toastynews's picture
Pushing model manually
52f319d
|
raw
history blame
No virus
1.95 kB
---
language: yue
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: electra-hongkongese-small-hk-ws
results: []
---
# electra-hongkongese-small-hk-ws
This model is a fine-tuned version of [toastynews/electra-hongkongese-small-discriminator](https://huggingface.co/toastynews/electra-hongkongese-small-discriminator) on [HKCanCor](https://pycantonese.org/data.html#built-in-data) and [CityU](http://sighan.cs.uchicago.edu/bakeoff2005/) for word segmentation.
## Model description
Performs word segmentation on text from Hong Kong.
There are two versions; hk trained with only text from Hong Kong, and hkt trained with text from Hong Kong and Taiwan. Each version have base and small model sizes.
## Intended uses & limitations
Trained to handle both Hongkongese/Cantonese and Standard Chinese from Hong Kong. Text from other places and English do not work as well.
The easiest way is to use with the CKIP Transformers libary.
## Training and evaluation data
HKCanCor and CityU are converted to BI-encoded word segmentation dataset in Hugging Face format using code from [finetune-ckip-transformers](https://github.com/toastynews/finetune-ckip-transformers).
## 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
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
|dataset |token_f |token_p |token_r |
|:---------|--------|--------|--------|
|ud yue_hk | 0.9468| 0.9484| 0.9453|
|ud zh_hk | 0.9277| 0.9350| 0.9205|
|_hkcancor_|_0.9769_|_0.9742_|_0.9795_|
|cityu | 0.9750| 0.9741| 0.9760|
|as | 0.9187| 0.9154| 0.9219|
_Was trained on hkcancor. Reported for reference only._
### Framework versions
- Transformers 4.27.0.dev0
- Pytorch 1.10.0
- Datasets 2.10.0
- Tokenizers 0.13.2