metadata

license: apache-2.0
base_model: Helsinki-NLP/opus-mt-tc-big-en-pt
tags:
  - generated_from_trainer
datasets:
  - kde4
model-index:
  - name: opus-en-to-pt-translate
    results: []
language:
  - pt
  - en
pipeline_tag: translation

Model description

This model is a fine tuning for translations from English to Portuguese.

How to Use


prompt = f"""
Trump received a Bachelor of Science in economics from the University of Pennsylvania in 1968, and his father named him president of his real estate business in 1971.
Trump renamed it the Trump Organization and reoriented the company toward building and renovating skyscrapers, hotels, casinos, and golf courses.
After a series of business failures in the late twentieth century, he successfully launched side ventures that required little capital, mostly by licensing the Trump name.
From 2004 to 2015, he co-produced and hosted the reality television series The Apprentice.
"""

from transformers import pipeline
pipe = pipeline("translation", model="rhaymison/opus-en-to-pt-translator")
print(pipe())

#Trump recebeu um título de bacharel em economia pela Universidade da Pensilvânia em 1968, e o seu pai deu- lhe o nome de presidente do seu negócio imobiliário em 1971.
#Trump mudou o nome para Organização Trump e voltou a orientar a companhia para a construção e reforma de arranha- céus, hotéis, casinos e campos de golfe.
#Depois de uma série de falhas de negócio no fim do século XX, ele lançou com sucesso projectos paralelos que necessitaram de pouca capital, principalmente através do
#licenciamento do nome Trump. De 2004 a 2015, ele produziu em conjunto e alojou a série de reality série The Apprentice.


from transformers import MarianMTModel, MarianTokenizer

texts = [
    ">>por<< Trump received a Bachelor of Science in economics from the University of Pennsylvania in 1968, and his father named him president of his real estate business in 1971.",
    ">>por<< Trump renamed it the Trump Organization and reoriented the company toward building and renovating skyscrapers, hotels, casinos, and golf courses."
]

model = "rhaymison/opus-en-to-pt-translator"
tokenizer = MarianTokenizer.from_pretrained(model)
model = MarianMTModel.from_pretrained(model)
translated = model.generate(**tokenizer(texts, return_tensors="pt", padding=True))

for t in translated:
    print( tokenizer.decode(t, skip_special_tokens=True) )

# output:
# Trump recebeu um título de bacharel em economia pela Universidade da Pensilvânia em 1968, e o seu pai deu- lhe o nome de presidente do seu negócio imobiliário em 1971..
# Trump mudou o nome para Organização Trump e voltou a orientar a companhia para a construção e reforma de arranha- céus, hotéis, casinos e campos de golfe.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 32
eval_batch_size: 16
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

Training Loss	Epoch	Step	Validation Loss
0.4982	0.08	500	0.6398
0.5475	0.15	1000	0.6370
0.5397	0.23	1500	0.6333
0.5267	0.31	2000	0.6272
0.5212	0.39	2500	0.6240
0.522	0.46	3000	0.6179
0.5213	0.54	3500	0.6124
0.5155	0.62	4000	0.6114
0.5143	0.7	4500	0.6053
0.5037	0.77	5000	0.6058
0.5093	0.85	5500	0.6002
0.5253	0.93	6000	0.5945
0.5138	1.01	6500	0.5892
0.4864	1.08	7000	0.5906
0.491	1.16	7500	0.5889
0.4993	1.24	8000	0.5849
0.4749	1.32	8500	0.5849
0.4911	1.39	9000	0.5812
0.487	1.47	9500	0.5796
0.4846	1.55	10000	0.5758
0.4863	1.63	10500	0.5739
0.4792	1.7	11000	0.5725
0.4816	1.78	11500	0.5704
0.4811	1.86	12000	0.5684
0.4773	1.94	12500	0.5676
0.4657	2.01	13000	0.5691
0.4246	2.09	13500	0.5683
0.4285	2.17	14000	0.5693
0.4241	2.25	14500	0.5676
0.422	2.32	15000	0.5669
0.4199	2.4	15500	0.5656
0.4273	2.48	16000	0.5650
0.4161	2.56	16500	0.5651
0.4243	2.63	17000	0.5635
0.4202	2.71	17500	0.5628
0.4152	2.79	18000	0.5627
0.4179	2.87	18500	0.5619
0.4241	2.94	19000	0.5618

opus-en-to-pt-translate

This model is a fine-tuned version of Helsinki-NLP/opus-mt-tc-big-en-pt on the kde4 dataset. It achieves the following results on the evaluation set:

Loss: 0.5618

Framework versions

Transformers 4.38.1
Pytorch 2.1.0+cu121
Datasets 2.18.0
Tokenizers 0.15.2