ExpertWeightTrafficScenario
Optimization scenarios · Implementation: mlsysim/engine/scenario_definitions/optimization.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
Weight bytes of the mixture-of-experts workload, resident and read per token, against a dense model of dense_parameters that reads every weight each step.
Notes
Equation: resident_bytes is the worked quantity; each property below states its equation and units.
Assumptions: Resident storage uses total MoE parameters while per-token expert traffic uses active parameters.
Excludes: Routing skew, expert-cache behavior, activation traffic, and communication.
References
Derivation: The equations below follow the stated counts, geometric relations, or accounting definitions; scenario inputs are supplied assumptions or observations, not fitted predictions.
Examples
Execute the registered worked inputs; this is a modeled calculation.
>>> from mlsysim import Scenarios
>>> case = Scenarios.DeepSeekV3WeightTraffic
>>> value = case.resident_bytes
>>> print(f"{value:.3f~P}")
1342.000 GBAttributes
| Name | Description |
|---|---|
| active_bytes | Active MoE weight traffic in GB using active parameter count. |
| bytes_per_param | |
| dense_bytes | Dense comparison-model weight traffic in GB. |
| dense_parameters | |
| dense_to_active | Dimensionless dense-to-active MoE weight ratio. |
| resident_bytes | Resident MoE weight storage in GB using total parameter count. |
| resident_to_active | Dimensionless resident-to-active MoE weight ratio. |
| workload |
workload
Type: Workload.
Declared default: required.
dense_parameters
Type: Quantity.
Declared default: required.
bytes_per_param
Type: Quantity.
Declared default: required.
resident_bytes
engine.scenario_types.ExpertWeightTrafficScenario.resident_bytes
Resident MoE weight storage in GB using total parameter count.
Equation: return memory_from_params(self.workload.parameters, self.bytes_per_param).to(‘GB’)
active_bytes
engine.scenario_types.ExpertWeightTrafficScenario.active_bytes
Active MoE weight traffic in GB using active parameter count.
Equation: return memory_from_params(self.workload.active_parameters, self.bytes_per_param).to(‘GB’)
dense_bytes
engine.scenario_types.ExpertWeightTrafficScenario.dense_bytes
Dense comparison-model weight traffic in GB.
Equation: return memory_from_params(self.dense_parameters, self.bytes_per_param).to(‘GB’)
resident_to_active
engine.scenario_types.ExpertWeightTrafficScenario.resident_to_active
Dimensionless resident-to-active MoE weight ratio.
Equation: return (self.resident_bytes / self.active_bytes).to(’’).magnitude
dense_to_active
engine.scenario_types.ExpertWeightTrafficScenario.dense_to_active
Dimensionless dense-to-active MoE weight ratio.
Equation: return (self.dense_bytes / self.active_bytes).to(’’).magnitude