SingleNodeInferenceScenario
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.
One inference of workload on hardware (roofline profile from SingleNodeModel).
Notes
Equation: Compose mlsysim.solvers.SingleNodeModel.
Assumptions: The generic SingleNodeModel controls work, traffic, and feasibility; optional precision and efficiency override its defaults.
Excludes: The generic activation-traffic heuristic is not a KV-aware serving trace.
References
Derivation: this scenario binds its input fields to the referenced implementation contracts: mlsysim.solvers.SingleNodeModel. The methods below identify any additional arithmetic.
Examples
>>> from mlsysim import Scenarios
>>> case = Scenarios.ResNet50A100Batch1
>>> type(case.result).__name__
'PerformanceProfile'Attributes
| Name | Description |
|---|---|
| batch_size | |
| efficiency | |
| hardware | |
| intensity | Inference work per modeled byte moved for the batch. Equation: |
| precision | |
| result | Evaluate this scenario through mlsysim.solvers.SingleNodeModel.solve. Equation: pass |
| ridge | Peak compute-to-bandwidth ratio of the evaluated device profile. Equation: |
| traffic | Memory bytes implied by the generic roofline profile. Equation: ``result.latency_memory |
| workload |
workload
Type: Workload.
Declared default: required.
hardware
Type: HardwareNode.
Declared default: required.
batch_size
Type: int.
Declared default: 1.
precision
Type: Optional[str].
Declared default: None.
efficiency
Type: Optional[float].
Declared default: None.
result
engine.scenario_types.SingleNodeInferenceScenario.result
Evaluate this scenario through mlsysim.solvers.SingleNodeModel.solve. Equation: pass the stated fields to that implementation; its contract defines the calculation and feasibility tests. Units: each returned field follows the referenced result schema; no independent equation is evaluated by this wrapper.
traffic
engine.scenario_types.SingleNodeInferenceScenario.traffic
Memory bytes implied by the generic roofline profile. Equation: result.latency_memory * result.peak_bw_actual. Units: bytes; this inherits the engine’s activation-traffic heuristic rather than a KV trace.
intensity
engine.scenario_types.SingleNodeInferenceScenario.intensity
Inference work per modeled byte moved for the batch. Equation: workload.inference_flops * batch_size / traffic. Units: FLOPs/byte; requires a workload with stated forward work.
ridge
engine.scenario_types.SingleNodeInferenceScenario.ridge
Peak compute-to-bandwidth ratio of the evaluated device profile. Equation: result.peak_flops_actual / result.peak_bw_actual. Units: FLOPs/byte; this is the peak ridge, not an efficiency-adjusted ridge.