oom problem during train

[hanjoonwon]

my gpu is gv100 with 12gb vram

and my config is
parent: projects/neuralangelo/configs/base.yaml
data:
readjust:
center:
- 0.0
- 0.0
- 0.0
scale: 1.0
root: /home/xiangyue/joonwon/neuralangelo/data/owl/
train:
image_size:
- 800
- 600
type: projects.neuralangelo.data
val:
image_size:
- 401
- 300
model:
appear_embed:
enabled: false
object:
sdf:
encoding:
dict_size: 19
dim: 4
coarse2fine:
init_active_level: 4
mlp:
inside_out: false

but occur oom problem
how can i solve it??
I tried reducing the train image size and also tried dict-size to 17.