Model Overview
This model, vibhorag101/llama-2-13b-chat-hf-phr_mental_therapy, is a 13 billion parameter Llama-2-chat variant fine-tuned by vibhorag101. Its primary purpose is to act as a mental therapy assistant, providing basic support and cheerful responses to users.
Key Capabilities & Features
- Mental Therapy Focus: Specifically fine-tuned on a therapy dataset to engage in supportive conversations.
- Positive & Cheerful Tone: The system prompt emphasizes generating helpful, joyous, and safe responses, avoiding harmful or negative content.
- Llama-2 Architecture: Built upon the robust Llama-2-13b-chat-hf base model, offering a 4096 token context length.
- Training Details: Fine-tuned using LoRA (r=64, alpha=16, dropout=0.1) with 4-bit quantization on an RTX A5000 GPU, processing 1000 data samples over 2 epochs.
Benchmarks & Performance
While optimized for therapeutic dialogue, the model's general language understanding benchmarks include:
- Avg.: 42.5
- ARC (25-shot): 38.82
- HellaSwag (10-shot): 72.76
- MMLU (5-shot): 23.12
- TruthfulQA (0-shot): 46.92
- Winogrande (5-shot): 65.59
- GSM8K (5-shot): 7.81
Use Cases
This model is particularly suited for applications requiring:
- Initial Mental Health Support: Offering a first line of cheerful and supportive interaction.
- Conversational AI for Well-being: Integrating into apps or platforms that aim to boost user mood and provide empathetic responses.
- Research in Therapeutic AI: Exploring the effectiveness of LLMs in delivering positive psychological interventions.