Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,50 +1,60 @@
|
|
1 |
-
import
|
2 |
-
import
|
3 |
-
from torchvision import models, transforms
|
4 |
-
from PIL import Image
|
5 |
-
import requests
|
6 |
-
from io import BytesIO
|
7 |
|
8 |
-
# Load
|
9 |
-
model =
|
10 |
-
|
|
|
11 |
|
12 |
-
#
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
|
20 |
-
#
|
21 |
-
|
22 |
-
class_idx = requests.get(LABELS_URL).json()
|
23 |
-
idx2label = [class_idx[str(k)][1] for k in range(len(class_idx))]
|
24 |
|
25 |
-
#
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
# Perform inference
|
31 |
-
with torch.no_grad():
|
32 |
-
output = model(image)
|
33 |
-
|
34 |
-
# Get the predicted label
|
35 |
-
_, predicted_class = torch.max(output, 1)
|
36 |
-
label = idx2label[predicted_class.item()]
|
37 |
-
return label
|
38 |
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
|
|
|
|
49 |
|
50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
|
2 |
+
from qwen_vl_utils import process_vision_info
|
|
|
|
|
|
|
|
|
3 |
|
4 |
+
# default: Load the model on the available device(s)
|
5 |
+
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
6 |
+
"Qwen/Qwen2-VL-2B-Instruct", torch_dtype="auto", device_map="auto"
|
7 |
+
)
|
8 |
|
9 |
+
# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
|
10 |
+
# model = Qwen2VLForConditionalGeneration.from_pretrained(
|
11 |
+
# "Qwen/Qwen2-VL-2B-Instruct",
|
12 |
+
# torch_dtype=torch.bfloat16,
|
13 |
+
# attn_implementation="flash_attention_2",
|
14 |
+
# device_map="auto",
|
15 |
+
# )
|
16 |
|
17 |
+
# default processer
|
18 |
+
processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
|
|
|
|
|
19 |
|
20 |
+
# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
|
21 |
+
# min_pixels = 256*28*28
|
22 |
+
# max_pixels = 1280*28*28
|
23 |
+
# processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
+
messages = [
|
26 |
+
{
|
27 |
+
"role": "user",
|
28 |
+
"content": [
|
29 |
+
{
|
30 |
+
"type": "image",
|
31 |
+
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
|
32 |
+
},
|
33 |
+
{"type": "text", "text": "Describe this image."},
|
34 |
+
],
|
35 |
+
}
|
36 |
+
]
|
37 |
|
38 |
+
# Preparation for inference
|
39 |
+
text = processor.apply_chat_template(
|
40 |
+
messages, tokenize=False, add_generation_prompt=True
|
41 |
+
)
|
42 |
+
image_inputs, video_inputs = process_vision_info(messages)
|
43 |
+
inputs = processor(
|
44 |
+
text=[text],
|
45 |
+
images=image_inputs,
|
46 |
+
videos=video_inputs,
|
47 |
+
padding=True,
|
48 |
+
return_tensors="pt",
|
49 |
+
)
|
50 |
+
inputs = inputs.to("cuda")
|
51 |
+
|
52 |
+
# Inference: Generation of the output
|
53 |
+
generated_ids = model.generate(**inputs, max_new_tokens=128)
|
54 |
+
generated_ids_trimmed = [
|
55 |
+
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
56 |
+
]
|
57 |
+
output_text = processor.batch_decode(
|
58 |
+
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
59 |
+
)
|
60 |
+
print(output_text)
|