Population simulation contracts

Module contract

The immutable simulation contract specifies the connected workload and timing boundary. PopulationInferenceProfile specifies a deployment recipe. These values are declarations, not measured speed results. Check the technical report’s measured throughput at the contact count and population scale relevant to your application.

Source revision: 306a51ed950b. Public export index.

LargePopulationSimulationContract

Source

Fully specified model and network workload for a population step.

LargePopulationSimulationContract(name: str, population_size: int, step_ms: float, input_step_ms: float, output_step_ms: float, target_latency_ms: float, model_family: str, patch_size: int, temporal_execution_mode: str, patch_phase_mode: str, patch_phase_assignment: str, patch_phase_storage_layout: str, lossless_ordered_temporal_io: bool, causal_block_feedback_required: bool, minimum_routing_delay_steps: int, synaptic_efficacy_values_per_neuron: int, adaptation_storage_bits: int, activation_storage_bits: int, recurrent_state_storage_bits: int, external_drive_storage_bits: int, morphology_classes: int, firing_rate: float, fanout: int, recurrent_contact_fraction: float, excitatory_fraction: float, local_connection_fraction: float, spatial_tile_neurons: int, delay_slots: int, branch_count: int, local_source_identity_required: bool, long_range_source_identity_required: bool, local_branch_input_mode: str, input_channel_collision_mode: str, local_source_summary_is_lossless_for_declared_router: bool, posthoc_source_projection_allowed: bool, morphology_assignment_independent: bool, activity_must_be_model_generated: bool, quota_activity_forcing_allowed: bool, activity_selection_mode: str, adaptation_reference_envelope_validation_required: bool, morphology_aware_branch_targeting: bool = True, source_type_aware_branch_targeting: bool = True, source_discovery_included: bool = True, event_delivery_included: bool = True, delay_queue_included: bool = True, every_neuron_has_custom_adaptation: bool = True, empirical_connectome: bool = False) -> None

Fields

Parameter	Type	Default
name	str	required
population_size	int	required
step_ms	float	required
input_step_ms	float	required
output_step_ms	float	required
target_latency_ms	float	required
model_family	str	required
patch_size	int	required
temporal_execution_mode	str	required
patch_phase_mode	str	required
patch_phase_assignment	str	required
patch_phase_storage_layout	str	required
lossless_ordered_temporal_io	bool	required
causal_block_feedback_required	bool	required
minimum_routing_delay_steps	int	required
synaptic_efficacy_values_per_neuron	int	required
adaptation_storage_bits	int	required
activation_storage_bits	int	required
recurrent_state_storage_bits	int	required
external_drive_storage_bits	int	required
morphology_classes	int	required
firing_rate	float	required
fanout	int	required
recurrent_contact_fraction	float	required
excitatory_fraction	float	required
local_connection_fraction	float	required
spatial_tile_neurons	int	required
delay_slots	int	required
branch_count	int	required
local_source_identity_required	bool	required
long_range_source_identity_required	bool	required
local_branch_input_mode	str	required
input_channel_collision_mode	str	required
local_source_summary_is_lossless_for_declared_router	bool	required
posthoc_source_projection_allowed	bool	required
morphology_assignment_independent	bool	required
activity_must_be_model_generated	bool	required
quota_activity_forcing_allowed	bool	required
activity_selection_mode	str	required
adaptation_reference_envelope_validation_required	bool	required
morphology_aware_branch_targeting	bool	True
source_type_aware_branch_targeting	bool	True
source_discovery_included	bool	True
event_delivery_included	bool	True
delay_queue_included	bool	True
every_neuron_has_custom_adaptation	bool	True
empirical_connectome	bool	False

population_size: Number of persistent neurons. patch_size: Native timesteps represented by one block.

LargePopulationSimulationContract.steps_per_second

Source

LargePopulationSimulationContract.steps_per_second: float

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

Returns float.

LargePopulationSimulationContract.target_neuron_steps_per_second

Source

LargePopulationSimulationContract.target_neuron_steps_per_second: float

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

Returns float.

LargePopulationSimulationContract.largest_patch_phase_population

Source

LargePopulationSimulationContract.largest_patch_phase_population: int

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

Returns int.

LargePopulationSimulationContract.active_sources_per_step

Source

LargePopulationSimulationContract.active_sources_per_step: int

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

Returns int.

LargePopulationSimulationContract.events_per_step

Source

LargePopulationSimulationContract.events_per_step: int

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

Returns int.

LargePopulationSimulationContract.synaptic_efficacy_bank_bytes

Source

LargePopulationSimulationContract.synaptic_efficacy_bank_bytes: int

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

Returns int.

LargePopulationSimulationContract.delay_queue_bytes

Source

LargePopulationSimulationContract.delay_queue_bytes: int

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

Returns int.

LargePopulationSimulationContract.external_drive_block_bytes

Source

LargePopulationSimulationContract.external_drive_block_bytes: int

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

Returns int.

LargePopulationSimulationContract.bytes_to_gib

Source

@staticmethod
bytes_to_gib(byte_count: int) -> float

Parameter	Type	Default
byte_count	int	required

Returns float.

PopulationInferenceProfile

Source

Named operating point within the million-neuron contract.

PopulationInferenceProfile(name: str, support_model_id: str, support_token_width: int, support_state_width: int, support_block_rows: int, support_block_token: int, support_block_state: int, external_block_width: int, project_block_width: int, compiled_adaptation_values_per_neuron: int, morphology_adapter_values_per_class: int, synaptic_efficacy_values_per_neuron: int, output_export_fraction: float, primary_metric: str, minimum_neuron_steps_per_second: float, measured_median_neuron_steps_per_second: float, measured_p90_neuron_steps_per_second: float, measured_voltage_sera_mv2: float, measured_voltage_root_sera_mv: float, measured_spike_mean_f1: float, measured_population_firing_rate_hz: float, population_adaptation_initialization: str, population_adaptation_individually_fitted: bool, individual_behavior_adaptation_supported: bool, individual_synaptic_adaptation_supported: bool) -> None

Fields

Parameter	Type	Default
name	str	required
support_model_id	str	required
support_token_width	int	required
support_state_width	int	required
support_block_rows	int	required
support_block_token	int	required
support_block_state	int	required
external_block_width	int	required
project_block_width	int	required
compiled_adaptation_values_per_neuron	int	required
morphology_adapter_values_per_class	int	required
synaptic_efficacy_values_per_neuron	int	required
output_export_fraction	float	required
primary_metric	str	required
minimum_neuron_steps_per_second	float	required
measured_median_neuron_steps_per_second	float	required
measured_p90_neuron_steps_per_second	float	required
measured_voltage_sera_mv2	float	required
measured_voltage_root_sera_mv	float	required
measured_spike_mean_f1	float	required
measured_population_firing_rate_hz	float	required
population_adaptation_initialization	str	required
population_adaptation_individually_fitted	bool	required
individual_behavior_adaptation_supported	bool	required
individual_synaptic_adaptation_supported	bool	required

MILLION_NEURON_REALTIME_CONTRACT

Source

MILLION_NEURON_REALTIME_CONTRACT = LargePopulationSimulationContract(name='million-neuron-realtime-rtx5090', population_size=1048576, step_ms=1.0, input_step_ms=1.0, output_step_ms=1.0, target_latency_ms=1.0, model_family='axosim-lite-causal-block-gru-regression', patch_size=4, temporal_execution_mode='causal_next_block_forecast', patch_phase_mode='staggered', patch_phase_assignment='morphology_role_stratified_random', patch_phase_storage_layout='phase_tile_major', lossless_ordered_temporal_io=True, causal_block_feedback_required=True, minimum_routing_delay_steps=4, synaptic_efficacy_values_per_neuron=362, adaptation_storage_bits=8, activation_storage_bits=16, recurrent_state_storage_bits=16, external_drive_storage_bits=8, morphology_classes=5, firing_rate=0.003869090909090909, fanout=1448, recurrent_contact_fraction=1.0 / 4.0, excitatory_fraction=0.8, local_connection_fraction=0.9, spatial_tile_neurons=4096, delay_slots=8, branch_count=118, local_source_identity_required=True, long_range_source_identity_required=True, local_branch_input_mode='binary_channels_then_exact_weighted_branch_currents', input_channel_collision_mode='binary_or_before_branch_weighting', local_source_summary_is_lossless_for_declared_router=True, posthoc_source_projection_allowed=False, morphology_assignment_independent=True, activity_must_be_model_generated=True, quota_activity_forcing_allowed=False, activity_selection_mode='fixed_model_score_threshold_crossing', adaptation_reference_envelope_validation_required=True)

AXOSIM_POPULATION_PROFILE

Source

AXOSIM_POPULATION_PROFILE = PopulationInferenceProfile(name='axosim-population', support_model_id='support_t24_s8_distilled_i600', support_token_width=24, support_state_width=8, support_block_rows=32, support_block_token=32, support_block_state=16, external_block_width=32, project_block_width=64, compiled_adaptation_values_per_neuron=88, morphology_adapter_values_per_class=88, synaptic_efficacy_values_per_neuron=362, output_export_fraction=0.01, primary_metric='neuron_steps_per_second', minimum_neuron_steps_per_second=MILLION_NEURON_REALTIME_CONTRACT.target_neuron_steps_per_second, measured_median_neuron_steps_per_second=1201327144.0976055, measured_p90_neuron_steps_per_second=1177454556.7997348, measured_voltage_sera_mv2=225.82762062058336, measured_voltage_root_sera_mv=15.02756203183282, measured_spike_mean_f1=0.35383167978104685, measured_population_firing_rate_hz=3.956115245819092, population_adaptation_initialization='five_fitted_morphology_rows_plus_direct_w8_behavior_residuals', population_adaptation_individually_fitted=False, individual_behavior_adaptation_supported=True, individual_synaptic_adaptation_supported=True)