Edit model card

flan-t5-portuguese-small-summarization

This model aims to help supply the needs of models in the Portuguese language for certain tasks. The model presents a good performance for summary tasks. Some errors due to word accentuation may occasionally occur due to the small version of the model.

model_max_length = 512

  • Loss: 1.6541
  • Rouge1: 16.3352
  • Rouge2: 6.2366
  • Rougel: 14.1335
  • Rougelsum: 15.2755
  • Gen Len: 19.0

GPU: RTX 3060, 12GB, =~3500 cuda cores

!pip install transformers
from transformers import pipeline
summarization = pipeline("summarization", model="rhaymison/flan-t5-portuguese-small-summarization", tokenizer="rhaymison/flan-t5-portuguese-small-summarization")

prompt =f"""
sumarize: No que consiste o transtorno dismórfico corporal? São pessoas que se acham feias e querem mudar sua aparência de forma obsessiva, mesmo que não tenham nenhum problema. Num dos estudos que fiz, detectamos que de 50% a 54% dos pacientes que procuram cirurgia de face, nariz ou abdômen apresentam essa condição. A cirurgia pode beneficiar aqueles com um quadro leve ou intermediário do transtorno. No entanto, os que apresentam um transtorno mais grave não devem ser operados, e sim encaminhados para tratamento psicológico. A maior dificuldade é que aceitem ajuda. Muitos preferem buscar um médico que dê sinal verde para a intervenção.
"""
output =  summarization(prompt)

#Transtorno dismórfico corporal: o que apresenta o transtorno no deve ser operados, e sim encaminhados para tratamento psicológico. 
#A cirurgia pode beneficiar aqueles com um quadro leve ou intermediário do transtornamento, nariz ou abdômen.

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: 5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 5

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum Gen Len
1.847 0.27 500 1.7443 15.4969 5.9408 13.5074 14.5518 19.0
1.8333 0.53 1000 1.7194 15.6496 5.8641 13.5584 14.669 19.0
1.8043 0.8 1500 1.7209 15.8523 6.0544 13.7563 14.8941 19.0
1.7903 1.07 2000 1.7156 15.8969 6.0071 13.7534 14.8513 19.0
1.7862 1.33 2500 1.7007 15.8441 5.958 13.66 14.7226 19.0
1.7687 1.6 3000 1.6949 15.9134 6.0486 13.9238 14.9171 19.0
1.7724 1.87 3500 1.6909 15.8827 5.8941 13.7195 14.8736 19.0
1.7653 2.13 4000 1.6811 16.0819 5.9791 13.8639 15.0031 19.0
1.7392 2.4 4500 1.6761 15.706 5.7384 13.5978 14.7374 19.0
1.7578 2.67 5000 1.6729 15.8926 5.9629 13.767 14.9088 19.0
1.7353 2.93 5500 1.6675 16.0266 5.9024 13.8471 14.9721 19.0
1.7425 3.2 6000 1.6626 16.0732 6.1141 13.9016 15.0673 19.0
1.73 3.47 6500 1.6631 16.1333 6.0951 13.9551 15.0686 19.0
1.7355 3.73 7000 1.6616 16.1704 6.1575 14.0481 15.079 19.0
1.7139 4.0 7500 1.6572 16.2592 6.25 14.0403 15.1851 19.0
1.7188 4.27 8000 1.6580 16.1572 6.0661 14.0029 15.0935 19.0
1.7045 4.53 8500 1.6560 16.1409 6.1478 13.9806 15.0795 19.0
1.7201 4.8 9000 1.6541 16.3352 6.2366 14.1335 15.2755 19.0

Framework versions

  • Transformers 4.38.2
  • Pytorch 2.2.1+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.2

Comments

Any idea, help or report will always be welcome.

email: rhaymisoncristian@gmail.com

Downloads last month
246
Safetensors
Model size
77M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for rhaymison/flan-t5-portuguese-small-summarization

Finetuned
(298)
this model

Dataset used to train rhaymison/flan-t5-portuguese-small-summarization