init upload

This commit is contained in:
WanX-Video-1
2025-02-25 22:07:47 +08:00
commit 65386b2e03
45 changed files with 8548 additions and 0 deletions

42
wan/configs/__init__.py Normal file
View File

@@ -0,0 +1,42 @@
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import copy
import os
os.environ['TOKENIZERS_PARALLELISM'] = 'false'
from .wan_i2v_14B import i2v_14B
from .wan_t2v_1_3B import t2v_1_3B
from .wan_t2v_14B import t2v_14B
# the config of t2i_14B is the same as t2v_14B
t2i_14B = copy.deepcopy(t2v_14B)
t2i_14B.__name__ = 'Config: Wan T2I 14B'
WAN_CONFIGS = {
't2v-14B': t2v_14B,
't2v-1.3B': t2v_1_3B,
'i2v-14B': i2v_14B,
't2i-14B': t2i_14B,
}
SIZE_CONFIGS = {
'720*1280': (720, 1280),
'1280*720': (1280, 720),
'480*832': (480, 832),
'832*480': (832, 480),
'1024*1024': (1024, 1024),
}
MAX_AREA_CONFIGS = {
'720*1280': 720 * 1280,
'1280*720': 1280 * 720,
'480*832': 480 * 832,
'832*480': 832 * 480,
}
SUPPORTED_SIZES = {
't2v-14B': ('720*1280', '1280*720', '480*832', '832*480'),
't2v-1.3B': ('480*832', '832*480'),
'i2v-14B': ('720*1280', '1280*720', '480*832', '832*480'),
't2i-14B': tuple(SIZE_CONFIGS.keys()),
}

View File

@@ -0,0 +1,19 @@
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import torch
from easydict import EasyDict
#------------------------ Wan shared config ------------------------#
wan_shared_cfg = EasyDict()
# t5
wan_shared_cfg.t5_model = 'umt5_xxl'
wan_shared_cfg.t5_dtype = torch.bfloat16
wan_shared_cfg.text_len = 512
# transformer
wan_shared_cfg.param_dtype = torch.bfloat16
# inference
wan_shared_cfg.num_train_timesteps = 1000
wan_shared_cfg.sample_fps = 16
wan_shared_cfg.sample_neg_prompt = '色调艳丽过曝静态细节模糊不清字幕风格作品画作画面静止整体发灰最差质量低质量JPEG压缩残留丑陋的残缺的多余的手指画得不好的手部画得不好的脸部畸形的毁容的形态畸形的肢体手指融合静止不动的画面杂乱的背景三条腿背景人很多倒着走'

View File

@@ -0,0 +1,35 @@
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import torch
from easydict import EasyDict
from .shared_config import wan_shared_cfg
#------------------------ Wan I2V 14B ------------------------#
i2v_14B = EasyDict(__name__='Config: Wan I2V 14B')
i2v_14B.update(wan_shared_cfg)
i2v_14B.t5_checkpoint = 'models_t5_umt5-xxl-enc-bf16.pth'
i2v_14B.t5_tokenizer = 'google/umt5-xxl'
# clip
i2v_14B.clip_model = 'clip_xlm_roberta_vit_h_14'
i2v_14B.clip_dtype = torch.float16
i2v_14B.clip_checkpoint = 'models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth'
i2v_14B.clip_tokenizer = 'xlm-roberta-large'
# vae
i2v_14B.vae_checkpoint = 'Wan2.1_VAE.pth'
i2v_14B.vae_stride = (4, 8, 8)
# transformer
i2v_14B.patch_size = (1, 2, 2)
i2v_14B.dim = 5120
i2v_14B.ffn_dim = 13824
i2v_14B.freq_dim = 256
i2v_14B.num_heads = 40
i2v_14B.num_layers = 40
i2v_14B.window_size = (-1, -1)
i2v_14B.qk_norm = True
i2v_14B.cross_attn_norm = True
i2v_14B.eps = 1e-6

View File

@@ -0,0 +1,29 @@
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
from easydict import EasyDict
from .shared_config import wan_shared_cfg
#------------------------ Wan T2V 14B ------------------------#
t2v_14B = EasyDict(__name__='Config: Wan T2V 14B')
t2v_14B.update(wan_shared_cfg)
# t5
t2v_14B.t5_checkpoint = 'models_t5_umt5-xxl-enc-bf16.pth'
t2v_14B.t5_tokenizer = 'google/umt5-xxl'
# vae
t2v_14B.vae_checkpoint = 'Wan2.1_VAE.pth'
t2v_14B.vae_stride = (4, 8, 8)
# transformer
t2v_14B.patch_size = (1, 2, 2)
t2v_14B.dim = 5120
t2v_14B.ffn_dim = 13824
t2v_14B.freq_dim = 256
t2v_14B.num_heads = 40
t2v_14B.num_layers = 40
t2v_14B.window_size = (-1, -1)
t2v_14B.qk_norm = True
t2v_14B.cross_attn_norm = True
t2v_14B.eps = 1e-6

View File

@@ -0,0 +1,29 @@
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
from easydict import EasyDict
from .shared_config import wan_shared_cfg
#------------------------ Wan T2V 1.3B ------------------------#
t2v_1_3B = EasyDict(__name__='Config: Wan T2V 1.3B')
t2v_1_3B.update(wan_shared_cfg)
# t5
t2v_1_3B.t5_checkpoint = 'models_t5_umt5-xxl-enc-bf16.pth'
t2v_1_3B.t5_tokenizer = 'google/umt5-xxl'
# vae
t2v_1_3B.vae_checkpoint = 'Wan2.1_VAE.pth'
t2v_1_3B.vae_stride = (4, 8, 8)
# transformer
t2v_1_3B.patch_size = (1, 2, 2)
t2v_1_3B.dim = 1536
t2v_1_3B.ffn_dim = 8960
t2v_1_3B.freq_dim = 256
t2v_1_3B.num_heads = 12
t2v_1_3B.num_layers = 30
t2v_1_3B.window_size = (-1, -1)
t2v_1_3B.qk_norm = True
t2v_1_3B.cross_attn_norm = True
t2v_1_3B.eps = 1e-6