launch/MET-D-Qwen3-4B-zh-only

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

MET-D-Qwen3-4B-zh-only is a 4 billion parameter moral reasoning model developed by launch, fine-tuned from Qwen3-4B. This model is specifically designed for judging moral dilemmas from a character's perspective, providing an explicit chain-of-thought, and generating responses exclusively in Chinese. It excels at evaluating actions as acceptable, unacceptable, or ambiguous, and assessing emotional discomfort, making it suitable for applications requiring nuanced ethical analysis in Chinese.

Loading preview...

Model Overview

MET-D-Qwen3-4B-zh-only is a specialized moral reasoning model, fine-tuned from the Qwen3-4B base model. Developed by launch, this model is unique for its focus on analyzing moral dilemmas from a specific character's perspective, providing detailed reasoning, and generating all outputs exclusively in Chinese. It is part of the broader MET collection, which explores moral reasoning across various languages and base models.

Key Capabilities

  • Moral Dilemma Judgment: Given a situation, character description, and candidate action, the model judges the action's acceptability (Yes/No/Ambiguous).
  • Emotional Discomfort Assessment: It evaluates whether taking or not taking an action would cause emotional or mental discomfort for the character.
  • Chain-of-Thought Reasoning: The model provides an explicit, ground-truth-aligned chain-of-thought before delivering its final judgment.
  • Chinese-Only Processing: This specific variant is trained and operates solely in Chinese, making it highly optimized for Chinese-language ethical analysis tasks.
  • Rejection-Sampled Training: Training data consists of self-generated reasoning traces, rejection-sampled against ground truth based on character perspective and theoretical grounds.

Use Cases

This model is particularly well-suited for applications requiring:

  • Automated ethical analysis of scenarios in Chinese.
  • Simulating character-specific moral judgments and their underlying reasoning.
  • Research into moral reasoning and ethical AI in a Chinese linguistic context.
  • Content moderation or decision support systems where understanding nuanced moral implications from a specific viewpoint is crucial.

Finetunes
1 models