PersonalizationStorageScenario
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.
contexts personalized models stored as full copies or as one backbone plus adapters.
Notes
Equation: Independent copies=contexts*backbone_bytes; shared storage=backbone_bytes+contexts*adapter_bytes.
Assumptions: One backbone can be reused across all contexts and parameters share the declared storage width.
Excludes: Activation caches, optimizer state, and context-switching overhead.
References
Implementation references: mlsysim.physics.memory_from_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.CameraModelTenContextPersonalization
>>> round(scenario.total_savings, 6)
9.52381Attributes
| Name | Description |
|---|---|
| adapter_params | |
| adapter_size | One context’s adapter storage via the same parameter-to-memory conversion. |
| bytes_per_param | |
| contexts | |
| model_params | |
| model_size | Backbone weight storage via memory_from_params; bytes follow the supplied parameter-width quantity. |
| per_context_savings | Backbone-to-adapter size ratio for one context; excludes the amortized shared backbone. |
| total_copies | Storage for one complete backbone copy per context. |
| total_savings | Dimensionless ratio of independent copies to shared-backbone storage; a multiplicative saving, not a fractional reduction. |
| total_shared | Storage for one shared backbone plus one adapter per context. |
model_params
Type: Quantity.
Declared default: required.
adapter_params
Type: Quantity.
Declared default: required.
bytes_per_param
Type: Quantity.
Declared default: required.
contexts
Type: int.
Declared default: required.
model_size
engine.scenario_types.PersonalizationStorageScenario.model_size
Backbone weight storage via memory_from_params; bytes follow the supplied parameter-width quantity.
Equation: The calculation follows the stated accounting rule; the method body shows each intermediate term.
adapter_size
engine.scenario_types.PersonalizationStorageScenario.adapter_size
One context’s adapter storage via the same parameter-to-memory conversion.
Equation: The calculation follows the stated accounting rule; the method body shows each intermediate term.
total_copies
engine.scenario_types.PersonalizationStorageScenario.total_copies
Storage for one complete backbone copy per context.
Equation: self.contexts * self.model_size
total_savings
engine.scenario_types.PersonalizationStorageScenario.total_savings
Dimensionless ratio of independent copies to shared-backbone storage; a multiplicative saving, not a fractional reduction.
Equation: (self.total_copies / self.total_shared).to_base_units().magnitude
per_context_savings
engine.scenario_types.PersonalizationStorageScenario.per_context_savings
Backbone-to-adapter size ratio for one context; excludes the amortized shared backbone.
Equation: (self.model_size / self.adapter_size).to_base_units().magnitude