Language

unilab.training.tensor_diagnostics.NanGuardCfg¶

class unilab.training.tensor_diagnostics.NanGuardCfg[source]¶

Bases: object

Pickleable configuration for the tensor finite-state guard.

Parameters:

enabled (bool)
buffer_size (int)
max_envs_to_dump (int)
output_dir (str | None)

Methods

__init__([enabled, buffer_size, ...])

Attributes

`buffer_size`
`enabled`
`max_envs_to_dump`
`output_dir`

enabled: bool = False¶

buffer_size: int = 100¶

max_envs_to_dump: int = 5¶

output_dir: str | None = None¶

__init__(enabled=False, buffer_size=100, max_envs_to_dump=5, output_dir=None)¶

Parameters:

enabled (bool)
buffer_size (int)
max_envs_to_dump (int)
output_dir (str | None)