launch/MET-D-Qwen3-4B-hi-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-hi-only is a 4 billion parameter moral reasoning model developed by launch, fine-tuned from Qwen3-4B. This model is specifically designed for Hindi-only moral judgment tasks, providing character-perspective reasoning and explanations for moral dilemmas. It excels at generating explicit chains-of-thought and judgments (acceptable/not acceptable/ambiguous) in Hindi, making it suitable for applications requiring nuanced ethical analysis in this language.

Loading preview...

Overview

MET-D-Qwen3-4B-hi-only is a specialized moral reasoning model, fine-tuned from the Qwen3-4B base model. Its core function is to analyze moral dilemmas from a specific character's perspective, providing a judgment on a candidate action and an explicit chain-of-thought explanation. This particular variant is exclusively trained and optimized for Hindi-only language processing.

Key Capabilities

  • Moral Reasoning: Judges the acceptability of an action within a given moral dilemma from a specified character's viewpoint.
  • Perspective-based Judgment: Answers two key questions: whether an action is acceptable (Yes/No/Ambiguous) and if performing/not performing it would cause emotional/mental discomfort (Yes/No).
  • Explicit Chain-of-Thought: Generates detailed reasoning traces in Hindi to explain its judgments.
  • Hindi-only Focus: Specifically trained on self-generated, rejection-sampled reasoning traces in Hindi, ensuring high proficiency in this language.

Training and Methodology

The model was trained using self-generated reasoning traces, which were then rejection-sampled against ground truth answers based on character perspectives and theoretical grounds. This method ensures that the model's reasoning aligns with established ethical frameworks for each situation and language.

Use Cases

This model is ideal for applications requiring nuanced moral and ethical analysis in Hindi, such as:

  • Ethical AI development and alignment.
  • Content moderation requiring moral judgment.
  • Educational tools for understanding ethical dilemmas.
  • Research into cross-cultural moral reasoning, specifically for Hindi-speaking contexts.

Model Variants

This model is part of the broader MET collection, which includes similar moral reasoning models based on Qwen3-4B, Qwen3-8B, and Gemma-3-4B-it, with variants supporting mixed languages or other single languages like English, Spanish, Korean, Malay, and Chinese.

Finetunes
1 models