message.Message):
+    """A trace estimated by the Latency Hiding Scheduler."""
+
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    @typing.final
+    class Instruction(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        ID_FIELD_NUMBER: builtins.int
+        START_TIMESTAMP_CYCLES_FIELD_NUMBER: builtins.int
+        END_TIMESTAMP_CYCLES_FIELD_NUMBER: builtins.int
+        id: builtins.int
+        """Instruction id (matches the id in HloInstructionProto)."""
+        start_timestamp_cycles: builtins.float
+        """Start and end timestamps in cycles."""
+        end_timestamp_cycles: builtins.float
+        def __init__(
+            self,
+            *,
+            id: builtins.int | None = ...,
+            start_timestamp_cycles: builtins.float | None = ...,
+            end_timestamp_cycles: builtins.float | None = ...,
+        ) -> None: ...
+        def ClearField(
+            self,
+            field_name: typing.Literal[
+                "end_timestamp_cycles", b"end_timestamp_cycles", "id", b"id", "start_timestamp_cycles", b"start_timestamp_cycles"
+            ],
+        ) -> None: ...
+
+    INSTRUCTIONS_FIELD_NUMBER: builtins.int
+    COMPUTATION_ID_FIELD_NUMBER: builtins.int
+    HLO_MODULE_FIELD_NUMBER: builtins.int
+    CYCLES_PER_MICROSECOND_FIELD_NUMBER: builtins.int
+    computation_id: builtins.int
+    """Computation id (matches the id in HloComputationProto)."""
+    cycles_per_microsecond: builtins.int
+    @property
+    def instructions(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___ScheduleProto.Instruction]: ...
+    @property
+    def hlo_module(self) -> tensorflow.compiler.xla.service.hlo_pb2.HloModuleProto: ...
+    def __init__(
+        self,
+        *,
+        instructions: collections.abc.Iterable[global___ScheduleProto.Instruction] | None = ...,
+        computation_id: builtins.int | None = ...,
+        hlo_module: tensorflow.compiler.xla.service.hlo_pb2.HloModuleProto | None = ...,
+        cycles_per_microsecond: builtins.int | None = ...,
+    ) -> None: ...
+    def HasField(self, field_name: typing.Literal["hlo_module", b"hlo_module"]) -> builtins.bool: ...
+    def ClearField(
+        self,
+        field_name: typing.Literal[
+            "computation_id",
+            b"computation_id",
+            "cycles_per_microsecond",
+            b"cycles_per_microsecond",
+            "hlo_module",
+            b"hlo_module",
+            "instructions",
+            b"instructions",
+        ],
+    ) -> None: ...
+
+global___ScheduleProto = ScheduleProto