MCPcopy Create free account
hub / github.com/CrazyBoyM/dreambooth-for-diffusion / create_vae_diffusers_config

Function create_vae_diffusers_config

tools/ckpt2diffusers.py:246–267  ·  view source on GitHub ↗

Creates a config for the diffusers based on the config of the LDM model.

(original_config)

Source from the content-addressed store, hash-verified

244
245
246def create_vae_diffusers_config(original_config):
247 """
248 Creates a config for the diffusers based on the config of the LDM model.
249 """
250 vae_params = original_config.model.params.first_stage_config.params.ddconfig
251 _ = original_config.model.params.first_stage_config.params.embed_dim
252
253 block_out_channels = [vae_params.ch * mult for mult in vae_params.ch_mult]
254 down_block_types = ["DownEncoderBlock2D"] * len(block_out_channels)
255 up_block_types = ["UpDecoderBlock2D"] * len(block_out_channels)
256
257 config = dict(
258 sample_size=vae_params.resolution,
259 in_channels=vae_params.in_channels,
260 out_channels=vae_params.out_ch,
261 down_block_types=tuple(down_block_types),
262 up_block_types=tuple(up_block_types),
263 block_out_channels=tuple(block_out_channels),
264 latent_channels=vae_params.z_channels,
265 layers_per_block=vae_params.num_res_blocks,
266 )
267 return config
268
269
270def create_diffusers_schedular(original_config):

Callers 1

ckpt2diffusers.pyFile · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected