Wmcs91's picture
Duplicate from shi-labs/Versatile-Diffusion
fb53ec8
# vd_base:
# symbol: vd
# find_unused_parameters: true
############
# vd basic #
############
vd_basic:
super_cfg: sd_t2i
type: vd_basic
symbol: vd
find_unused_parameters: true
args:
cond_stage_config: MODEL(clip_frozen_encode_vision)
vd_basic_noema:
super_cfg: vd_basic
args:
use_ema: false
###################
# vd dual-context #
###################
vd_dc:
super_cfg: sd_t2i_fullclip
type: vd_dc
symbol: vd
find_unused_parameters: true
args:
unet_config: MODEL(openai_unet_dual_context)
vd_dc_noema:
super_cfg: vd_dc
args:
use_ema: false
######
# vd #
######
vd:
type: vd
symbol: vd
find_unused_parameters: true
args:
autokl_cfg: MODEL(sd_autoencoder)
optimus_cfg: MODEL(optimus_vae)
clip_cfg: MODEL(clip_frozen)
unet_config: MODEL(openai_unet_vd)
beta_linear_start: 0.00085
beta_linear_end: 0.012
timesteps: 1000
scale_factor: 0.18215
use_ema: true
vd_noema:
super_cfg: vd
args:
use_ema: false