mirror of https://github.com/vladmandic/automatic
66 lines
2.6 KiB
Python
66 lines
2.6 KiB
Python
import transformers
|
|
import diffusers
|
|
from modules import shared, devices, sd_models, shared_items, sd_hijack_te
|
|
from modules.logger import log
|
|
|
|
|
|
def load_meissonic(checkpoint_info, diffusers_load_config=None):
|
|
from pipelines.meissonic.transformer import Transformer2DModel as TransformerMeissonic
|
|
from pipelines.meissonic.scheduler import Scheduler as MeissonicScheduler
|
|
from pipelines.meissonic.pipeline import MeissonicPipeline
|
|
from pipelines.meissonic.pipeline_img2img import MeissonicImg2ImgPipeline
|
|
from pipelines.meissonic.pipeline_inpaint import MeissonicInpaintPipeline
|
|
shared_items.pipelines['Meissonic'] = MeissonicPipeline
|
|
|
|
if diffusers_load_config is None:
|
|
diffusers_load_config = {}
|
|
repo_id = sd_models.path_to_repo(checkpoint_info)
|
|
sd_models.hf_auth_check(checkpoint_info)
|
|
|
|
diffusers_load_config['variant'] = 'fp16'
|
|
diffusers_load_config['trust_remote_code'] = True
|
|
|
|
log.debug(f'Load model: type=Meissonic repo="{repo_id}" config={diffusers_load_config} offload={shared.opts.diffusers_offload_mode} dtype={devices.dtype} args={diffusers_load_config}')
|
|
model = TransformerMeissonic.from_pretrained(
|
|
repo_id,
|
|
subfolder="transformer",
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
**diffusers_load_config,
|
|
)
|
|
vqvae = diffusers.VQModel.from_pretrained(
|
|
repo_id,
|
|
subfolder="vqvae",
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
**diffusers_load_config,
|
|
)
|
|
text_encoder = transformers.CLIPTextModelWithProjection.from_pretrained(
|
|
repo_id,
|
|
subfolder="text_encoder",
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
)
|
|
tokenizer = transformers.CLIPTokenizer.from_pretrained(
|
|
repo_id,
|
|
subfolder="tokenizer",
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
)
|
|
scheduler = MeissonicScheduler.from_pretrained(
|
|
repo_id,
|
|
subfolder="scheduler",
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
)
|
|
pipe = MeissonicPipeline(
|
|
vqvae=vqvae.to(devices.dtype),
|
|
text_encoder=text_encoder.to(devices.dtype),
|
|
transformer=model.to(devices.dtype),
|
|
tokenizer=tokenizer,
|
|
scheduler=scheduler,
|
|
)
|
|
|
|
diffusers.pipelines.auto_pipeline.AUTO_TEXT2IMAGE_PIPELINES_MAPPING["meissonic"] = MeissonicPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["meissonic"] = MeissonicImg2ImgPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_INPAINT_PIPELINES_MAPPING["meissonic"] = MeissonicInpaintPipeline
|
|
sd_hijack_te.init_hijack(pipe)
|
|
|
|
devices.torch_gc(force=True, reason='load')
|
|
return pipe
|