Very stupid question on using models generated from cfgs

[MuXiaoYun]

I am trying to do a simple classfication task with 9 classes using xyzs of points. Here's the cfg file I am using:

    model:
 NAME: BaseCls
 encoder_args:
   NAME: PointNextEncoder
   blocks: [1, 1, 1, 1, 1, 1]
   strides: [1, 2, 2, 2, 2, 1]
   width: 32
   in_channels: 3
   radius: 0.15
   radius_scaling: 1.5
   sa_layers: 2
   sa_use_res: True
   nsample: 32
   expansion: 4
   aggr_args:
     feature_type: 'dp_fj'
     reduction: 'max'
   group_args:
     NAME: 'ballquery'
     normalize_dp: True
   conv_args:
     order: conv-norm-act
   act_args:
     act: 'relu'
   norm_args:
     norm: 'bn'
 cls_args: 
   NAME: ClsHead
   num_classes: 9
   mlps: [512, 256]
   norm_args: 
     norm: 'bn1d'

My inputs shape is [32, 2048, 3]. Of course I'm expecting an output with [32, 2048, 9] or [32, 9, 2048]. However I got [32,9]. I guess the model takes [2048, 3] as one entry but I am expecting it to take [3] as one. I can't figure out how to adjust my args to make it work.
I am sorry that I'm not famaliar with this cfg-build-model process and ask this stupid question. Could u pls help me?