transformers/docs/source/en/model_doc/donut.md

This model was published in HF papers on 2021-11-30 and contributed to Hugging Face Transformers on 2022-08-12.

Donut

Donut (Document Understanding Transformer) is a visual document understanding model that doesn't require an Optical Character Recognition (OCR) engine. Unlike traditional approaches that extract text using OCR before processing, Donut employs an end-to-end Transformer-based architecture to directly analyze document images. This eliminates OCR-related inefficiencies making it more accurate and adaptable to diverse languages and formats.

Donut features vision encoder (Swin) and a text decoder (BART). Swin converts document images into embeddings and BART processes them into meaningful text sequences.

You can find all the original Donut checkpoints under the Naver Clova Information Extraction organization.

Tip

Click on the Donut models in the right sidebar for more examples of how to apply Donut to different language and vision tasks.

The examples below demonstrate how to perform document understanding tasks using Donut with [Pipeline] and [AutoModel]

# pip install datasets
from transformers import pipeline


pipeline = pipeline(
    task="document-question-answering",
    model="naver-clova-ix/donut-base-finetuned-docvqa",
    device=0,
)
dataset = load_dataset("hf-internal-testing/example-documents", split="test")
image = dataset[0]["image"]

pipeline(image=image, question="What time is the coffee break?")

# pip install datasets
from datasets import load_dataset

from transformers import AutoModelForImageTextToText, AutoProcessor


processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
model = AutoModelForImageTextToText.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa", device_map="auto")

dataset = load_dataset("hf-internal-testing/example-documents", split="test")
image = dataset[0]["image"]
question = "What time is the coffee break?"
task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
inputs = processor(image, task_prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    input_ids=inputs.input_ids,
    pixel_values=inputs.pixel_values,
    max_length=512
)
answer = processor.decode(outputs[0], skip_special_tokens=True)
print(answer)

Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the Quantization overview for more available quantization backends.

The example below uses torchao to only quantize the weights to int4.

# pip install datasets torchao
from datasets import load_dataset

from transformers import AutoModelForImageTextToText, AutoProcessor, TorchAoConfig


quantization_config = TorchAoConfig("int4_weight_only", group_size=128)
processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
model = AutoModelForImageTextToText.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa", quantization_config=quantization_config, device_map="auto")

dataset = load_dataset("hf-internal-testing/example-documents", split="test")
image = dataset[0]["image"]
question = "What time is the coffee break?"
task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
inputs = processor(image, task_prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    input_ids=inputs.input_ids,
    pixel_values=inputs.pixel_values,
    max_length=512
)
answer = processor.decode(outputs[0], skip_special_tokens=True)
print(answer)

Notes

• Use Donut for document image classification as shown below.

  import re
  from transformers import DonutProcessor, VisionEncoderDecoderModel
      from datasets import load_dataset
  import torch
  
  processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
  model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip", device_map="auto")
  
  model.to(model.device)  # doctest: +IGNORE_RESULT
  
  # load document image
  dataset = load_dataset("hf-internal-testing/example-documents", split="test")
  image = dataset[1]["image"]
  
  # prepare decoder inputs
  task_prompt = "<s_rvlcdip>"
  decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").to(model.device).input_ids
  
  pixel_values = processor(image, return_tensors="pt").to(model.device).pixel_values
  
  outputs = model.generate(
      pixel_values.to(model.device),
      decoder_input_ids=decoder_input_ids.to(model.device),
      max_length=model.decoder.config.max_position_embeddings,
      pad_token_id=processor.tokenizer.pad_token_id,
      eos_token_id=processor.tokenizer.eos_token_id,
      use_cache=True,
      bad_words_ids=[[processor.tokenizer.unk_token_id]],
      return_dict_in_generate=True,
  )
  
  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
  print(processor.token2json(sequence))
  {'class': 'advertisement'}
  

• Use Donut for document parsing as shown below.

  import re
      from datasets import load_dataset
  from transformers import DonutProcessor, VisionEncoderDecoderModel
  import torch
  
  processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
  model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2", device_map="auto")
  
  model.to(model.device)  # doctest: +IGNORE_RESULT
  
  # load document image
  dataset = load_dataset("hf-internal-testing/example-documents", split="test")
  image = dataset[2]["image"]
  
  # prepare decoder inputs
  task_prompt = "<s_cord-v2>"
  decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").to(model.device).input_ids
  
  pixel_values = processor(image, return_tensors="pt").to(model.device).pixel_values
  
  outputs = model.generate(
      pixel_values.to(model.device),
      decoder_input_ids=decoder_input_ids.to(model.device),
      max_length=model.decoder.config.max_position_embeddings,
      pad_token_id=processor.tokenizer.pad_token_id,
      eos_token_id=processor.tokenizer.eos_token_id,
      use_cache=True,
      bad_words_ids=[[processor.tokenizer.unk_token_id]],
      return_dict_in_generate=True,
  )
  
  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
  print(processor.token2json(sequence))
  {'menu': {'nm': 'CINNAMON SUGAR', 'unitprice': '17,000', 'cnt': '1 x', 'price': '17,000'}, 'sub_total': {'subtotal_price': '17,000'}, 'total': 
  {'total_price': '17,000', 'cashprice': '20,000', 'changeprice': '3,000'}}
  

DonutSwinConfig

autodoc DonutSwinConfig

DonutImageProcessor

autodoc DonutImageProcessor - preprocess

DonutImageProcessorPil

autodoc DonutImageProcessorPil - preprocess

DonutProcessor

autodoc DonutProcessor - call - from_pretrained - save_pretrained - batch_decode - decode

DonutSwinModel

autodoc DonutSwinModel - forward

DonutSwinForImageClassification

autodoc transformers.DonutSwinForImageClassification - forward