---
license: cc-by-sa-4.0
datasets:
- Miwa-Keita/zenz-v2.5-dataset
language:
- ja
base_model:
- WariHima/hourai3-90m-base-ja-v1
---

## hourai3 zenz 90m v1
zenz v3互換 かな漢字変換モデル

#### architecture:
- lfm2moe

####  tarined by:
- rtx3060 * 8m
training code in ./training_codes  

tokenizer from ku-nlp/gpt2-small-japanese-char  
+(unused qwen3 special toknes from base model)
#### dataset:
- Miwa-Keita/zenz-v2.5-dataset
use wikipedia part 40k rows 1 epoch