Which model is pre-trained without bidirection?

[onchiptech]

All standard pre-trained models in the Model Zoo (e.g., VideoMamba-Ti, VideoMamba-S, VideoMamba-M) are initialized with bimamba=True (Bidirectional Mamba). I require a model for causal streaming (where the model cannot see future frames).

How to use the official pre-trained weights effectively?
Do I need to pre-train your own model setting bimamba=False from scratch?
Can I use a bidirectional checkpoint in a unidirectional streaming setup?
Using bidirectional checkpoint result in severe performance degradation or functional errors?

Thanks,
Surendra