MzeroMiko / VMamba

VMamba: Visual State Space Models,code is based on mamba

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

tiny model ckpt not match the model weights

shengyu27 opened this issue · comments

When I use the tiny model and find that some of the shapes in the model do not match the shape of the model
this is the ckpt:vssm_tiny_0230_ckpt_epoch_262.pth
model
this is the model weights:
ckpt
this is the model config:
model param

So where do you get that config file?

So where do you get that config file?

from the readme table 'Classification on ImageNet-1K with VMambav2'
https://github.com/MzeroMiko/VMamba/blob/main/classification/configs/vssm/vmambav2_tiny_224.yaml

It seems that the hyper-parameter in the picture your provide mismatch what in https://github.com/MzeroMiko/VMamba/blob/main/classification/configs/vssm/vmambav2_tiny_224.yaml, e.g. ssm_d_state=16 in your fig, but SSM_D_STATE=1 in the yaml config. Fix that and try again.

It seems that the hyper-parameter in the picture your provide mismatch what in https://github.com/MzeroMiko/VMamba/blob/main/classification/configs/vssm/vmambav2_tiny_224.yaml, e.g. ssm_d_state=16 in your fig, but SSM_D_STATE=1 in the yaml config. Fix that and try again.

oh, this is my mistake, thanks.