AttentionDilutionScenario
Agents scenarios · Implementation: mlsysim/engine/scenario_definitions/agents.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
One attention head with fixed logits: a target among background keys, optionally look-alikes.
Notes
Equation: result = AttentionDilutionModel.solve(…) with the declared target_logit, background_logit, short_keys inputs.
Assumptions: The inputs instantiate the assumptions documented by mlsysim.solvers.AttentionDilutionModel.solve; accessing result evaluates that solver again.
Excludes: Any extra mechanism beyond AttentionDilutionModel; this record supplies configuration and does not collect performance measurements.
References
Derivation: This configuration forwards its stored inputs to mlsysim.solvers.AttentionDilutionModel.solve. The solver contract records the equations and their provenance.
Examples
The named scenario supplies illustrative inputs; the output is calculated.
>>> from mlsysim import Scenarios
>>> case = Scenarios.ToyHeadAttentionDilution
>>> round(float(case.result.short_weight), 3)
0.968Attributes
| Name | Description |
|---|---|
| background_logit | |
| long_keys | |
| lookalike_keys | |
| lookalike_logit | |
| result | Modeled result from mlsysim.solvers.AttentionDilutionModel.solve using this |
| short_keys | |
| target_logit | |
| unrelated_keys |
target_logit
Type: float.
Declared default: required.
background_logit
Type: float.
Declared default: required.
short_keys
Type: int.
Declared default: required.
long_keys
Type: int.
Declared default: required.
lookalike_keys
Type: Optional[int].
Declared default: None.
lookalike_logit
Type: Optional[float].
Declared default: None.
result
engine.scenario_types.AttentionDilutionScenario.result
Modeled result from mlsysim.solvers.AttentionDilutionModel.solve using this scenario’s declared inputs; see that method for units, limits, and feasibility.
Equation: Delegates the declared fields to mlsysim.solvers.AttentionDilutionModel.solve; returned quantities and feasibility follow that contract.