minimax-H3 FL2VA train error

[stihuangyuan]

[rank5]: noise_pred, noise_pred_audio = pipe.model_fn(
[rank5]: ^^^^^^^^^^^^^^
[rank5]: TypeError: model_fn_minimax_h3() missing 1 required positional argument: 'audio_latents'

[mi804]

Thanks for the report.

This error means audio_latents was never produced for that sample, which happens when a video in your dataset has no audio track — without input_audio, the audio encoding step is skipped and model_fn fails on the missing argument. Only rank 5 crashed, so likely just a subset of your data is affected.

Try adding --silent_on_missing_audio to your training command — it substitutes a silent track for those samples.

If the error persists, please share your full training command.