ResidualAdapterScenario
Training scenarios · Implementation: mlsysim/engine/scenario_definitions/training.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
Residual bottleneck adapters of width to bottleneck, one at each of sites.
Notes
Equation: Adapter parameters follow calc_residual_adapter_params; total scales by insertion-site count.
Assumptions: Identical adapter shape at every site and uniform parameter storage width.
Excludes: Base-model weights, optimizer state, and adapter quality.
References
Implementation references: mlsysim.physics.calc_residual_adapter_params. 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.CompactEncoderResidualAdapters
>>> scenario.params_per_adapter
25360Attributes
| Name | Description |
|---|---|
| adapter_bytes | Storage of one adapter in decimal KB at the supplied parameter width. |
| bottleneck | |
| bytes_per_param | |
| params_per_adapter | One bottleneck adapter’s parameter count from calc_residual_adapter_params, including its defined bias terms. |
| params_total | Adapter parameter count across identical insertion sites. |
| sites | |
| total_bytes | Storage across all adapter sites in decimal MB; excludes the base network and training state. |
| width |
width
Type: int.
Declared default: required.
bottleneck
Type: int.
Declared default: required.
sites
Type: int.
Declared default: required.
bytes_per_param
Type: Quantity.
Declared default: required.
params_per_adapter
engine.scenario_types.ResidualAdapterScenario.params_per_adapter
One bottleneck adapter’s parameter count from calc_residual_adapter_params, including its defined bias terms.
Equation: The calculation follows the stated accounting rule; the method body shows each intermediate term.
params_total
engine.scenario_types.ResidualAdapterScenario.params_total
Adapter parameter count across identical insertion sites.
Equation: self.params_per_adapter * self.sites
adapter_bytes
engine.scenario_types.ResidualAdapterScenario.adapter_bytes
Storage of one adapter in decimal KB at the supplied parameter width.
Equation: (self.params_per_adapter * self.bytes_per_param).to(‘KB’)
total_bytes
engine.scenario_types.ResidualAdapterScenario.total_bytes
Storage across all adapter sites in decimal MB; excludes the base network and training state.
Equation: (self.params_total * self.bytes_per_param).to(‘MB’)