Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import gradio as gr
|
|
|
2 |
from janus.models import MultiModalityCausalLM, VLChatProcessor
|
3 |
from janus.utils.io import load_pil_images
|
4 |
import numpy as np
|
@@ -6,10 +7,6 @@ from PIL import Image
|
|
6 |
from transformers import AutoConfig, AutoModelForCausalLM
|
7 |
import torch
|
8 |
|
9 |
-
##
|
10 |
-
# Code from deepseek-ai/Janus
|
11 |
-
# Space from huggingface/twodgirl.
|
12 |
-
|
13 |
def generate(input_ids,
|
14 |
width,
|
15 |
height,
|
@@ -58,10 +55,10 @@ def unpack(dec, width, height, parallel_size=1):
|
|
58 |
return visual_img
|
59 |
|
60 |
@torch.inference_mode()
|
|
|
61 |
def generate_image(prompt,
|
62 |
width,
|
63 |
height,
|
64 |
-
# num_steps,
|
65 |
guidance,
|
66 |
seed):
|
67 |
if seed > -1:
|
@@ -117,7 +114,6 @@ if __name__ == '__main__':
|
|
117 |
model_path = 'deepseek-ai/Janus-1.3B'
|
118 |
processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
|
119 |
tokenizer = processor.tokenizer
|
120 |
-
# model: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)
|
121 |
config = AutoConfig.from_pretrained(model_path)
|
122 |
language_config = config.language_config
|
123 |
language_config._attn_implementation = 'eager'
|
|
|
1 |
import gradio as gr
|
2 |
+
import spaces # Import spaces for ZeroGPU compatibility
|
3 |
from janus.models import MultiModalityCausalLM, VLChatProcessor
|
4 |
from janus.utils.io import load_pil_images
|
5 |
import numpy as np
|
|
|
7 |
from transformers import AutoConfig, AutoModelForCausalLM
|
8 |
import torch
|
9 |
|
|
|
|
|
|
|
|
|
10 |
def generate(input_ids,
|
11 |
width,
|
12 |
height,
|
|
|
55 |
return visual_img
|
56 |
|
57 |
@torch.inference_mode()
|
58 |
+
@spaces.GPU # Decorate the function for ZeroGPU compatibility
|
59 |
def generate_image(prompt,
|
60 |
width,
|
61 |
height,
|
|
|
62 |
guidance,
|
63 |
seed):
|
64 |
if seed > -1:
|
|
|
114 |
model_path = 'deepseek-ai/Janus-1.3B'
|
115 |
processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
|
116 |
tokenizer = processor.tokenizer
|
|
|
117 |
config = AutoConfig.from_pretrained(model_path)
|
118 |
language_config = config.language_config
|
119 |
language_config._attn_implementation = 'eager'
|