KernelShareScenario
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.
One kernel class’s multiply-adds against a whole model’s, per example.
Notes
Equation: Kernel share is kernel_macs/total_macs for the same example.
Assumptions: Both counts use the same MAC convention and the total is nonzero.
Excludes: Latency attribution and memory traffic; an operation share is not a runtime share.
References
Derivation: The kernel MAC count is a subset of total model MACs for the same example. Their ratio is an accounting fraction; converting that fraction into elapsed-time attribution would additionally require execution and memory information.
Examples
Attributes
| Name | Description |
|---|---|
| kernel_macs | |
| share | Dimensionless fraction of model multiply-accumulates attributed to this kernel class. |
| total_macs | |
| workload |
workload
Type: Workload.
Declared default: required.
kernel_macs
Type: Quantity.
Declared default: required.
total_macs
Type: Quantity.
Declared default: required.
share
engine.scenario_types.KernelShareScenario.share
Dimensionless fraction of model multiply-accumulates attributed to this kernel class.
Equation: (kernel_macs / total_macs).to('').magnitude.