Definitely! As long as you have an (image, text) pair dataset, you can fine-tune Donut to generate whatever you like.

However if you're planning to fine-tune on a different language than English/Chinese/Korean/Japanese, it makes sense to first pre-train the model on that language (since donut-base is only pre-trained on English, Chinese, Korean, Japanese). The authors create synthetic data for pre-training, as explained here: https://github.com/clovaai/donut#synthdog-datasets

I am using "naver-clova-ix/donut-base-finetuned-docvqa" model and want to print the full content of the result json after it reads the image without invoking any prompts or user input. I just want it to parse the image and give me the full json content. How can I achieve that, please help. I am using below code:

import re
import gradio as gr

import torch
from transformers import DonutProcessor, VisionEncoderDecoderModel

processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

def process_document(image, question):

prepare encoder inputs

pixel_values = processor(image, return_tensors="pt").pixel_values
print(pixel_values)

prepare decoder inputs

task_prompt = "{user_input}"
prompt = task_prompt.replace("{user_input}", question)
decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors="pt").input_ids
print(decoder_input_ids)

generate answer

outputs = model.generate(
pixel_values.to(device),
decoder_input_ids=decoder_input_ids.to(device),
max_length=model.decoder.config.max_position_embeddings,
early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True,
num_beams=1,
bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True,
)

postprocess

sequence = processor.batch_decode(outputs.sequences)[0]
sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
sequence = re.sub(r"<.*?>", "", sequence, count=1).strip() # remove first task start token

json_content = processor.token2json(sequence)
print(json_content) # Print the full JSON content

return json_content
#return processor.token2json(sequence)
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."
article = "

Donut: OCR-free Document Understanding Transformer | Github Repo

"
demo = gr.Interface(
fn=process_document,
inputs=["image", "text"],
outputs="json",
title="Demo: Donut 🍩 for DocVQA",
description=description,
article=article,
enable_queue=True,
examples=[["example_1.png", "When is the coffee break?"], ["example_2.jpeg", "What's the population of Stoddard?"]],
cache_examples=False)

demo.launch()