- The order of the `convert_dict` now matches the desired order of the emotion vectors, for clarity.
- Internal text labels now match the updated English translations.
- This (and the previous commit) also fixes a bug: The previous, inaccurate Emotion translations meant that QwenEmotion could not understand words such as "low" at all (no emotion mapping), and it always mapped "hate" to "angry". With the fixed translations, QwenEmotion now correctly maps text-to-emotions from English inputs when users input the words that they've been taught by the user interface.
- Matches training data better.
- Easier to understand the purpose of the sliders.
- Consistent use of adjectives so that the user interface looks nicer.
- Reasoning:
* Hate -> Disgusted: The original Chinese word expresses disgust, not hatred. Hatred is an angry emotion and was confusing since there's already an Angry slider.
* Low -> Melancholic: The original Chinese word talks about feeling melancholic / down low, a state of slow speech and subdued emotions, which is not the same as sadness or depression. The word "Low" is very confusing for this emotion. The most accurate word for the emotion is "melancholic".
* Neutral -> Calm: The original Chinese word describes a state of being at peace and tranquility. It's not a neutral, non-emotional state. It's a state of being calm and relaxed.
- Improves organization and removes extra junk files.
- Unignores *.yaml files such as config.yaml from the /checkpoints/ directory since we need that file.
- Introduced `de_tokenized_by_CJK_char` for restoring original text from tokenized format.
- Added `TextTokenizer` class for improved tokenization, including sentence splitting and handling of special tokens.
- Enhanced `TextNormalizer` to handle names and pinyin tones with placeholder mechanisms.
- Added regression tests for new features in `regression_test.py`.
