Language

unilab.training.tensor_diagnostics.NanGuard¶

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

Detect bad tensor rows on-device and export dumps at an explicit host boundary.

Parameters:

Methods

`__init__`(cfg, num_envs, supports_state_playback)
`capture`(physics_state)	Retain a detached device tensor snapshot for a later abnormal dump.
`check`(obs, reward[, step])	Return device-resident bad-row indices, or `None` when every row is finite.
`check_ctrl`(ctrl[, step])	Return device-resident bad control-row indices, or `None` when finite.
`dump`(nan_env_ids, model_file, step)	Export one abnormal-state artifact at an explicit host boundary.

__init__(cfg, num_envs, supports_state_playback)[source]¶

Parameters:

capture(physics_state)[source]¶

Retain a detached device tensor snapshot for a later abnormal dump.

check(obs, reward, step=0)[source]¶

Return device-resident bad-row indices, or None when every row is finite.

Parameters:

Return type:

Tensor | None

check_ctrl(ctrl, step=0)[source]¶

Return device-resident bad control-row indices, or None when finite.

Parameters:

Return type:

Tensor | None

dump(nan_env_ids, model_file, step)[source]¶

Export one abnormal-state artifact at an explicit host boundary.

Parameters:

Return type:

str | None