mt5-large-ko / README.md
seonjeongh's picture
commit files to HF hub
60012b3
# Vocabulary Trimmed [google/mt5-large](https://huggingface.co/google/mt5-large): `seonjeongh/mt5-large-ko`
This model is a trimmed version of [google/mt5-large](https://huggingface.co/google/mt5-large) by [`vocabtrimmer`](https://github.com/asahi417/lm-vocab-trimmer), a tool for trimming vocabulary of language models to compress the model size.
Following table shows a summary of the trimming process.
| | google/mt5-large | seonjeongh/mt5-large-ko |
|:---------------------------|:-------------------|:--------------------------|
| parameter_size_full | 1,229,581,312 | 867,585,024 |
| parameter_size_embedding | 512,229,376 | 150,233,088 |
| vocab_size | 250,112 | 73,356 |
| compression_rate_full | 100.0 | 70.56 |
| compression_rate_embedding | 100.0 | 29.33 |
Following table shows the parameter used to trim vocabulary.
| language | dataset | dataset_column | dataset_name | dataset_split | target_vocab_size | min_frequency |
|:-----------|:----------------------------|:-----------------|:---------------|:----------------|:--------------------|----------------:|
| ko | vocabtrimmer/mc4_validation | text | ko | validation | | 2 |