measurements#

Runtime application helpers for named vLLM measurement contracts.

Classes

VllmMeasurement

Functions

apply_vllm_measurement

Return an isolated runtime config for exactly one measurement.

measurement_workload

Project one measurement to the exact MIP selector fields.

normalize_vllm_measurements

Normalize named settings while preserving legacy output behavior.
class VllmMeasurement#

Bases: object

__init__(measurement_id, prefill_seq_len, generation_seq_len, batch_size, max_num_seqs, granularity, runtime_stats, model_hidden_sizes=(), description='', legacy=False)#
Parameters:

  • measurement_id (str)

  • prefill_seq_len (int)

  • generation_seq_len (int)

  • batch_size (int)

  • max_num_seqs (int)

  • granularity (str)

  • runtime_stats (Mapping[str, Any])

  • model_hidden_sizes (tuple[int, ...])

  • description (str)

  • legacy (bool)

Return type:

None

batch_size: int#
description: str = ''#
generation_seq_len: int#
property gpu_group_size: int#
granularity: str#
property identity: str#
legacy: bool = False#
max_num_seqs: int#
measurement_id: str#
model_hidden_sizes: tuple[int, ...] = ()#
prefill_seq_len: int#
property relative_stats_path: Path#
runtime_stats: Mapping[str, Any]#
to_dict()#
Return type:

dict[str, Any]

property topology: Mapping[str, Any]#
apply_vllm_measurement(config, measurement)#

Return an isolated runtime config for exactly one measurement.

Parameters:
Return type:

dict[str, Any]

measurement_workload(measurement)#

Project one measurement to the exact MIP selector fields.

Parameters:

measurement (VllmMeasurement)

Return type:

dict[str, int | str]

normalize_vllm_measurements(config)#

Normalize named settings while preserving legacy output behavior.

Parameters:

config (Mapping[str, Any])

Return type:

OrderedDict[str, VllmMeasurement]