The LTXBaseModel.forward and _forward methods required attention_mask as
a positional argument with no default value. However, LTXV.extra_conds
only conditionally adds attention_mask to model_conds when it is present
in kwargs. If attention_mask is not provided by the text encoder, the
diffusion_model forward call fails with:
TypeError: LTXBaseModel.forward() missing 1 required positional argument: 'attention_mask'
The model already handles attention_mask=None correctly in both
_prepare_attention_mask and _prepare_context, so making the parameter
optional is the minimal safe fix. This also aligns with how
LTXAVDoubleStreamBlock.forward handles the parameter.
b8bb5427f9