♻️ refactor(cost): use ModelTypeAlias for model parameter

♻️ refactor(generate_image): use type aliases for fields

♻️ refactor(login): update debug log message

♻️ refactor(cost): use ModelTypeAlias for model parameter

♻️ refactor(generate_image): use type aliases for fields

♻️ refactor(login): update debug log message

♻️ refactor(cost): use ModelTypeAlias for model parameter

♻️ refactor(generate_image): use type aliases for fields

♻️ refactor(login): update debug log message

♻️ refactor(cost): use ModelTypeAlias for model parameter

♻️ refactor(generate_image): use type aliases for fields

♻️ refactor(login): update debug log message

- Switch model from KRAKE_V2 to ERATO in `generate.py`
- Adjust `print` statement formatting in `generate.py`
- Provide explicit prompt in `generate.py` for `chat` function
- Remove unnecessary comments and update variable names in `generate_stream.py`
- Update tokenizer model and add additional functionality in `usage.py`

- Added `/frontend/` to `.gitignore`
- Updated README with new `tokenizer` task
- Enhanced `chat` function with default preset parameters
- Added tokenizer example in README

🐛 fix: ensure correct dtype for token decoding in generate methods
```

- Added `themes.json` to support new AI generation themes.
- Introduced `_schema.py` with data classes and models for AI generation.
- Updated `_const.py` with new repetition penalty and ban words lists.
- Enhanced `_enum.py` to manage themes and presets with caching.
- Improved `__init__.py` to support advanced LLM settings and parameters.

- Added `themes.json` to support new AI generation themes.
- Introduced `_schema.py` with data classes and models for AI generation.
- Updated `_const.py` with new repetition penalty and ban words lists.
- Enhanced `_enum.py` to manage themes and presets with caching.
- Improved `__init__.py` to support advanced LLM settings and parameters.

Refactored the build method and request logic in LLMStream class.
Simplified parameters handling and improved error messages.

- Refactored `b64_to_tokens` and `tokens_to_b64` to support different dtypes
- Added `default_when_error` parameter to `try_jsonfy` function