56 lines
1.6 KiB
Python
56 lines
1.6 KiB
Python
import sys
|
|
from argparse import Namespace
|
|
|
|
|
|
sys.path.append("../")
|
|
from helios.modules.transformer_helios import HeliosTransformer3DModel
|
|
from helios.pipelines.pipeline_helios import HeliosPipeline
|
|
from helios.utils.utils_base import load_extra_components
|
|
|
|
|
|
transformer_additional_kwargs = {
|
|
"has_multi_term_memory_patch": False,
|
|
"zero_history_timestep": False,
|
|
"guidance_cross_attn": False,
|
|
"restrict_self_attn": False,
|
|
"is_train_restrict_lora": False,
|
|
"restrict_lora": False,
|
|
"restrict_lora_rank": 128,
|
|
}
|
|
|
|
transformer = HeliosTransformer3DModel.from_pretrained(
|
|
"Wan-AI/Wan2.1-T2V-14B-Diffusers",
|
|
subfolder="transformer",
|
|
transformer_additional_kwargs=transformer_additional_kwargs,
|
|
)
|
|
pipe = HeliosPipeline.from_pretrained(
|
|
"Wan-AI/Wan2.1-T2V-14B-Diffusers",
|
|
transformer=transformer,
|
|
)
|
|
|
|
pipe.load_lora_weights(
|
|
"ablation_14B_single-res_length-21_t2v_rank128_bs128_new/checkpoint-1000/pytorch_lora_weights.safetensors",
|
|
adapter_name="default",
|
|
)
|
|
pipe.set_adapters(["default"], adapter_weights=[1.0])
|
|
|
|
|
|
args = Namespace()
|
|
if not hasattr(args, "training_config"):
|
|
args.training_config = Namespace()
|
|
args.training_config.is_enable_stage1 = True
|
|
args.training_config.restrict_self_attn = True
|
|
args.training_config.is_amplify_history = True
|
|
args.training_config.is_use_gan = True
|
|
load_extra_components(
|
|
args,
|
|
transformer,
|
|
"ablation_14B_single-res_length-21_t2v_rank128_bs128_new/checkpoint-1000/transformer_partial.pth",
|
|
)
|
|
|
|
pipe.fuse_lora()
|
|
pipe.unload_lora_weights()
|
|
pipe.transformer.save_pretrained(
|
|
"1_formal_ckpts/ablation_14B_single-res_length-21_t2v_rank128_bs128_new_e1000/transformer"
|
|
)
|