Skip to content

Latest commit

 

History

History
56 lines (44 loc) · 2.84 KB

README.md

File metadata and controls

56 lines (44 loc) · 2.84 KB

aca-model

Core lifecycle model for the ACA structural retirement project.

The IID transitory medical-cost shock is read only as its draw next_hcc_transitory, so pylcm integrates it out and no regime carries it as a state. The benchmark declares the asset transition as resources plus the negated medical bill; to measure the plain asset law instead, pass grid_config=replace(BENCHMARK_GRID_CONFIG, subtract_hcc_bill=False) to create_benchmark_model, and the same grid configuration to get_benchmark_initial_conditions.

build_model_spec(...) returns the model's domain: regimes, ages, edges, broadcast slots and the default entry initial_nodes. policy=None builds the baseline; a PolicyVariant overlays its ACA schedules. create_model(spec=..., fixed_params=..., **model_options) forwards every other lcm.Model option unchanged, and an option the spec also holds, such as initial_nodes, overrides it. create_benchmark_model takes the same options.

Without execution_config, pylcm's default applies: the budget is the smallest selected device pool limit less headroom, which requires a preallocated allocator pool (XLA_PYTHON_CLIENT_PREALLOCATE not false); CPU runs unbudgeted.

To combine a measured budget with device selection and sharding:

from dataclasses import replace

from aca_model.execution import execution_config_for_devices

execution_config = replace(
    execution_config_for_devices(devices=(0, 1, 2)),
    sharded_states=("pref_type",),
)

Pass this policy as execution_config to create_model or create_benchmark_model. Record its exact device_memory_bytes, the selected devices' allocator observations, executed precision and selected program widths with performance evidence. The budget is an allocator ceiling; it does not establish that all workload phases fit.

Explicit policies pass through unchanged. In particular, ExecutionConfig(device_memory_bytes=None) is an unbudgeted control and retains pylcm's bootstrap width selection. Economic grids, numerical solver settings and result retention are separate choices.

GridConfig accepts economic grid sizes and numerical solver choices only. Execution controls belong to ExecutionConfig: sharded state names go in sharded_states, declared program widths in axis_widths, and the per-device byte ceiling in device_memory_bytes. Obsolete grid execution keywords are rejected. A state-specific batch size has no automatic conversion to a flattened program width; callers must choose and record an explicit execution policy.

For comparisons with an older pylcm version, retain its model/configuration pin and use a separately reviewed API adapter. Preserve grids, transition declarations, numerical solver choices, retention and seeds across the comparison. An unspecified width in the new policy means planner selection, not a translation of an old zero-valued batch size.