IterativeDecoderBudgetScenario

Robotics scenarios · Implementation: mlsysim/engine/scenario_definitions/robotics/learning.py.

Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.

machine’s one-pass chunk decoder against an iterative head that encodes once and runs denoising_steps dependent evaluations in denoising_latency, and a flow_steps head at the same cost per evaluation, on the setpoint step and the renewal period with renewal_jitter held free.

Notes

Equation: Per-evaluation time=(denoising_latency-encode)/denoising_steps; latency=encode+evaluations*per_evaluation.

Assumptions: Equal sequential evaluation cost shared by compared decoder heads and a supplied jitter allowance.

Excludes: Convergence/quality differences, overlap, and measured worst-case execution guarantees.

References

Implementation references: mlsysim.physics.robotics.learning.calc_iterative_decoder_latency, mlsysim.physics.robotics.learning.calc_max_decoder_evaluations. These helper contracts state the source assumptions; the scenario binds their inputs and derives the displayed ratios and budgets.

Examples

The named registry scenario supplies the stated inputs; this is a modeled calculation, not a new measurement.

>>> from mlsysim import Scenarios
>>> scenario = Scenarios.LatchDecoderBudget
>>> scenario.evaluations_within_renewal
14

Attributes

Name	Description
denoising_latency
denoising_overrun_under_jitter	Denoising completion minus renewal deadline in ms after adding jitter; negative values retain slack.
denoising_slack	Renewal period minus denoising latency in ms, without jitter reserve.
denoising_steps
encode
evaluations_within_renewal	Whole evaluations admitted by renewal time after reserving jitter, using `max_evaluations`.
evaluations_within_step	Whole decoder evaluations admitted by the setpoint period, using `max_evaluations`.
flow_latency	Encoding plus `flow_steps` equal-cost evaluations via `calc_iterative_decoder_latency`; assumes the inferred evaluation cost transfers to this head.
flow_steps
machine
one_pass_end_under_jitter	One-pass completion time in ms after adding the declared jitter allowance.
one_pass_latency	Declared one-pass chunk inference latency in ms; no timing experiment is performed here.
one_pass_slack	Renewal period minus one-pass latency in ms; negative values indicate an overrun.
per_evaluation	Average dependent decoder-evaluation time in ms after subtracting the one-time encode cost from denoising latency.
renewal_jitter
renewal_less_jitter	Usable renewal time in ms after reserving jitter.
renewal_period	Machine chunk-renewal period in ms, copied from `machine.chunk_period`.
setpoint_step	Machine setpoint period in ms, copied from `machine.chunk_step`.

Methods

Name	Description
evaluation_room	Unrounded count of equal-cost evaluations fitting a supplied time budget after encoding; may be negative when encoding alone exceeds budget.
latency_at	Time for encoding plus a supplied integer evaluation count via `calc_iterative_decoder_latency`.
max_evaluations	Whole evaluations fitting a time budget via `calc_max_decoder_evaluations`; inherits the helper’s infeasible-budget handling.

evaluation_room

engine.scenario_types.IterativeDecoderBudgetScenario.evaluation_room(budget)

Unrounded count of equal-cost evaluations fitting a supplied time budget after encoding; may be negative when encoding alone exceeds budget.

Equation: ((budget - self.encode) / self.per_evaluation).to(’’).magnitude

latency_at

engine.scenario_types.IterativeDecoderBudgetScenario.latency_at(evaluations)

Time for encoding plus a supplied integer evaluation count via calc_iterative_decoder_latency.

Equation: The calculation follows the stated accounting rule; the method body shows each intermediate term.

max_evaluations

engine.scenario_types.IterativeDecoderBudgetScenario.max_evaluations(budget)

Whole evaluations fitting a time budget via calc_max_decoder_evaluations; inherits the helper’s infeasible-budget handling.

Equation: The calculation follows the stated accounting rule; the method body shows each intermediate term.