Spaces:
Running
on
Zero
Running
on
Zero
Linoy Tsaban
commited on
Commit
·
1b62550
1
Parent(s):
502ed04
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,25 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
-
from
|
|
|
|
| 4 |
from utils import video_to_frames, add_dict_to_yaml_file, save_video, seed_everything
|
| 5 |
# from diffusers.utils import export_to_video
|
| 6 |
from tokenflow_pnp import TokenFlow
|
| 7 |
from preprocess_utils import *
|
| 8 |
from tokenflow_utils import *
|
|
|
|
| 9 |
# load sd model
|
| 10 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
def randomize_seed_fn():
|
| 16 |
seed = random.randint(0, np.iinfo(np.int32).max)
|
|
@@ -65,7 +74,12 @@ def prep(config):
|
|
| 65 |
else:
|
| 66 |
save_path = None
|
| 67 |
|
| 68 |
-
model = Preprocess(device, config
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
print(type(model.config["batch_size"]))
|
| 70 |
frames, latents, total_inverted_latents, rgb_reconstruction = model.extract_latents(
|
| 71 |
num_steps=model.config["steps"],
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
+
from transformers import CLIPTextModel, CLIPTokenizer, logging
|
| 4 |
+
from diffusers import AutoencoderKL, UNet2DConditionModel, DDIMScheduler
|
| 5 |
from utils import video_to_frames, add_dict_to_yaml_file, save_video, seed_everything
|
| 6 |
# from diffusers.utils import export_to_video
|
| 7 |
from tokenflow_pnp import TokenFlow
|
| 8 |
from preprocess_utils import *
|
| 9 |
from tokenflow_utils import *
|
| 10 |
+
|
| 11 |
# load sd model
|
| 12 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 13 |
+
model_id = "stabilityai/stable-diffusion-2-1-base"
|
| 14 |
+
|
| 15 |
+
scheduler = DDIMScheduler.from_pretrained(model_id, subfolder="scheduler")
|
| 16 |
+
vae = AutoencoderKL.from_pretrained(model_id, subfolder="vae", revision="fp16",
|
| 17 |
+
torch_dtype=torch.float16).to(device)
|
| 18 |
+
tokenizer = CLIPTokenizer.from_pretrained(model_id, subfolder="tokenizer")
|
| 19 |
+
text_encoder = CLIPTextModel.from_pretrained(model_id, subfolder="text_encoder", revision="fp16",
|
| 20 |
+
torch_dtype=torch.float16).to(device)
|
| 21 |
+
unet = UNet2DConditionModel.from_pretrained(model_id, subfolder="unet", revision="fp16",
|
| 22 |
+
torch_dtype=torch.float16).to(device)
|
| 23 |
|
| 24 |
def randomize_seed_fn():
|
| 25 |
seed = random.randint(0, np.iinfo(np.int32).max)
|
|
|
|
| 74 |
else:
|
| 75 |
save_path = None
|
| 76 |
|
| 77 |
+
model = Preprocess(device, config,
|
| 78 |
+
vae=vae,
|
| 79 |
+
text_encoder=text_encoder,
|
| 80 |
+
scheduler=scheduler,
|
| 81 |
+
tokenizer=tokenizer,
|
| 82 |
+
unet=unet)
|
| 83 |
print(type(model.config["batch_size"]))
|
| 84 |
frames, latents, total_inverted_latents, rgb_reconstruction = model.extract_latents(
|
| 85 |
num_steps=model.config["steps"],
|