amritzeon commited on
Commit
88a6cc0
1 Parent(s): 72fd318

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +57 -0
app.py ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import re
2
+ import gradio as gr
3
+
4
+ import torch
5
+ from transformers import DonutProcessor, VisionEncoderDecoderModel
6
+
7
+ processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
8
+ model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
9
+
10
+ device = "cuda" if torch.cuda.is_available() else "cpu"
11
+ model.to(device)
12
+
13
+ def process_document(image, question):
14
+ # prepare encoder inputs
15
+ pixel_values = processor(image, return_tensors="pt").pixel_values
16
+
17
+ # prepare decoder inputs
18
+ task_prompt = "<s_docvqa><s_question>{user_input}</s_question><s_answer>"
19
+ prompt = task_prompt.replace("{user_input}", question)
20
+ decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors="pt").input_ids
21
+
22
+ # generate answer
23
+ outputs = model.generate(
24
+ pixel_values.to(device),
25
+ decoder_input_ids=decoder_input_ids.to(device),
26
+ max_length=model.decoder.config.max_position_embeddings,
27
+ early_stopping=True,
28
+ pad_token_id=processor.tokenizer.pad_token_id,
29
+ eos_token_id=processor.tokenizer.eos_token_id,
30
+ use_cache=True,
31
+ num_beams=1,
32
+ bad_words_ids=[[processor.tokenizer.unk_token_id]],
33
+ return_dict_in_generate=True,
34
+ )
35
+
36
+ # postprocess
37
+ sequence = processor.batch_decode(outputs.sequences)[0]
38
+ sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
39
+ sequence = re.sub(r"<.*?>", "", sequence, count=1).strip() # remove first task start token
40
+
41
+ return processor.token2json(sequence)
42
+
43
+ description = "Gradio Demo for Donut, an instance of `VisionEncoderDecoderModel` fine-tuned on DocVQA (document visual question answering). To use it, simply upload your image and type a question and click 'submit', or click one of the examples to load them. Read more at the links below."
44
+ article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2111.15664' target='_blank'>Donut: OCR-free Document Understanding Transformer</a> | <a href='https://github.com/clovaai/donut' target='_blank'>Github Repo</a></p>"
45
+
46
+ demo = gr.Interface(
47
+ fn=process_document,
48
+ inputs=["image", "text"],
49
+ outputs="json",
50
+ title="Demo: DocumentAI for Entity Extraction with DONUT",
51
+ description=description,
52
+ article=article,
53
+ enable_queue=True,
54
+ #examples=[["example_1.png", "What is the invoice number?"], ["example_2.jpeg", "What's the population of Stoddard?"]],
55
+ cache_examples=False)
56
+
57
+ demo.launch()