w11wo/lao-roberta-base

Lao RoBERTa Base

Lao RoBERTa Base is a masked language model based on the RoBERTa model. It was trained on the OSCAR-2109 dataset, specifically the deduplicated_lo subset. The model was trained from scratch and achieved an evaluation loss of 1.4556 and an evaluation perplexity of 4.287.

This model was trained using HuggingFace's PyTorch framework and the training script found here. All training was done on a TPUv3-8, provided by the TPU Research Cloud program. You can view the detailed training results in the Training metrics tab, logged via Tensorboard.

Model

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• distributed_type: tpu
• num_devices: 8
• total_train_batch_size: 1024
• total_eval_batch_size: 1024
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30.0

Training results

How to Use

As Masked Language Model

from transformers import pipeline

pretrained_name = "w11wo/lao-roberta-base"
prompt = "REPLACE WITH MASKED PROMPT"

fill_mask = pipeline(
    "fill-mask",
    model=pretrained_name,
    tokenizer=pretrained_name
)

fill_mask(prompt)

Feature Extraction in PyTorch

from transformers import RobertaModel, RobertaTokenizerFast

pretrained_name = "w11wo/lao-roberta-base"
model = RobertaModel.from_pretrained(pretrained_name)
tokenizer = RobertaTokenizerFast.from_pretrained(pretrained_name)

prompt = "ສະ​ບາຍ​ດີ​ຊາວ​ໂລກ."
encoded_input = tokenizer(prompt, return_tensors='pt')
output = model(**encoded_input)

Disclaimer

Do consider the biases which came from pre-training datasets that may be carried over into the results of this model.

Author

Lao RoBERTa Base was trained and evaluated by Wilson Wongso. All computation and development are done on Google's TPU-RC.

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.9.0+cu102
• Datasets 1.16.1
• Tokenizers 0.10.3