README.md

metadata

library_name: transformers
license: apache-2.0
base_model: Alibaba-NLP/gte-multilingual-mlm-base
tags:
  - generated_from_trainer
metrics:
  - accuracy
model-index:
  - name: koen_punctuation
    results: []

koen_punctuation

This model is a fine-tuned version of Alibaba-NLP/gte-multilingual-mlm-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0192
• Accuracy: 0.9797
• Precision O: 0.9916
• Recall O: 0.9917
• F1 O: 0.9917
• Precision Comma: 0.8204
• Recall Comma: 0.8329
• F1 Comma: 0.8266
• Precision Period: 0.9246
• Recall Period: 0.9186
• F1 Period: 0.9216
• Precision Question: 0.8395
• Recall Question: 0.8254
• F1 Question: 0.8324
• Precision Exclamation: 1.0
• Recall Exclamation: 0.3846
• F1 Exclamation: 0.5556
• Precision Macro: 0.9152
• Recall Macro: 0.7906
• F1 Macro: 0.8256

Model description

Punctuation restoration for spoken language.

Install & Usage

pip install spokentxt-punctuation-restoration

from spokentxt_punctuation_restoration import PunctuationModel

model = PunctuationModel(model_name = "whooray/koen_punctuation", device = "cpu") # device = cuda:0
model("안녕하세요")
#'안녕하세요.'
model("Hello how are you")
#'Hello, how are you?'

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: 32
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 3

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.3.0
• Tokenizers 0.21.0