michaeljcliao/invoice_extraction_20240809_base_non_0_aug4x_retrain

donut-base-invoice-resize_splitbydate_roc_240401

This model is a fine-tuned version of naver-clova-ix/donut-base on the imagefolder dataset.

Model description

Trained from Donut base model (naver-clova-ix/donut-base) with non-type-0 invoice data with original Gregorian date instead of ROC date with Chinese characters

Intended uses & limitations

More information needed

Training and evaluation data

Train data: 1124 samples of non type 0 images with invoice date between 2024/07/01 and 2024/07/15 (original 281 samples + 3 times augmented data)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10 (checkpoint-5620)

Training results

TrainOutput(global_step=2420, training_loss=1.0457300442309418, metrics={'train_runtime': 9844.8278, 'train_samples_per_second': 0.49, 'train_steps_per_second': 0.246, 'total_flos': 6.47612717723136e+18, 'train_loss': 1.0457300442309418, 'epoch': 20.0})

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.14.5
• Tokenizers 0.15.2