Hard to train mamba in other downstream tasks

state-spaces / mamba

Mamba SSM architecture

Apache License 2.0

11.88k stars 988 forks source link

Open Jamesgender opened 3 months ago

Jamesgender commented 3 months ago

When I want to train mamba in other downstream tasks, it is hard to get good results. Any ideas?

poult-lab commented 2 months ago

What kind of downstream tasks did you try? classification tasks?

Jamesgender commented 2 months ago

Monocular depth estimation. A regression task. Using mamba as a feature encoder and then estimate the depth of the picture.