mirror of
https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
synced 2024-12-27 07:39:53 +08:00
33 lines
1.4 KiB
Python
33 lines
1.4 KiB
Python
import torch
|
|
|
|
from modules import sd_hijack_clip, devices
|
|
|
|
|
|
class FrozenXLMREmbedderWithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWithCustomWords):
|
|
def __init__(self, wrapped, hijack):
|
|
super().__init__(wrapped, hijack)
|
|
|
|
self.id_start = wrapped.config.bos_token_id
|
|
self.id_end = wrapped.config.eos_token_id
|
|
self.id_pad = wrapped.config.pad_token_id
|
|
|
|
self.comma_token = self.tokenizer.get_vocab().get(',', None) # alt diffusion doesn't have </w> bits for comma
|
|
|
|
def encode_with_transformers(self, tokens):
|
|
# there's no CLIP Skip here because all hidden layers have size of 1024 and the last one uses a
|
|
# trained layer to transform those 1024 into 768 for unet; so you can't choose which transformer
|
|
# layer to work with - you have to use the last
|
|
|
|
attention_mask = (tokens != self.id_pad).to(device=tokens.device, dtype=torch.int64)
|
|
features = self.wrapped(input_ids=tokens, attention_mask=attention_mask)
|
|
z = features['projection_state']
|
|
|
|
return z
|
|
|
|
def encode_embedding_init_text(self, init_text, nvpt):
|
|
embedding_layer = self.wrapped.roberta.embeddings
|
|
ids = self.wrapped.tokenizer(init_text, max_length=nvpt, return_tensors="pt", add_special_tokens=False)["input_ids"]
|
|
embedded = embedding_layer.token_embedding.wrapped(ids.to(devices.device)).squeeze(0)
|
|
|
|
return embedded
|