Instructions to use hf-tuner/donut-base-finetuned-docvqa with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hf-tuner/donut-base-finetuned-docvqa with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="hf-tuner/donut-base-finetuned-docvqa")

# Load model directly
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("hf-tuner/donut-base-finetuned-docvqa")
model = AutoModel.from_pretrained("hf-tuner/donut-base-finetuned-docvqa")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use hf-tuner/donut-base-finetuned-docvqa with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hf-tuner/donut-base-finetuned-docvqa"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hf-tuner/donut-base-finetuned-docvqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/hf-tuner/donut-base-finetuned-docvqa

SGLang

How to use hf-tuner/donut-base-finetuned-docvqa with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "hf-tuner/donut-base-finetuned-docvqa" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hf-tuner/donut-base-finetuned-docvqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "hf-tuner/donut-base-finetuned-docvqa" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hf-tuner/donut-base-finetuned-docvqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use hf-tuner/donut-base-finetuned-docvqa with Docker Model Runner:
```
docker model run hf.co/hf-tuner/donut-base-finetuned-docvqa
```

donut-base-finetuned-docvqa

This model is a fine-tuned version of naver-clova-ix/donut-base on hf-tuner/docvqa-10k-donut dataset.

Model description

Donut consists of a vision encoder (Swin Transformer) and a text decoder (BART). Given an image, the encoder first encodes the image into a tensor of embeddings (of shape batch_size, seq_len, hidden_size), after which the decoder autoregressively generates text, conditioned on the encoding of the encoder.

How to use

import torch
from PIL import Image
from transformers import DonutProcessor, VisionEncoderDecoderConfig, VisionEncoderDecoderModel

model_id = "hf-tuner/donut-base-finetuned-docvqa"

config = VisionEncoderDecoderConfig.from_pretrained(model_id)
processor = DonutProcessor.from_pretrained(model_id)
device = "cuda" if torch.cuda.is_available() else "cpu"
config.dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model = VisionEncoderDecoderModel.from_pretrained(ckpt, config=config)
model.to(device)

PROMPT_TEMPLATE = "<doc_vqa><s><s_question>{}</s_question><s_answer>"

def predict(image, question):
  prompt = PROMPT_TEMPLATE.format(question)

  pixel_values = processor(image,
                           return_tensors="pt"
                           ).pixel_values.to(device)
  decoder_input_ids = processor.tokenizer(prompt,
                                          add_special_tokens=False,
                                          return_tensors="pt"
                                          ).input_ids.to(device)

  generated_ids = model.generate(pixel_values,
                                decoder_input_ids=decoder_input_ids,
                                max_length=64,
                                bad_words_ids=[[processor.tokenizer.unk_token_id]]
                                )
  generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
  output_json = processor.token2json(generated_text)
  return output_json

output = predict(image=Image.open('doc-image.png'), question="Which is the date of the approval form?")
# {"question": "Which is the date of the approval form?", "answer": "april 5, 1995"}

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 1
eval_batch_size: 8
seed: 42
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
num_epochs: 1
mixed_precision_training: Native AMP