Connectome routing

Module contract

The routing modules construct procedural contact identities and delayed event delivery. ProceduralMorphologyConnectomeRouter retains morphology-conditioned branch targeting. Population IDs, local/long-range source identities, delay slots, and route topology must agree with the declared workload; procedural routing does not imply an empirical connectome.

Source revision: 306a51ed950b. Public export index.

DelayBucket

Source

DelayBucket(delay_steps: int, first_fanout_offset: int, event_count_per_source: int, local_targets: bool) -> None

Fields

Parameter	Type	Default
delay_steps	int	required
first_fanout_offset	int	required
event_count_per_source	int	required
local_targets	bool	required

DelayBucket.fanout_offsets

Source

DelayBucket.fanout_offsets: tuple[int, ...]

Read-only property. Access as instance.fanout_offsets; do not call it as a function.

Returns tuple[int, ...].

procedural_delay_buckets

Source

procedural_delay_buckets(*, fanout: int, local_fanout: int, minimum_delay_steps: int=1) -> tuple[DelayBucket, ...]

Partition fanout offsets into local/long-range delay classes.

Parameter	Type	Default
fanout	int	required
local_fanout	int	required
minimum_delay_steps	int	1

Returns tuple[DelayBucket, ...].

compact_active_sources

Source

compact_active_sources(outputs: torch.Tensor, *, threshold: float=0.0) -> torch.Tensor

Return source-sorted int32 indices whose scalar output fires.

Parameter	Type	Default
outputs	torch.Tensor	required
threshold	float	0.0

Returns torch.Tensor.

build_typed_tile_pools

Source

build_typed_tile_pools(source_is_inhibitory: torch.Tensor, physical_to_logical: torch.Tensor, *, spatial_tile_neurons: int) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]

Build invertible logical E/I ranks within every spatial tile.

Parameter	Type	Default
source_is_inhibitory	torch.Tensor	required
physical_to_logical	torch.Tensor	required
spatial_tile_neurons	int	required

Returns tuple[torch.Tensor, torch.Tensor, torch.Tensor].

build_external_count_branch_bank

Source

build_external_count_branch_bank(*, slot_map: torch.Tensor, morphology_gain: torch.Tensor, feature_gain: torch.Tensor, synapses_per_branch: int, max_event_count: int=32, channels_per_type: int | None=None) -> torch.Tensor

Precompute exact branch currents for a consecutive event pattern.

Parameter	Type	Default
slot_map	torch.Tensor	required
morphology_gain	torch.Tensor	required
feature_gain	torch.Tensor	required
synapses_per_branch	int	required
max_event_count	int	32
channels_per_type	int | None	None

Returns torch.Tensor.

TrajectoryExternalBranchDrive

Source

Exact native-time trajectory drive at the learned branch boundary.

TrajectoryExternalBranchDrive.init

Source

__init__(self, *, trajectories: torch.Tensor, branch_bank: torch.Tensor, logical_ids: torch.Tensor, morphology: torch.Tensor, seed: int, excitatory_gain: float, inhibitory_gain: float, slot_map: torch.Tensor | None=None, morphology_gain: torch.Tensor | None=None, feature_gain: torch.Tensor | None=None, synapses_per_branch: int | None=None, block_rows: int=8, block_branches: int=128) -> None

Parameter	Type	Default
trajectories	torch.Tensor	required
branch_bank	torch.Tensor	required
logical_ids	torch.Tensor	required
morphology	torch.Tensor	required
seed	int	required
excitatory_gain	float	required
inhibitory_gain	float	required
slot_map	torch.Tensor | None	None
morphology_gain	torch.Tensor | None	None
feature_gain	torch.Tensor | None	None
synapses_per_branch	int | None	None
block_rows	int	8
block_branches	int	128

seed: Random seed for the declared operation.

Returns None.

TrajectoryExternalBranchDrive.materialize

Source

materialize(self, step: int, *, output: torch.Tensor | None=None, add_existing: bool=False, existing: torch.Tensor | None=None, recurrent_bits: torch.Tensor | None=None) -> torch.Tensor

Materialize exact branch currents for one native timestep.

Parameter	Type	Default
step	int	required
output	torch.Tensor | None	None
add_existing	bool	False
existing	torch.Tensor | None	None
recurrent_bits	torch.Tensor | None	None

Returns torch.Tensor.

ProceduralBinaryChannelConnectomeRouter

Source

Exact delayed routing in AxoBench’s signed binary-channel space.

ProceduralBinaryChannelConnectomeRouter.init

Source

__init__(self, contract: LargePopulationSimulationContract, *, slot_map: torch.Tensor, morphology: torch.Tensor, source_is_inhibitory: torch.Tensor, morphology_gain: torch.Tensor, feature_gain: torch.Tensor, synapses_per_branch: int, physical_to_logical: torch.Tensor | None=None, logical_to_physical: torch.Tensor | None=None, route_block_size: int=512, route_num_warps: int=4, branch_block_rows: int=8, branch_block_size: int=128) -> None

Parameter	Type	Default
contract	LargePopulationSimulationContract	required
slot_map	torch.Tensor	required
morphology	torch.Tensor	required
source_is_inhibitory	torch.Tensor	required
morphology_gain	torch.Tensor	required
feature_gain	torch.Tensor	required
synapses_per_branch	int	required
physical_to_logical	torch.Tensor | None	None
logical_to_physical	torch.Tensor | None	None
route_block_size	int	512
route_num_warps	int	4
branch_block_rows	int	8
branch_block_size	int	128

Returns None.

ProceduralBinaryChannelConnectomeRouter.queue_bytes

Source

ProceduralBinaryChannelConnectomeRouter.queue_bytes: int

Read-only property. Access as instance.queue_bytes; do not call it as a function.

Returns int.

ProceduralBinaryChannelConnectomeRouter.current_channel_bits

Source

current_channel_bits(self, current_slot: int) -> torch.Tensor

Parameter	Type	Default
current_slot	int	required

Returns torch.Tensor.

ProceduralBinaryChannelConnectomeRouter.current_inputs

Source

current_inputs(self, current_slot: int, *, output: torch.Tensor | None=None) -> torch.Tensor

Convert one exact binary channel slot to learned branch currents.

Parameter	Type	Default
current_slot	int	required
output	torch.Tensor | None	None

Returns torch.Tensor.

ProceduralBinaryChannelConnectomeRouter.clear_and_route

Source

clear_and_route(self, active_sources: torch.Tensor, *, current_slot: int, clear_consumed: bool=True) -> torch.Tensor

Clear a consumed slot and OR new delayed recurrent channels.

Parameter	Type	Default
active_sources	torch.Tensor	required
current_slot	int	required
clear_consumed	bool	True

Returns torch.Tensor.

ProceduralExactBranchConnectomeRouter

Source

Sparse exact-OR routing with ready-to-consume branch currents.

Bases: ProceduralBinaryChannelConnectomeRouter.

ProceduralExactBranchConnectomeRouter.init

Source

__init__(self, *args, external_trajectories: torch.Tensor | None=None, external_seed: int=0, external_excitatory_gain: float=1.0, external_inhibitory_gain: float=1.0, max_external_event_count: int=32, **kwargs) -> None

Parameter	Type	Default
args	unspecified	variadic
external_trajectories	torch.Tensor | None	None
external_seed	int	0
external_excitatory_gain	float	1.0
external_inhibitory_gain	float	1.0
max_external_event_count	int	32
kwargs	unspecified	variadic

Returns None.

ProceduralExactBranchConnectomeRouter.queue_bytes

Source

ProceduralExactBranchConnectomeRouter.queue_bytes: int

Read-only property. Access as instance.queue_bytes; do not call it as a function.

Returns int.

ProceduralExactBranchConnectomeRouter.current_inputs

Source

current_inputs(self, current_slot: int, *, output: torch.Tensor | None=None) -> torch.Tensor

Parameter	Type	Default
current_slot	int	required
output	torch.Tensor | None	None

Returns torch.Tensor.

ProceduralExactBranchConnectomeRouter.clear_and_route

Source

clear_and_route(self, active_sources: torch.Tensor, *, current_slot: int, current_step: int | None=None, clear_consumed: bool=True) -> torch.Tensor

Clear consumed state and route each binary channel at most once.

Parameter	Type	Default
active_sources	torch.Tensor	required
current_slot	int	required
current_step	int | None	None
clear_consumed	bool	True

Returns torch.Tensor.

ProceduralMorphologyConnectomeRouter

Source

Persistent delay-queue router for the scalable connectome control.

ProceduralMorphologyConnectomeRouter.init

Source

__init__(self, contract: LargePopulationSimulationContract, *, slot_map: torch.Tensor, morphology: torch.Tensor, source_is_inhibitory: torch.Tensor, morphology_gain: torch.Tensor, feature_gain: torch.Tensor, synapses_per_branch: int, physical_to_logical: torch.Tensor | None=None, logical_to_physical: torch.Tensor | None=None, route_block_size: int=512, route_num_warps: int=4) -> None

Parameter	Type	Default
contract	LargePopulationSimulationContract	required
slot_map	torch.Tensor	required
morphology	torch.Tensor	required
source_is_inhibitory	torch.Tensor	required
morphology_gain	torch.Tensor	required
feature_gain	torch.Tensor	required
synapses_per_branch	int	required
physical_to_logical	torch.Tensor | None	None
logical_to_physical	torch.Tensor | None	None
route_block_size	int	512
route_num_warps	int	4

Returns None.

ProceduralMorphologyConnectomeRouter.current_inputs

Source

current_inputs(self, current_slot: int) -> torch.Tensor

Parameter	Type	Default
current_slot	int	required

Returns torch.Tensor.

ProceduralMorphologyConnectomeRouter.clear_and_route

Source

clear_and_route(self, active_sources: torch.Tensor, *, current_slot: int, clear_consumed: bool=True, touched_offsets: torch.Tensor | None=None) -> torch.Tensor

Clear a consumed queue slot and schedule new delayed events.

Parameter	Type	Default
active_sources	torch.Tensor	required
current_slot	int	required
clear_consumed	bool	True
touched_offsets	torch.Tensor | None	None

Returns torch.Tensor.

ProceduralMorphologyConnectomeRouter.clear_recorded_branches

Source

clear_recorded_branches(self, touched_offsets: torch.Tensor) -> None

Clear branch queue destinations recorded by event delivery.

Parameter	Type	Default
touched_offsets	torch.Tensor	required

Returns None.

ProceduralUniqueChannelConnectomeRouter

Source

Collision-free typed fan-in with invertible event-wise routing.

Bases: ProceduralMorphologyConnectomeRouter.

ProceduralUniqueChannelConnectomeRouter.init

Source

__init__(self, *args, synaptic_efficacy_bank: QuantizedSynapticEfficacyBank | None=None, external_trajectories: torch.Tensor | None=None, external_seed: int=0, external_excitatory_gain: float=1.0, external_inhibitory_gain: float=1.0, max_external_event_count: int=32, allow_repeated_source_target_pairs: bool=False, **kwargs) -> None

Parameter	Type	Default
args	unspecified	variadic
synaptic_efficacy_bank	QuantizedSynapticEfficacyBank | None	None
external_trajectories	torch.Tensor | None	None
external_seed	int	0
external_excitatory_gain	float	1.0
external_inhibitory_gain	float	1.0
max_external_event_count	int	32
allow_repeated_source_target_pairs	bool	False
kwargs	unspecified	variadic

Returns None.

ProceduralUniqueChannelConnectomeRouter.queue_bytes

Source

ProceduralUniqueChannelConnectomeRouter.queue_bytes: int

Read-only property. Access as instance.queue_bytes; do not call it as a function.

Returns int.

ProceduralUniqueChannelConnectomeRouter.recorded_offsets_per_source

Source

ProceduralUniqueChannelConnectomeRouter.recorded_offsets_per_source: int

Read-only property. Access as instance.recorded_offsets_per_source; do not call it as a function.

Candidate offset slots needed to record one routed source.

Returns int.

ProceduralUniqueChannelConnectomeRouter.clear_and_route

Source

clear_and_route(self, active_sources: torch.Tensor, *, current_slot: int, current_step: int | None=None, clear_consumed: bool=True, touched_offsets: torch.Tensor | None=None, unique_rows: torch.Tensor | None=None, unique_row_flags: torch.Tensor | None=None, unique_row_count: torch.Tensor | None=None) -> torch.Tensor

Route the exact selected typed channel for each active source.

Parameter	Type	Default
active_sources	torch.Tensor	required
current_slot	int	required
current_step	int | None	None
clear_consumed	bool	True
touched_offsets	torch.Tensor | None	None
unique_rows	torch.Tensor | None	None
unique_row_flags	torch.Tensor | None	None
unique_row_count	torch.Tensor | None	None

Returns torch.Tensor.