search

jacklishufan / Mamba-ND

Ofiicial Implementation for Mamba-ND: Selective State Space Modeling for Multi-Dimensional Data
46 stars 4 forks source link

Regarding Block2D #5

Closed arduR-O closed 3 months ago

arduR-O commented 3 months ago

Hello,

This is in regards to Block2D class. In Image_classification > src > mamba.py, in my understanding, create_block function instantiates Block2D class as

block = Block2D( d_model, mixer_cls, norm_cls=norm_cls, fused_add_norm=fused_add_norm, residual_in_fp32=residual_in_fp32, reverse=reverse, drop_rate=drop_rate, transpose=transpose, drop_path_rate=drop_path_rate, ) But I am unable to find a proper implementation of Block2D, it's simply set to nn.Identity in beginning of code.

Block2D=SplitHead2D=nn.Identity # TODO: Clan implementation and release

Is this intended or the implementation for this class yet to be released. The comment suggests the latter but I am not sure.

Thanks for you time!!

arduR-O commented 3 months ago

Ok so is_2D is set to false so Block2D is never called, no issues then.