polars.lazyframe.engine_config.GPUEngine

class polars.lazyframe.engine_config.GPUEngine(
    *,
    device: int | None = None,
    memory_resource: Any | None = None,
    raise_on_fail: bool = False,
    monitoring: bool = False,
    **kwargs: Any,
)

Configuration options for the GPU execution engine.

Use this if you want control over details of the execution.

Parameters:

deviceint, default None

Select the GPU used to run the query. If not provided, the query uses the current CUDA device.

memory_resourcermm.mr.DeviceMemoryResource, default None

Provide a memory resource for GPU memory allocations.

Warning

If passing a memory_resource, you must ensure that it is valid for the selected device. See the RMM documentation for more details.

raise_on_failbool, default False

If True, do not fall back to the Polars CPU engine if the GPU engine cannot execute the query, but instead raise an error.

__init__(
    *,
    device: int | None = None,
    memory_resource: Any | None = None,
    raise_on_fail: bool = False,
    monitoring: bool = False,
    **kwargs: Any,
) → None

Methods

__init__(*[, device, memory_resource, ...])
collect(lf, *, optimizations[, background, ...])	Execute lf, returning a DataFrame or a background query handle.
collect_all(lfs, *, optimizations)	Execute several queries, potentially in parallel.
collect_all_async(lfs, *, optimizations[, ...])	Execute several queries asynchronously.
collect_async(lf, *, optimizations[, gevent])	Execute lf asynchronously.
collect_batches(lf, *, optimizations[, ...])	Execute lf, yielding its result in batches.
execute(lf, *, optimizations)	Execute the query into a QueryResult.
sink_batches(lf, function, *, chunk_size, ...)	See polars.LazyFrame.sink_batches().
sink_csv(lf, path, *, include_bom, ...)	See polars.LazyFrame.sink_csv().
sink_ipc(lf, path, *, compression, ...)	See polars.LazyFrame.sink_ipc().
sink_ndjson(lf, path, *, compression, ...)	See polars.LazyFrame.sink_ndjson().
sink_parquet(lf, path, *, compression, ...)	See polars.LazyFrame.sink_parquet().

Attributes

monitoring	Whether queries are monitored (None uses the configured default).
name	Name of the engine.
plan_engine	Engine name used to render query plans.
device	Device on which to run query.
memory_resource	Memory resource to use for device allocations.
raise_on_fail	Whether unsupported queries should raise an error, rather than falling back to the CPU engine.
config	Additional configuration options for the engine.