298977887/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-04-19 19:03:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Limit T5 to 128 tokens for HiDream: #7620

Browse Source
This commit is contained in:
comfyanonymous 2025-04-16 18:07:55 -04:00
parent f00f340a56
commit 9899d187b1
2 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
comfy/text_encoders/hidream.py
View File

@ -11,14 +11,15 @@ class HiDreamTokenizer:
def __init__(self, embedding_directory=None, tokenizer_data={}):
self.clip_l = sd1_clip.SDTokenizer(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data)
self.clip_g = sdxl_clip.SDXLClipGTokenizer(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data)
self.t5xxl = sd3_clip.T5XXLTokenizer(embedding_directory=embedding_directory, min_length=128, tokenizer_data=tokenizer_data)
self.t5xxl = sd3_clip.T5XXLTokenizer(embedding_directory=embedding_directory, min_length=128, max_length=128, tokenizer_data=tokenizer_data)
self.llama = hunyuan_video.LLAMA3Tokenizer(embedding_directory=embedding_directory, min_length=128, pad_token=128009, tokenizer_data=tokenizer_data)
def tokenize_with_weights(self, text:str, return_word_ids=False, **kwargs):
out = {}
out["g"] = self.clip_g.tokenize_with_weights(text, return_word_ids)
out["l"] = self.clip_l.tokenize_with_weights(text, return_word_ids)
out["t5xxl"] = self.t5xxl.tokenize_with_weights(text, return_word_ids)
t5xxl = self.t5xxl.tokenize_with_weights(text, return_word_ids)
out["t5xxl"] = [t5xxl[0]] # Use only first 128 tokens
out["llama"] = self.llama.tokenize_with_weights(text, return_word_ids)
return out

4
comfy/text_encoders/sd3_clip.py
View File

@ -32,9 +32,9 @@ def t5_xxl_detect(state_dict, prefix=""):
return out
class T5XXLTokenizer(sd1_clip.SDTokenizer):
def __init__(self, embedding_directory=None, tokenizer_data={}, min_length=77):
def __init__(self, embedding_directory=None, tokenizer_data={}, min_length=77, max_length=99999999):
tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "t5_tokenizer")
super().__init__(tokenizer_path, embedding_directory=embedding_directory, pad_with_end=False, embedding_size=4096, embedding_key='t5xxl', tokenizer_class=T5TokenizerFast, has_start_token=False, pad_to_max_length=False, max_length=99999999, min_length=min_length, tokenizer_data=tokenizer_data)
super().__init__(tokenizer_path, embedding_directory=embedding_directory, pad_with_end=False, embedding_size=4096, embedding_key='t5xxl', tokenizer_class=T5TokenizerFast, has_start_token=False, pad_to_max_length=False, max_length=max_length, min_length=min_length, tokenizer_data=tokenizer_data)
class SD3Tokenizer: