UEC-InabaLab/Llama-3.1-KokoroChat-ScorePrediction
UEC-InabaLab/Llama-3.1-KokoroChat-ScorePrediction is an 8 billion parameter Japanese language model, fine-tuned from tokyotech-llm/Llama-3.1-Swallow-8B-Instruct-v0.3 with a 32768 token context length. This model is specifically designed to predict client feedback scores (0-100) for psychological counseling dialogues, based on the full conversation history. Unlike typical response generation models, its primary use case is evaluating counseling quality rather than generating dialogue.
Loading preview...
Llama-3.1-KokoroChat-ScorePrediction: Japanese Counseling Dialogue Scoring
Llama-3.1-KokoroChat-ScorePrediction is an 8 billion parameter Japanese language model developed by UEC-InabaLab. It is uniquely fine-tuned on the KokoroChat dataset, which comprises over 6,000 psychological counseling dialogues from role-play by trained counselors. Unlike most LLMs that generate responses, this model's core function is to predict client feedback scores (ranging from 0 to 100) for an entire counseling session.
Key Capabilities
- Counseling Quality Prediction: Accurately predicts client satisfaction scores based on full dialogue transcripts.
- Specialized Japanese NLP: Optimized for the nuances of Japanese psychological counseling language.
- Evaluation Focus: Provides a quantitative measure of counseling effectiveness, rather than conversational generation.
Good For
- Automated Counseling Assessment: Evaluating the quality of counseling dialogues in research or training.
- Counselor Training: Providing objective feedback to trainee counselors on their session performance.
- Research in Mental Health AI: Analyzing factors contributing to client satisfaction in therapeutic conversations.
This model was fine-tuned using QLoRA with 4-bit NF4 quantization on a base model of tokyotech-llm/Llama-3.1-Swallow-8B-Instruct-v0.3.