Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import (
|
2 |
+
AutoModelForCausalLM,
|
3 |
+
AutoTokenizer,
|
4 |
+
__version__,
|
5 |
+
GenerationConfig,
|
6 |
+
)
|
7 |
+
from PIL import Image
|
8 |
+
import gradio as gr
|
9 |
+
import argparse
|
10 |
+
import tempfile
|
11 |
+
|
12 |
+
import os
|
13 |
+
from PIL import Image
|
14 |
+
import json
|
15 |
+
from tqdm import tqdm
|
16 |
+
import easyocr
|
17 |
+
|
18 |
+
assert (
|
19 |
+
__version__ == "4.32.0"
|
20 |
+
), "Please use transformers version 4.32.0, pip install transformers==4.32.0"
|
21 |
+
|
22 |
+
reader = easyocr.Reader(
|
23 |
+
["en"]
|
24 |
+
) # this needs to run only once to load the model into memory
|
25 |
+
|
26 |
+
|
27 |
+
def get_easy_text(img_file):
|
28 |
+
out = reader.readtext(img_file, detail=0, paragraph=True)
|
29 |
+
if isinstance(out, list):
|
30 |
+
return "\n".join(out)
|
31 |
+
return out
|
32 |
+
|
33 |
+
model_name = "DigitalAgent/Captioner"
|
34 |
+
model = (
|
35 |
+
AutoModelForCausalLM.from_pretrained(
|
36 |
+
model_name, device_map="cuda", trust_remote_code=True
|
37 |
+
)
|
38 |
+
.eval()
|
39 |
+
.half()
|
40 |
+
)
|
41 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
42 |
+
generation_config = GenerationConfig.from_dict(
|
43 |
+
{
|
44 |
+
"chat_format": "chatml",
|
45 |
+
"do_sample": True,
|
46 |
+
"eos_token_id": 151643,
|
47 |
+
"max_new_tokens": 2048,
|
48 |
+
"max_window_size": 6144,
|
49 |
+
"pad_token_id": 151643,
|
50 |
+
"repetition_penalty": 1.2,
|
51 |
+
"top_k": 0,
|
52 |
+
"top_p": 0.3,
|
53 |
+
"transformers_version": "4.31.0",
|
54 |
+
}
|
55 |
+
)
|
56 |
+
|
57 |
+
|
58 |
+
def generate(image: Image):
|
59 |
+
with tempfile.NamedTemporaryFile(suffix=".jpg", delete=True) as tmp:
|
60 |
+
image.save(tmp.name)
|
61 |
+
ocr_result = get_easy_text(tmp.name)
|
62 |
+
text = f"Please describe the screenshot above in details.\nOCR Result:\n{ocr_result}"
|
63 |
+
history = []
|
64 |
+
input_data = [{"image": tmp.name}, {"text": text}]
|
65 |
+
query = tokenizer.from_list_format(input_data)
|
66 |
+
response, _ = model.chat(
|
67 |
+
tokenizer, query=query, history=history, generation_config=generation_config
|
68 |
+
)
|
69 |
+
return response
|
70 |
+
|
71 |
+
|
72 |
+
def main(port, share):
|
73 |
+
demo = gr.Interface(
|
74 |
+
fn=generate, inputs=[gr.Image(type="pil")], outputs="text", concurrency_limit=1
|
75 |
+
)
|
76 |
+
demo.queue().launch(server_port=port, share=share)
|
77 |
+
|
78 |
+
|
79 |
+
if __name__ == "__main__":
|
80 |
+
parser = argparse.ArgumentParser()
|
81 |
+
parser.add_argument("--port", type=int)
|
82 |
+
parser.add_argument("--share", action="store_true", default=False)
|
83 |
+
args = parser.parse_args()
|
84 |
+
main(args.port, args.share)
|