jvilchesf/donut-base-finetuned-receipts

Donut Receipt Parser

This is a fine-tuned Donut model for receipt parsing.

Model Details

• Base Model: naver-clova-ix/donut-base
• Fine-tuned on: Receipt dataset
• Task: Parse receipts and extract structured information
• Training Date: 2025-10-13

Usage

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

# Load model and processor
processor = DonutProcessor.from_pretrained("jvilchesf/donut-base-finetuned-receipts")
model = VisionEncoderDecoderModel.from_pretrained("jvilchesf/donut-base-finetuned-receipts")

# Load and process image
image = Image.open("receipt.jpg").convert("RGB")
pixel_values = processor(image, return_tensors="pt").pixel_values

# Generate
task_prompt = "<parsing>"
decoder_input_ids = processor.tokenizer(
    task_prompt, 
    add_special_tokens=False, 
    return_tensors="pt"
).input_ids

outputs = model.generate(
    pixel_values,
    decoder_input_ids=decoder_input_ids,
    max_length=model.config.decoder.max_length,
    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,
)

# Decode
sequence = processor.batch_decode(outputs.sequences)[0]
sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
sequence = sequence.strip()
print(sequence)

Training Configuration

• Epochs: 30
• Batch Size: 2 (effective: 4 with gradient accumulation)
• Learning Rate: 2e-5
• LR Scheduler: Cosine with warmup
• Max Length: 1200 tokens
• Image Size: 720x960

Model Card

For more details, visit the model repository.