Spaces:
Running
on
Zero
Running
on
Zero
update
Browse files- inference.py +2 -3
inference.py
CHANGED
@@ -14,7 +14,6 @@ import sys
|
|
14 |
import gc
|
15 |
from transformers import CLIPTextModel, CLIPTokenizer, BertModel, BertTokenizer
|
16 |
|
17 |
-
from hf_demo import dtype
|
18 |
# import train_util
|
19 |
|
20 |
from utils.train_util import get_noisy_image, encode_prompts
|
@@ -320,8 +319,8 @@ def inference(network: LoRANetwork, tokenizer: CLIPTokenizer, text_encoder: CLIP
|
|
320 |
uncond_embeddings = uncond_embed.repeat(bcz, 1, 1)
|
321 |
else:
|
322 |
uncond_embeddings = uncond_embed
|
323 |
-
style_text_embeddings = torch.cat([uncond_embeddings, style_embeddings]
|
324 |
-
original_embeddings = torch.cat([uncond_embeddings, original_embeddings]
|
325 |
|
326 |
generator = torch.manual_seed(single_seed) if single_seed is not None else None
|
327 |
noise_scheduler.set_timesteps(steps)
|
|
|
14 |
import gc
|
15 |
from transformers import CLIPTextModel, CLIPTokenizer, BertModel, BertTokenizer
|
16 |
|
|
|
17 |
# import train_util
|
18 |
|
19 |
from utils.train_util import get_noisy_image, encode_prompts
|
|
|
319 |
uncond_embeddings = uncond_embed.repeat(bcz, 1, 1)
|
320 |
else:
|
321 |
uncond_embeddings = uncond_embed
|
322 |
+
style_text_embeddings = torch.cat([uncond_embeddings, style_embeddings]).to(weight_dtype)
|
323 |
+
# original_embeddings = torch.cat([uncond_embeddings, original_embeddings]).to(weight_dtype)
|
324 |
|
325 |
generator = torch.manual_seed(single_seed) if single_seed is not None else None
|
326 |
noise_scheduler.set_timesteps(steps)
|