package pblczero

Get desktop application:
View/edit binary Protocol Buffers messages

• optional fixed32 magic = 1

• optional string license = 2

• optional EngineVersion version = 3

• repeated Game game = 4

Used in: Chunk, Net

• optional uint32 major = 1

• optional uint32 minor = 2

• optional uint32 patch = 3

Used in: Net

• optional Format.Encoding weights_encoding = 1

• optional NetworkFormat network_format = 2

  If network_format is missing, it's assumed to have INPUT_CLASSICAL_112_PLANE / OUTPUT_CLASSICAL / NETWORK_CLASSICAL format.

Used in: Format

• UNKNOWN = 0

• LINEAR16 = 1

Used in: Chunk

• repeated State state = 1

• repeated Policy policy = 2

• repeated float value = 3

• repeated uint32 move = 4

• optional Game.Result result = 5

Used in: Game

• WHITE = 0

• BLACK = 1

• DRAW = 2

• optional fixed32 magic = 1

• optional string license = 2

• optional EngineVersion min_version = 3

• optional Format format = 4

• optional TrainingParams training_params = 5

• optional Weights weights = 10

Used in: Format

• 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

Format to encode the input planes with. Used by position encoder.

Used in: NetworkFormat

• INPUT_UNKNOWN = 0

• INPUT_CLASSICAL_112_PLANE = 1

  INPUT_WITH_COORDINATE_PLANES = 2; // Example. Uncomment/rename.

Network architecture. Used by backends to build the network.

Used in: NetworkFormat

• 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

Output format of the NN. Used by search code to interpret results.

Used in: NetworkFormat

• OUTPUT_UNKNOWN = 0

• OUTPUT_CLASSICAL = 1

• OUTPUT_WDL = 2

Policy head architecture

Used in: NetworkFormat

• POLICY_UNKNOWN = 0

• POLICY_CLASSICAL = 1

• POLICY_CONVOLUTION = 2

Value head architecture

Used in: NetworkFormat

• VALUE_UNKNOWN = 0

• VALUE_CLASSICAL = 1

• VALUE_WDL = 2

Used in: Game

• repeated uint32 index = 1

• repeated float prior = 2

Used in: Game

• repeated fixed64 plane = 1

• optional uint32 us_ooo = 2

• optional uint32 us_oo = 3

• optional uint32 them_ooo = 4

• optional uint32 them_oo = 5

• optional uint32 side_to_move = 6

• optional uint32 rule_50 = 7

Used in: Net

• 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

Used in: Net

• optional Weights.ConvBlock input = 1

  Input convnet.

• repeated Weights.Residual residual = 2

  Residual tower.

• optional Weights.ConvBlock policy1 = 11

  Policy head Extra convolution for AZ-style policy head

• optional Weights.ConvBlock policy = 3

• optional Weights.Layer ip_pol_w = 4

• optional Weights.Layer ip_pol_b = 5

• optional Weights.ConvBlock value = 6

  Value head

• optional Weights.Layer ip1_val_w = 7

• optional Weights.Layer ip1_val_b = 8

• optional Weights.Layer ip2_val_w = 9

• optional Weights.Layer ip2_val_b = 10

Used in: Weights, Residual

• optional Layer weights = 1

• optional Layer biases = 2

• optional Layer bn_means = 3

• optional Layer bn_stddivs = 4

• optional Layer bn_gammas = 5

• optional Layer bn_betas = 6

Used in: Weights, ConvBlock, SEunit

• optional float min_val = 1

• optional float max_val = 2

• optional bytes params = 3

Used in: Weights

• optional ConvBlock conv1 = 1

• optional ConvBlock conv2 = 2

• optional SEunit se = 3

Used in: Residual

• optional Layer w1 = 1

  Squeeze-excitation unit (https://arxiv.org/abs/1709.01507) weights and biases of the two fully connected layers.

• optional Layer b1 = 2

• optional Layer w2 = 3

• optional Layer b2 = 4