Model Description

This model, named traintogpb-ke-t5-base-aihub-koen-translation-integrated-10m-en-to-ko, is a machine translation model that translates English to Korean. It is fine-tuned from the KETI-AIR/ke-t5-base model using the aihub-koen-translation-integrated-base-10m dataset.

Model Architecture

The model uses the ke-t5-base architecture, which is based on the T5 (Text-to-Text Transfer Transformer) model.

Training Data

The model was trained on the aihub-koen-translation-integrated-base-10m dataset, which is designed for English-to-Korean translation tasks.

Training Procedure

Training Parameters

The model was trained with the following parameters:

• Learning Rate: 0.0005
• Weight Decay: 0.01
• Batch Size: 64 (training), 128 (evaluation)
• Number of Epochs: 2
• Save Steps: 500
• Max Save Checkpoints: 2
• Evaluation Strategy: At the end of each epoch
• Logging Strategy: No logging
• Use of FP16: No
• Gradient Accumulation Steps: 2
• Reporting: None

Hardware

The training was performed on a single GPU system with an NVIDIA A100 (40GB).

Performance

The model achieved the following BLEU scores during training:

• Epoch 1: 18.006119
• Epoch 2: 18.838066

Usage

This model is suitable for applications involving translation from English to Korean. Here is an example on how to use this model in Hugging Face's Transformers:

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model = AutoModelForSeq2SeqLM.from_pretrained("traintogpb-ke-t5-base-aihub-koen-translation-integrated-10m-en-to-ko")
tokenizer = AutoTokenizer.from_pretrained("traintogpb-ke-t5-base-aihub-koen-translation-integrated-10m-en-to-ko")

inputs = tokenizer.encode("This is a sample text.", return_tensors="pt")
outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))