Common Fields Config

These fields are generic and can be adapted based on the user's machine.

Class DefinitionConfig Example

class CommonFieldsConfig(ProjectConfig, extra=Extra.ignore):
    """"""
    artifact_path: str = "/cache"
    batch_size: int = 32
    use_cuda: Union[Literal["auto"], bool] = "auto"
    large_dask_cluster: bool = False
    read_only_config: bool = False

Example to append to the config to override the default batch_size.

{
  "batch_size": 64,
}

Artifact Path

Default value: /cache

Where to store the caching artifacts (HDF5 files and HF datasets). The value needs to be available inside Docker (see docker-compose.yml). /cache is available by default on the docker image.

Not using Docker?

If Azimuth is run without Docker, the cache needs to be a path with write access (/cache will not work).

Batch Size

Default value: 32

Batch size to use during inference. A higher batch size will make computation faster, depending on the memory available on your machine.

Use Cuda

Default value: auto

If cuda is available on your machine, set to true, otherwise false. Can also be set to "auto" and let the user-code take care of it.

Large Dask Cluster

Default value: False

The memory of the dask cluster is usually 6GB. If your models are big or if you encounter garbage collection errors, you can set the memory to 12GB by setting large_dask_cluster to True.

Read-Only Config

Default value: False

This field allows to block the changes to the config when set to True. This can be useful in certain context, such as when hosting a demo.