Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -7,7 +7,8 @@ import torchvision.transforms as T
|
|
7 |
from transformers import AutoTokenizer
|
8 |
import gradio as gr
|
9 |
from resnet50 import build_model
|
10 |
-
from utils import generate_similiarity_map, post_process, load_tokenizer, build_transform_R50
|
|
|
11 |
from utils import IMAGENET_MEAN, IMAGENET_STD
|
12 |
from internvl.train.dataset import dynamic_preprocess
|
13 |
from internvl.model.internvl_chat import InternVLChatModel
|
@@ -42,20 +43,16 @@ def load_model(check_type):
|
|
42 |
elif 'TokenFD' in check_type:
|
43 |
model_path = CHECKPOINTS[check_type]
|
44 |
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, use_fast=False, use_auth_token=HF_TOKEN)
|
45 |
-
model = InternVLChatModel.from_pretrained(model_path, torch_dtype=torch.bfloat16).eval()
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
T.ToTensor(),
|
50 |
-
T.Normalize(IMAGENET_MEAN, IMAGENET_STD)
|
51 |
-
])
|
52 |
-
|
53 |
return model.to(device), tokenizer, transform, device
|
54 |
|
55 |
def process_image(model, tokenizer, transform, device, check_type, image, text):
|
56 |
|
57 |
src_size = image.size
|
58 |
-
if '
|
59 |
images, target_ratio = dynamic_preprocess(image, min_num=1, max_num=12,
|
60 |
image_size=model.config.force_image_size,
|
61 |
use_thumbnail=model.config.use_thumbnail,
|
|
|
7 |
from transformers import AutoTokenizer
|
8 |
import gradio as gr
|
9 |
from resnet50 import build_model
|
10 |
+
# from utils import generate_similiarity_map, post_process, load_tokenizer, build_transform_R50
|
11 |
+
from utils import generate_similiarity_map, get_transform, post_process, load_tokenizer, build_transform_R50
|
12 |
from utils import IMAGENET_MEAN, IMAGENET_STD
|
13 |
from internvl.train.dataset import dynamic_preprocess
|
14 |
from internvl.model.internvl_chat import InternVLChatModel
|
|
|
43 |
elif 'TokenFD' in check_type:
|
44 |
model_path = CHECKPOINTS[check_type]
|
45 |
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, use_fast=False, use_auth_token=HF_TOKEN)
|
46 |
+
# model = InternVLChatModel.from_pretrained(model_path, torch_dtype=torch.bfloat16).eval()
|
47 |
+
model = InternVLChatModel.from_pretrained(checkpoint_vit_english, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16 ,load_in_8bit=False, load_in_4bit=False).eval()
|
48 |
+
transform = get_transform(is_train=False, image_size=model.config.force_image_size)
|
49 |
+
|
|
|
|
|
|
|
|
|
50 |
return model.to(device), tokenizer, transform, device
|
51 |
|
52 |
def process_image(model, tokenizer, transform, device, check_type, image, text):
|
53 |
|
54 |
src_size = image.size
|
55 |
+
if 'TokenFD' in check_type:
|
56 |
images, target_ratio = dynamic_preprocess(image, min_num=1, max_num=12,
|
57 |
image_size=model.config.force_image_size,
|
58 |
use_thumbnail=model.config.use_thumbnail,
|