FiLMMonitorCallback#

class FiLMMonitorCallback(*args, **kwargs)#

Bases: Callback

Logs a compact FiLM diagnostic text report to wandb.

Parameters:

  • log_every_n_steps (int) – How often to log (in global steps).

  • num_film_layers (int) – Number of FiLM layers per generator.

  • film_on_pos_embed (bool) – If True, the first FiLM layer is pos-embed.

  • film_after_pos_embed (bool) – If True, the pos-embed FiLM is applied after sin() (i.e. gamma * sin(x) + beta), not before.

__init__(
log_every_n_steps=50,
num_film_layers=3,
film_on_pos_embed=True,
film_after_pos_embed=True,
)#
Parameters:

  • log_every_n_steps (int)

  • num_film_layers (int)

  • film_on_pos_embed (bool)

  • film_after_pos_embed (bool)

on_fit_start(trainer, pl_module)#
Parameters:

  • trainer (pytorch_lightning.Trainer)

  • pl_module (pytorch_lightning.LightningModule)

Return type:

None

on_train_batch_end(
trainer,
pl_module,
outputs,
batch,
batch_idx,
)#
on_fit_end(trainer, pl_module)#