VlaEvaluationScenario
Robotics scenarios · Implementation: mlsysim/engine/scenario_definitions/robotics/proposers.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
One fresh evaluation of machine’s intent model on its application processor.
Each input left as None takes the machine’s deployed value: the weight precision, the input and action token counts, and the decoding mode. The sustained-bandwidth fraction and compute utilization are the machine’s.
Notes
Equation: evaluation = VlaEvaluationModel.solve(model, processor, selected precision and token counts, supplied efficiencies).
Assumptions: An omitted override takes its value from the machine configuration; compute and bandwidth efficiencies remain declared inputs.
Excludes: Scheduling and control guarantees beyond the underlying VLA model contract.
References
Derivation: This configuration forwards its stored inputs to mlsysim.solvers.VlaEvaluationModel.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.IntentModelEvaluationFp16
>>> round(float(case.bytes_per_param.to('B/param').magnitude), 3)
2.0Attributes
| Name | Description |
|---|---|
| action_tokens | |
| bytes_per_param | Stored weight bytes per parameter, taking the machine’s deployed format unless |
| input_tokens | |
| machine | |
| parallel_decode | |
| result | Modeled result from mlsysim.solvers.VlaEvaluationModel.solve using this |
| tokens_in | Input-token count, taking the deployed machine setting unless explicitly |
| tokens_out | Output-token count, taking the model’s action dimension unless explicitly |
| weight_bytes_per_param |
machine
Type: Any.
Declared default: required.
weight_bytes_per_param
Type: Optional[Quantity].
Declared default: None.
input_tokens
Type: Optional[int].
Declared default: None.
action_tokens
Type: Optional[int].
Declared default: None.
parallel_decode
Type: Optional[bool].
Declared default: None.
bytes_per_param
engine.scenario_types.VlaEvaluationScenario.bytes_per_param
Stored weight bytes per parameter, taking the machine’s deployed format unless overridden.
Equation: return machine.intent_weight_bytes_per_param if weight_bytes_per_param is None else weight_bytes_per_param.
tokens_in
engine.scenario_types.VlaEvaluationScenario.tokens_in
Input-token count, taking the deployed machine setting unless explicitly overridden.
Equation: return machine.intent_input_tokens if input_tokens is None else input_tokens.
tokens_out
engine.scenario_types.VlaEvaluationScenario.tokens_out
Output-token count, taking the model’s action dimension unless explicitly overridden.
Equation: return machine.intent_model.action_dim if action_tokens is None else action_tokens.
result
engine.scenario_types.VlaEvaluationScenario.result
Modeled result from mlsysim.solvers.VlaEvaluationModel.solve using this scenario’s declared inputs; see that method for units, limits, and feasibility.
Equation: Delegates the declared fields to mlsysim.solvers.VlaEvaluationModel.solve; returned quantities and feasibility follow that contract.