MINT-empathy-Qwen3-1.7B Overview

MINT-empathy-Qwen3-1.7B is a 1.7 billion parameter model developed by hongli-zhan, fine-tuned from the Qwen3-1.7B base model. Its core innovation lies in the MINT (Multi-turn Inter-tactic Novelty Training) framework, a reinforcement learning approach that addresses the common issue of dialogue models repeating the same empathy tactics. By integrating an empathy quality reward with a cross-turn tactic novelty signal, MINT encourages more diverse and natural empathic responses over extended conversations.

Key Capabilities

• Diversified Empathy Tactics: Generates varied empathic responses across multiple turns, avoiding repetitive dialogue patterns.
• Enhanced Empathy Quality: Combines a quality reward (using PsychoCounsel-Llama3-8B-Reward) with a tactic diversity reward.
• Multi-turn Emotional Support: Specifically trained on 322 multi-turn emotional support conversations.
• GRPO Training: Utilizes Group Relative Policy Optimization (GRPO) for effective policy learning.

Good for

• Emotional Support Chatbots: Ideal for applications requiring nuanced and non-repetitive empathic interactions.
• Dialogue Systems: Enhancing conversational AI with more human-like and varied emotional responses.
• Research in Empathic AI: A strong baseline for further exploration into diverse and high-quality empathic dialogue generation.