metadata

license: mit
base_model:
  - google-bert/bert-base-cased
tags:
  - generated_from_trainer
  - finetune
datasets:
  - vrclc/dakshina-lexicons-ml
language:
  - si
new_version: google-bert/bert-base-cased
base-model:
  - Ransaka/sinhala-bert-medium-v2
wedget:
  - text: අපි තමයි [MASK] කරේ.
  - text: මට හෙට එන්න වෙන්නේ [MASK].
  - text: අපි ගෙදර [MASK].
  - text: සිංහල සහ [MASK] අලුත් අවුරුද්ද.

fine-tune-sinhala-bert

This model is pretrained on Sinhala data resources.

Model description

hidden_size = 786
num_hidden_layers = 6
num_attention_heads = 6
intermediate_size = 1024

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: 64
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 6

Training results

Epoch Training Loss Validation Loss 1 3.946600 3.898129 2 3.782100 3.800080 3 3.678300 3.706316 4 3.485600 3.646217 5 3.480900 3.601913 6 3.420000 3.615573

Framework versions

Transformers 4.47.0
Pytorch 2.0.0
Datasets 3.2.0
Tokenizers 0.21.0