package pblczero
Get desktop application:
View/edit binary Protocol Buffers messages
message
onnx.proto:75Used in:
optional string name = 1
optional string doc_string = 13
optional float f = 2
Exactly ONE of the following fields must be present
float
optional int64 i = 3
int
optional bytes s = 4
UTF-8 string
optional t = 5
tensor value
repeated float floats = 7
list of floats
repeated int64 ints = 8
list of ints
repeated bytes strings = 9
list of UTF-8 strings
repeated tensors = 10
list of tensors
optional AttributeProto.AttributeType type = 20
Used in:
UNDEFINED = 0
FLOAT = 1
INT = 2
STRING = 3
TENSOR = 4
GRAPH = 5;
FLOATS = 6
INTS = 7
STRINGS = 8
TENSORS = 9
GRAPHS = 10;
message
hlo.proto:352Used in:
required string key = 1
required value = 2
message
hlo.proto:442repeated argument_layouts = 1
optional bool parameter_is_tupled_arguments = 2
optional executable_build_options = 3
optional bool compile_portable_executable = 4
optional int64 profile_version = 5
optional bytes serialized_multi_slice_config = 6
repeated env_options = 7
message
net.proto:31Used in:
optional uint32 major = 1
optional uint32 minor = 2
optional uint32 patch = 3
message
hlo.proto:366Used in:
optional int64 device_ordinal = 1
If set, this is the device to build the computation for. Valid device_ordinal values are: 0 to # of devices - 1. These values are identical to the device ordinal values used by StreamExecutor. The built executable will be executable on any device equivalent to the specified device as determined by Backend::devices_equivalent(). A value of -1 indicates this option has not been set.
optional result_layout = 2
If set, this specifies the layout of the result of the computation. If not set, the service will chose the layout of the result. A Shape is used to store the layout to accommodate tuple result shapes. A value of nullptr indicates the option has not been set.
optional int64 num_replicas = 4
The number of replicas of this computation that are to be executed. Defaults to 1.
optional int64 num_partitions = 5
The number of partitions in this computation. Defaults to 1.
optional bool use_spmd_partitioning = 6
Indicates whether to use SPMD (true) or MPMD (false) partitioning when num_partitions > 1 and XLA is requested to partition the input program.
optional bool use_auto_spmd_partitioning = 7
Whether to automatically generate XLA shardings for SPMD partitioner.
optional bool deduplicate_hlo = 8
Whether HLOs should be deduplicated.
optional device_assignment = 9
If set, this specifies a static device assignment for the computation. Otherwise, the computation will be compiled generically and can be run with any device assignment compatible with the computation's replica and partition counts.
optional bool alias_passthrough_params = 10
Whether input and output buffers are aliased if the associated parameter is passed-through XLA modules without being changed.
optional bool run_backend_only = 11
By default, XLA builds an executable by invoking standard compilation, i.e. running Compiler::Compile, or both Compiler::RunHloPasses and Compiler::RunBackend. When run_backend_only is set to true, XLA builds an executable by invoking only RunBackend and skip invoking RunHloPasses, which can be used to compile post-optimizations HLO modules.
repeated bool allow_spmd_sharding_propagation_to_output = 12
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.
optional bytes fdo_profile = 14
Opaque profile data for any feedback directed optimizations.
optional int64 device_memory_size = 15
repeated int64 auto_spmd_partitioning_mesh_shape = 16
Mesh shape in auto sharding options.
repeated int64 auto_spmd_partitioning_mesh_ids = 17
Mesh ids in auto sharding options.
message
net.proto:368Used in:
optional Format.Encoding weights_encoding = 1
Any encoding specified in a Layer overides this.
optional network_format = 2
If network_format is missing, it's assumed to have INPUT_CLASSICAL_112_PLANE / OUTPUT_CLASSICAL / NETWORK_CLASSICAL format.
Used in:
UNKNOWN = 0
LINEAR16 = 1
message
onnx.proto:141Used in:
repeated node = 1
optional string name = 2
repeated initializer = 5
optional string doc_string = 10
repeated input = 11
repeated output = 12
message
hlo.proto:314Used in:
required string name = 1
repeated instructions = 2
The array of instructions is always in a valid dependency order, where operands appear before their users.
required program_shape = 4
required int64 id = 5
The id of this computation.
required int64 root_id = 6
The id of the root of the computation.
message
hlo.proto:263Used in:
required string name = 1
required string opcode = 2
required shape = 3
optional metadata = 7
optional literal = 8
Literal, only present for kConstant.
optional int64 parameter_number = 9
Parameter number is only present for kParameter.
optional int64 tuple_index = 13
Index for kGetTupleElement.
optional window = 15
Describes the window in a windowed operation such as convolution.
optional convolution_dimension_numbers = 16
Describes the dimension numbers used for a convolution.
repeated slice_dimensions = 17
optional dot_dimension_numbers = 30
repeated int64 dimensions = 14
Dimensions present for some operations that require reshaping or broadcasting, including Reshape, Reduce, ReduceWindow, and Reverse.
optional gather_dimension_numbers = 33
Gather dimension numbers.
repeated int64 gather_slice_sizes = 34
optional bool indices_are_sorted = 67
optional bool unique_indices = 69
required int64 id = 35
The id of this instruction.
repeated int64 operand_ids = 36
repeated int64 called_computation_ids = 38
optional string comparison_direction = 63
message
hlo.proto:286Describes the [begin, end) index range and stride for slices.
Used in:
optional int64 start = 1
optional int64 limit = 2
optional int64 stride = 3
message
hlo.proto:329required string name = 1
required string entry_computation_name = 2
required int64 entry_computation_id = 6
repeated computations = 3
The array of computations is always in a valid dependency order, where callees appear before their callers.
required host_program_shape = 4
The host program shape (with layout) of the entry computation.
required int64 id = 5
The id of this module.
message
onnx.proto:156The subset of Onnx protos that are used in lczero.
optional int64 ir_version = 1
optional string producer_name = 2
optional string producer_version = 3
optional string domain = 4
'org.lczero.models.*'
optional int64 model_version = 5
optional string doc_string = 6
optional graph = 7
repeated opset_import = 8
message
net.proto:401optional fixed32 magic = 1
optional string license = 2
optional min_version = 3
optional format = 4
optional training_params = 5
optional weights = 10
Either weights or onnx_model is set, but not both.
optional onnx_model = 11
message
net.proto:274Used in:
optional NetworkFormat.InputFormat input = 1
optional NetworkFormat.OutputFormat output = 2
optional NetworkFormat.NetworkStructure network = 3
optional NetworkFormat.PolicyFormat policy = 4
optional NetworkFormat.ValueFormat value = 5
optional NetworkFormat.MovesLeftFormat moves_left = 6
optional NetworkFormat.DefaultActivation default_activation = 7
optional NetworkFormat.ActivationFunction smolgen_activation = 8
optional NetworkFormat.ActivationFunction ffn_activation = 9
optional NetworkFormat.InputEmbeddingFormat input_embedding = 10
Used in:
ACTIVATION_DEFAULT = 0
ACTIVATION_MISH = 1
ACTIVATION_RELU = 2
ACTIVATION_NONE = 3
ACTIVATION_TANH = 4
ACTIVATION_SIGMOID = 5
ACTIVATION_SELU = 6
ACTIVATION_SWISH = 7
ACTIVATION_RELU_2 = 8
ACTIVATION_SOFTMAX = 9
Activation used everywhere except head outputs or otherwise specified.
Used in:
DEFAULT_ACTIVATION_RELU = 0
DEFAULT_ACTIVATION_MISH = 1
Used in:
INPUT_EMBEDDING_NONE = 0
INPUT_EMBEDDING_PE_MAP = 1
INPUT_EMBEDDING_PE_DENSE = 2
Format to encode the input planes with. Used by position encoder.
Used in:
INPUT_UNKNOWN = 0
INPUT_CLASSICAL_112_PLANE = 1
INPUT_112_WITH_CASTLING_PLANE = 2
INPUT_112_WITH_CANONICALIZATION = 3
INPUT_112_WITH_CANONICALIZATION_HECTOPLIES = 4
INPUT_112_WITH_CANONICALIZATION_HECTOPLIES_ARMAGEDDON = 132
INPUT_112_WITH_CANONICALIZATION_V2 = 5
INPUT_112_WITH_CANONICALIZATION_V2_ARMAGEDDON = 133
Moves left head architecture
Used in:
MOVES_LEFT_NONE = 0
MOVES_LEFT_V1 = 1
Network architecture. Used by backends to build the network.
Used in:
NETWORK_UNKNOWN = 0
Networks without PolicyFormat or ValueFormat specified
NETWORK_CLASSICAL = 1
NETWORK_SE = 2
NETWORK_CLASSICAL_WITH_HEADFORMAT = 3
Networks with PolicyFormat and ValueFormat specified
NETWORK_SE_WITH_HEADFORMAT = 4
NETWORK_ONNX = 5
NETWORK_ATTENTIONBODY_WITH_HEADFORMAT = 6
NETWORK_ATTENTIONBODY_WITH_MULTIHEADFORMAT = 7
NETWORK_AB_LEGACY_WITH_MULTIHEADFORMAT = 134
Output format of the NN. Used by search code to interpret results.
Used in:
OUTPUT_UNKNOWN = 0
OUTPUT_CLASSICAL = 1
OUTPUT_WDL = 2
Policy head architecture
Used in:
POLICY_UNKNOWN = 0
POLICY_CLASSICAL = 1
POLICY_CONVOLUTION = 2
POLICY_ATTENTION = 3
Value head architecture
Used in:
VALUE_UNKNOWN = 0
VALUE_CLASSICAL = 1
VALUE_WDL = 2
VALUE_PARAM = 3
message
onnx.proto:109Used in:
repeated string input = 1
repeated string output = 2
optional string name = 3
optional string op_type = 4
repeated attribute = 5
optional string doc_string = 6
optional string domain = 7
message
net.proto:380Used in:
optional bytes model = 1
Serialized OnnxProto model.
optional OnnxModel.DataType data_type = 2
optional string input_planes = 3
Name of the input tensor to populate.
optional string output_value = 4
Names of the output tensors to get results from. If some feature is not present, corresponding values are not set.
optional string output_wdl = 5
optional string output_policy = 6
optional string output_mlh = 7
Used in:
UNKNOWN_DATATYPE = 0
FLOAT = 1
FLOAT16 = 10
BFLOAT16 = 16
message
onnx.proto:150Used in:
optional string domain = 1
optional int64 version = 2
message
hlo.proto:345Used in:
optional string string_field = 1
optional bool bool_field = 2
optional int64 int_field = 3
optional double double_field = 4
message
onnx.proto:70Used in:
optional string key = 1
optional string value = 2
message
onnx.proto:32Used in: ,
repeated int64 dims = 1
optional TensorProto.DataType data_type = 2
optional string name = 8
optional bytes raw_data = 9
optional string doc_string = 12
repeated external_data = 13
optional TensorProto.DataLocation data_location = 14
Used in:
DEFAULT = 0
EXTERNAL = 1
Used in: ,
UNDEFINED = 0
FLOAT = 1
Basic types.
float
UINT8 = 2
uint8_t
INT8 = 3
int8_t
UINT16 = 4
uint16_t
INT16 = 5
int16_t
INT32 = 6
int32_t
INT64 = 7
int64_t
STRING = 8
string
BOOL = 9
bool
FLOAT16 = 10
DOUBLE = 11
UINT32 = 12
UINT64 = 13
COMPLEX64 = 14
COMPLEX128 = 15
BFLOAT16 = 16
FLOAT8E4M3FN = 17
FLOAT8E4M3FNUZ = 18
FLOAT8E5M2 = 19
FLOAT8E5M2FNUZ = 20
message
onnx.proto:119Used in:
repeated dim = 1
message
onnx.proto:120Used in:
optional int64 dim_value = 1
optional string dim_param = 2
message
net.proto:265Used in:
optional uint32 training_steps = 1
optional float learning_rate = 2
optional float mse_loss = 3
optional float policy_loss = 4
optional float accuracy = 5
optional string lc0_params = 6
message
onnx.proto:127Used in:
optional tensor_type = 1
message
onnx.proto:128Used in:
optional TensorProto.DataType elem_type = 1
optional shape = 2
message
onnx.proto:135Used in:
optional string name = 1
optional type = 2
optional string doc_string = 3
message
net.proto:37Used in:
optional input = 1
Input convnet.
repeated residual = 2
Residual tower.
optional ip_emb_preproc_w = 37
optional ip_emb_preproc_b = 38
optional ip_emb_w = 25
optional ip_emb_b = 26
optional ip_emb_ln_gammas = 39
optional ip_emb_ln_betas = 40
optional ip_mult_gate = 33
Input gating (NETWORK_ATTENTIONBODY_WITH_HEADFORMAT).
optional ip_add_gate = 34
optional ip_emb_ffn = 41
optional ip_emb_ffn_ln_gammas = 42
optional ip_emb_ffn_ln_betas = 43
repeated encoder = 27
Encoder stack (NETWORK_ATTENTIONBODY_WITH_HEADFORMAT).
optional uint32 headcount = 28
repeated pol_encoder = 21
Policy encoder stack The ffn activation up to and including NETWORK_SE_WITH_HEADFORMAT is SELU, otherwise it follows the ffn activation setting.
optional uint32 pol_headcount = 24
optional policy1 = 11
Policy head Extra convolution for AZ-style policy head
optional policy = 3
optional ip_pol_w = 4
"embedding" in policy attention
optional ip_pol_b = 5
optional ip2_pol_w = 17
For policy attention, up to and including NETWORK_SE_WITH_HEADFORMAT the "embedding" activation is SELU, otherwise it is the default activation.
"wq" in policy attention
optional ip2_pol_b = 18
optional ip3_pol_w = 19
"wk" in policy attention
optional ip3_pol_b = 20
optional ip4_pol_w = 22
"ppo" in policy attention
optional value = 6
Value head
optional ip_val_w = 29
"embedding" for attention body value
optional ip_val_b = 30
optional ip1_val_w = 7
optional ip1_val_b = 8
optional ip2_val_w = 9
optional ip2_val_b = 10
optional value_heads = 44
optional policy_heads = 45
optional moves_left = 12
Moves left head
optional ip_mov_w = 31
"embedding" for attention body moves left
optional ip_mov_b = 32
optional ip1_mov_w = 13
optional ip1_mov_b = 14
optional ip2_mov_w = 15
optional ip2_mov_b = 16
optional smolgen_w = 35
Global smolgen weights (NETWORK_ATTENTIONBODY_WITH_HEADFORMAT).
optional smolgen_b = 36
message
net.proto:52Used in: , , ,
optional weights = 1
optional biases = 2
optional bn_means = 3
optional bn_stddivs = 4
optional bn_gammas = 5
optional bn_betas = 6
message
net.proto:115Used in: ,
optional mha = 1
optional ln1_gammas = 2
optional ln1_betas = 3
optional ffn = 4
optional ln2_gammas = 5
optional ln2_betas = 6
message
net.proto:107Used in: ,
optional dense1_w = 1
optional dense1_b = 2
optional dense2_w = 3
optional dense2_b = 4
reserved 5 - 10 for int8 quantization
message
net.proto:38Used in: , , , , , , , , ,
optional float min_val = 1
optional float max_val = 2
optional bytes params = 3
optional Layer.Encoding encoding = 4
repeated uint32 dims = 5
Used in:
UNKNOWN_ENCODING = 0
LINEAR16 = 1
FLOAT16 = 2
BFLOAT16 = 3
message
net.proto:89Used in:
optional q_w = 1
optional q_b = 2
optional k_w = 3
optional k_b = 4
optional v_w = 5
optional v_b = 6
optional dense_w = 7
optional dense_b = 8
optional smolgen = 9
optional rpe_q = 10
optional rpe_k = 11
optional rpe_v = 12
message
net.proto:124Used in: ,
optional ip_pol_w = 1
optional ip_pol_b = 2
optional ip2_pol_w = 3
"wq" in policy attention
optional ip2_pol_b = 4
optional ip3_pol_w = 5
"wk" in policy attention
optional ip3_pol_b = 6
optional ip4_pol_w = 7
"ppo" in policy attention
repeated pol_encoder = 8
Optional policy encoders for policy head.
optional uint32 pol_headcount = 9
optional policy1 = 10
Convolutions for legacy policy head.
optional policy = 11
message
net.proto:158Used in:
required string key = 1
name of the policy head
required value = 2
message
net.proto:163Used in:
optional ip_pol_w = 1
"embedding" in policy attention
optional ip_pol_b = 2
optional vanilla = 3
optional optimistic_st = 4
optional soft = 5
optional opponent = 6
repeated policy_head_map = 7
map<string, PolicyHead> policy_head_map = 7;
message
net.proto:70Used in:
optional conv1 = 1
optional conv2 = 2
optional se = 3
message
net.proto:61Used in:
optional w1 = 1
Squeeze-excitation unit (https://arxiv.org/abs/1709.01507) weights and biases of the two fully connected layers.
optional b1 = 2
optional w2 = 3
optional b2 = 4
message
net.proto:76Used in:
optional compress = 1
For NETWORK_ATTENTIONBODY_WITH_HEADFORMAT.
optional dense1_w = 2
optional dense1_b = 3
optional ln1_gammas = 4
optional ln1_betas = 5
optional dense2_w = 6
optional dense2_b = 7
optional ln2_gammas = 8
optional ln2_betas = 9
message
net.proto:142Used in: ,
optional ip_val_w = 1
"embedding" for attention body value
optional ip_val_b = 2
optional ip1_val_w = 3
optional ip1_val_b = 4
optional ip2_val_w = 5
optional ip2_val_b = 6
optional ip_val_err_w = 7
optional ip_val_err_b = 8
optional ip_val_cat_w = 9
optional ip_val_cat_b = 10
optional value = 11
Legacy value head support.
message
net.proto:174Used in:
required string key = 1
name of the value head
required value = 2
message
net.proto:179Used in:
optional winner = 1
optional q = 2
optional st = 3
repeated value_head_map = 4
map<string, ValueHead> value_head_map = 4;
message
hlo.proto:237Used in:
optional int64 input_batch_dimension = 7
optional int64 input_feature_dimension = 8
repeated int64 input_spatial_dimensions = 11
optional int64 kernel_input_feature_dimension = 3
optional int64 kernel_output_feature_dimension = 4
repeated int64 kernel_spatial_dimensions = 6
optional int64 output_batch_dimension = 9
optional int64 output_feature_dimension = 10
repeated int64 output_spatial_dimensions = 12
message
hlo.proto:357Used in:
optional int32 replica_count = 1
optional int32 computation_count = 2
repeated computation_devices = 3
message
hlo.proto:360Used in:
repeated int64 replica_device_ids = 1
message
hlo.proto:249Used in:
repeated int64 lhs_contracting_dimensions = 1
repeated int64 rhs_contracting_dimensions = 2
repeated int64 lhs_batch_dimensions = 3
repeated int64 rhs_batch_dimensions = 4
message
hlo.proto:256Used in:
repeated int64 offset_dims = 1
repeated int64 collapsed_slice_dims = 2
repeated int64 start_index_map = 3
optional int64 index_vector_dim = 4
message
hlo.proto:32Used in:
repeated int64 minor_to_major = 1
Sequence of dimension numbers, from minor (fastest varying index) to major (slowest varying index). This field is required.
message
hlo.proto:195Used in:
required shape = 1
repeated bool preds = 2
optional bytes s4s = 21
optional bytes u4s = 22
optional bytes s8s = 15
optional bytes u8s = 3
repeated int32 s32s = 4
repeated int64 s64s = 5
repeated uint32 u32s = 6
repeated uint64 u64s = 7
repeated float f32s = 8
repeated double f64s = 9
repeated float c64s = 12
Stored as interleaved real, imag floats.
repeated double c128s = 18
Stored as interleaved real, imag doubles.
repeated tuple_literals = 10
optional bytes f16s = 11
The F16s, BF16s, U16s and S16s are encoded in little endian byte order
optional bytes bf16s = 13
optional bytes u16s = 16
optional bytes s16s = 17
optional bytes f8e5m2s = 19
optional bytes f8e4m3fns = 20
optional bytes f8e4m3b11fnuzs = 23
optional bytes f8e5m2fnuzs = 24
optional bytes f8e4m3fnuzs = 25
repeated int64 sparse_indices = 14
Next = 26
message
hlo.proto:174Symbolization metadata for HLO Instructions. This metadata is used for debugging XLA code generation, as well as performance profiling of XLA-generated executables.
Used in:
optional string op_type = 1
The framework op name that generated this XLA op. Frameworks that build on top of XLA should mirror the names of their ops back to users by specifying the op_type. In this way, even if the framework's "ops" are implemented as multiple XLA HLO Ops, they can be grouped appropriately. (e.g. if a SoftMax layer is emitted into XLA as multiple ops, then each op should have the op_type be "SoftMax".)
optional string op_name = 2
The user-specified name of the op. This name is often unique within a computation. Note: some frameworks add auto-generated names if the user does not provide one.
optional string source_file = 3
Indicate a file and line that this op is associated to in a user's program. e.g. it could be the file and line of user code that generated the op.
optional int32 source_line = 4
message
hlo.proto:164Shape of the parameters and output of a computation (like a traditional function signature).
Used in: ,
repeated parameters = 1
required result = 2
repeated string parameter_names = 3
message
hlo.proto:38Used in: , , , ,
required XlaShapeProto.Type element_type = 2
The element type for this shape.
repeated int64 dimensions = 3
The size (number of elements) for each dimension, or an upper bound on the size if the dimension is dynamic. In XLA, dimensions are numbered from 0 to N-1 for an N-dimensional array. The first element of 'dimensions' is the size of dimension 0, the second element is the size of dimension 1, and so forth. Empty list indicates a scalar. If the respective element in 'is_dimension_dynamic' is true then the value in this field represents an upper bound on the size of the dimension.
repeated tuple_shapes = 4
For tuples only, the shapes of constituent shapes in the tuple sequence.
required layout = 5
The layout used to back this shape.
repeated bool is_dynamic_dimension = 6
For arrays, this indicates whether or not each dimension is dynamically-sized. The number of elements in this repeated field should be zero (indicating that no dimensions are dynamic) or equal to the number of elements in the 'dimensions' field.
Used in:
PRIMITIVE_TYPE_INVALID = 0
PRED = 1
Predicates are two-state booleans.
S4 = 21
Signed integral values of fixed width.
S8 = 2
S16 = 3
S32 = 4
S64 = 5
U4 = 22
Unsigned integral values of fixed width.
U8 = 6
U16 = 7
U32 = 8
U64 = 9
F16 = 10
Floating-point values of fixed width. Note: if f16s are not natively supported on the device, they will be converted to f16 from f32 at arbirary points in the computation.
F32 = 11
BF16 = 16
Truncated 16 bit floating-point format. This is similar to IEEE's 16 bit floating-point format, but uses 1 bit for the sign, 8 bits for the exponent and 7 bits for the mantissa.
F64 = 12
F8E5M2 = 19
F8E4M3FN = 20
F8E4M3B11FNUZ = 23
F8E5M2FNUZ = 24
FP8 dtypes, as described in this paper: https://arxiv.org/abs/2206.02915 F8E5M2FNUZ has 5 exponent bits and 2 mantissa bits. F8E4M3FNUZ has 4 exponent bits and 3 mantissa bits. The "FNUZ" means only Finite and NaN values are supported; zero is unsigned. Unlike IEEE types, infinities are not supported. NaN is represented when the exponent and mantissa bits are all 0s with a sign bit of 1. All other values are finite. These differences mean there's an additional exponent value available. To keep the same dynamic range as an IEEE-like FP8 type, the exponent is biased one more than would be expected given the number of exponent bits (8 for Float8E4M3FNUZ and 16 for Float8E5M2FNUZ).
F8E4M3FNUZ = 25
C64 = 15
Complex values of fixed width.
Paired F32 (real, imag), as in std::complex<float>.
C128 = 18
Paired F64 (real, imag), as in std::complex<double>.
TUPLE = 13
A tuple is a polymorphic sequence; e.g. a shape that holds different sub-shapes. They are used for things like returning multiple values from a computation; e.g. a computation that returns weights and biases may have a signature that results in a tuple like (f32[784x2000], f32[2000]) If a shape proto has the tuple element type, it may not have any entries in the dimensions field.
OPAQUE_TYPE = 14
An opaque type used for passing context-specific data to a custom operation. Shapes of this primitive type will have empty dimensions and tuple_shapes fields. (OPAQUE would be a better name for this identifier, but that conflicts with a macro defined in windows.h.)
TOKEN = 17
A token type threaded between side-effecting operations. Shapes of this primitive type will have empty dimensions and tuple_shapes fields.
message
hlo.proto:235Used in:
repeated dimensions = 1
message
hlo.proto:225Used in:
optional int64 size = 1
optional int64 stride = 2
optional int64 padding_low = 3
optional int64 padding_high = 4
optional int64 window_dilation = 5
optional int64 base_dilation = 6
optional bool window_reversal = 7