theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-24 18:28:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #374 from modelscope/wan-tokenizer-bugfix

Browse Source 
align wan tokenizer to official
This commit is contained in:
Zhongjie Duan
2025-02-28 16:05:36 +08:00
committed by GitHub
parent a48822ec00 61a30673c2
commit f9f49e3c78
3 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
diffsynth/models/wan_video_vae.py
View File

@@ -785,6 +785,7 @@ class WanVideoVAE(nn.Module):
video = self.single_decode(hidden_state, device) video = self.single_decode(hidden_state, device)
video = video.squeeze(0) video = video.squeeze(0)
videos.append(video) videos.append(video)
videos = torch.stack(videos)
return videos return videos

5
diffsynth/prompters/wan_prompter.py
View File

@@ -2,20 +2,24 @@ from .base_prompter import BasePrompter
from ..models.wan_video_text_encoder import WanTextEncoder from ..models.wan_video_text_encoder import WanTextEncoder
from transformers import AutoTokenizer from transformers import AutoTokenizer
import os, torch import os, torch
import ftfy
import html import html
import string import string
import regex as re import regex as re
def basic_clean(text): def basic_clean(text):
text = ftfy.fix_text(text)
text = html.unescape(html.unescape(text)) text = html.unescape(html.unescape(text))
return text.strip() return text.strip()
def whitespace_clean(text): def whitespace_clean(text):
text = re.sub(r'\s+', ' ', text) text = re.sub(r'\s+', ' ', text)
text = text.strip() text = text.strip()
return text return text
def canonicalize(text, keep_punctuation_exact_string=None): def canonicalize(text, keep_punctuation_exact_string=None):
text = text.replace('_', ' ') text = text.replace('_', ' ')
if keep_punctuation_exact_string: if keep_punctuation_exact_string:
@@ -28,6 +32,7 @@ def canonicalize(text, keep_punctuation_exact_string=None):
text = re.sub(r'\s+', ' ', text) text = re.sub(r'\s+', ' ', text)
return text.strip() return text.strip()
class HuggingfaceTokenizer: class HuggingfaceTokenizer:
def __init__(self, name, seq_len=None, clean=None, **kwargs): def __init__(self, name, seq_len=None, clean=None, **kwargs):

1
requirements.txt
View File

@@ -10,3 +10,4 @@ einops
sentencepiece sentencepiece
protobuf protobuf
modelscope modelscope
ftfy