fine-tuned_ar-en / README.md
Patt's picture
update model card README.md
c04b5bc
---
license: apache-2.0
tags:
- translation
- generated_from_trainer
datasets:
- tatoeba_mt
metrics:
- bleu
model-index:
- name: fine-tuned_ar-en
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: tatoeba_mt
type: tatoeba_mt
config: ara-eng
split: validation
args: ara-eng
metrics:
- name: Bleu
type: bleu
value: 51.81577100911173
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# fine-tuned_ar-en
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-ar-en](https://huggingface.co/Helsinki-NLP/opus-mt-ar-en) on the tatoeba_mt dataset.
It achieves the following results on the evaluation set:
- Loss: 0.8464
- Bleu: 51.8158
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2