The training speed is abnormally slow.

[baldyouth]

May I ask a question? I've already installed selective_scan==0.0.2, but training is extremely slow. Why is that?
I’m using an RTX 3090, and

self.vmamba = Backbone_VSSM(
    depths=[2, 2, 20, 2], dims=32, drop_path_rate=0.5, 
    patch_size=16, in_chans=3, num_classes=1, 
    ssm_d_state=1, ssm_ratio=1.0, ssm_dt_rank="auto", ssm_act_layer="silu",
    ssm_conv=3, ssm_conv_bias=False, ssm_drop_rate=0.0, 
    ssm_init="v0", forward_type="v05_noz", 
    mlp_ratio=4.0, mlp_act_layer="gelu", mlp_drop_rate=0.0, gmlp=False,
    patch_norm=True, norm_layer=("ln2d" if channel_first else "ln"), 
    downsample_version="v3", patchembed_version="v1", 
    use_checkpoint=False, posembed=False, imgsize=448,
    out_indices=[3]
)

It would be great if the author could shed some light on this.

[TIanCat]

You can try this solution, my training speed is normal
#340 (comment)

[baldyouth]

You can try this solution, my training speed is normal #340 (comment)

Thanks! I’ll give it a try.