mirror of
https://github.com/vladmandic/sdnext.git
synced 2026-01-27 15:02:48 +03:00
51 lines
2.0 KiB
Python
51 lines
2.0 KiB
Python
import transformers
|
|
import diffusers
|
|
from modules import shared, devices, sd_models, model_quant, sd_hijack_te
|
|
from pipelines import generic
|
|
|
|
|
|
def load_nunchaku():
|
|
import nunchaku
|
|
nunchaku_precision = nunchaku.utils.get_precision()
|
|
nunchaku_rank = 128
|
|
nunchaku_repo = f"nunchaku-tech/nunchaku-z-image-turbo/svdq-{nunchaku_precision}_r{nunchaku_rank}-z-image-turbo.safetensors"
|
|
shared.log.debug(f'Load module: quant=Nunchaku module=transformer repo="{nunchaku_repo}" attention={shared.opts.nunchaku_attention}')
|
|
transformer = nunchaku.NunchakuZImageTransformer2DModel.from_pretrained(
|
|
nunchaku_repo,
|
|
torch_dtype=devices.dtype,
|
|
cache_dir=shared.opts.hfcache_dir,
|
|
)
|
|
return transformer
|
|
|
|
|
|
def load_z_image(checkpoint_info, diffusers_load_config=None):
|
|
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)
|
|
|
|
load_args, _quant_args = model_quant.get_dit_args(diffusers_load_config, allow_quant=False)
|
|
shared.log.debug(f'Load model: type=ZImage repo="{repo_id}" config={diffusers_load_config} offload={shared.opts.diffusers_offload_mode} dtype={devices.dtype} args={diffusers_load_config}')
|
|
|
|
if model_quant.check_nunchaku('Model'): # only available model
|
|
transformer = load_nunchaku()
|
|
else:
|
|
transformer = generic.load_transformer(repo_id, cls_name=diffusers.ZImageTransformer2DModel, load_config=diffusers_load_config)
|
|
|
|
text_encoder = generic.load_text_encoder(repo_id, cls_name=transformers.Qwen3ForCausalLM, load_config=diffusers_load_config)
|
|
|
|
pipe = diffusers.ZImagePipeline.from_pretrained(
|
|
repo_id,
|
|
cache_dir=shared.opts.diffusers_dir,
|
|
transformer=transformer,
|
|
text_encoder=text_encoder,
|
|
**load_args,
|
|
)
|
|
|
|
del transformer
|
|
del text_encoder
|
|
sd_hijack_te.init_hijack(pipe)
|
|
|
|
devices.torch_gc(force=True, reason='load')
|
|
return pipe
|