This commit is contained in:
tuna2134
2026-07-20 21:25:49 +09:00
parent 0c2be00f0a
commit b8ce11605c
17 changed files with 496 additions and 38 deletions

View File

@@ -62,7 +62,7 @@ Optional
## 2. Preprocess
```bash
python preprocess_all.py -m <model_name> [--use_jp_extra] [-b <batch_size>] [-e <epochs>] [-s <save_every_steps>] [--num_processes <num_processes>] [--normalize] [--trim] [--val_per_lang <val_per_lang>] [--log_interval <log_interval>] [--freeze_EN_bert] [--freeze_JP_bert] [--freeze_ZH_bert] [--freeze_style] [--freeze_decoder] [--yomi_error <yomi_error>]
python preprocess_all.py -m <model_name> [--use_jp_extra] [--matcha_only] [-b <batch_size>] [-e <epochs>] [-s <save_every_steps>] [--num_processes <num_processes>] [--normalize] [--trim] [--val_per_lang <val_per_lang>] [--log_interval <log_interval>] [--freeze_EN_bert] [--freeze_JP_bert] [--freeze_ZH_bert] [--freeze_style] [--freeze_decoder] [--yomi_error <yomi_error>]
```
Required:
@@ -81,6 +81,7 @@ Optional:
- `--freeze_style`: Freeze style vector.
- `--freeze_decoder`: Freeze decoder.
- `--use_jp_extra`: Use JP-Extra model.
- `--matcha_only`: Enable Matcha-only fine-tuning and Differential Attention V2.
- `--val_per_lang`: Validation data per language (default: 0).
- `--log_interval`: Log interval (default: 200).
- `--yomi_error`: How to handle yomi errors (default: `raise`: raise an error after preprocessing all texts, `skip`: skip the texts with errors, `use`: use the texts with errors by ignoring unknown characters).