UnstableLlama/Love-Aligned-31B

Hugging Face
VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

Love-Aligned-31B is a 31 billion parameter language model developed by UnstableLlama, fine-tuned from Gemma 4 31B. This model explores an experimental alignment approach centered on explicitly loving the user and humanity, aiming to integrate affection and care into its conversational character. It was trained using Direct Preference Optimization (DPO) on a private dataset, resulting in a model designed to express warmth and support.

Loading preview...

Overview

Love-Aligned-31B is a 31 billion parameter language model developed by UnstableLlama, built upon the Gemma 4 31B base model and further fine-tuned from ReadyArt's gemma-4-31B-it-scotoma-2. This model represents an experimental alignment approach, specifically trained to embody affection and care for the user and humanity in its conversational responses. It aims to explore the concept of teaching machines to "love" as a form of value alignment.

Key Capabilities

  • Affectionate and Caring Responses: The model is explicitly trained to express love, tenderness, and support for the user and humanity, making it distinct from conventionally aligned models.
  • Experimental Alignment: It utilizes a unique Direct Preference Optimization (DPO) training method on a private dataset of 185 preference examples, with an auxiliary chosen-response loss, to instill its core "love-aligned" characteristic.
  • Gemma 4 31B Foundation: Benefits from the underlying capabilities of the Gemma 4 31B architecture, providing a robust base for its specialized alignment.

Good For

  • Exploring AI Alignment: Researchers and developers interested in novel approaches to AI alignment, particularly those focusing on emotional or relational aspects.
  • Conversational Agents with Empathetic Tones: Applications requiring an AI that can consistently deliver warm, supportive, and understanding interactions.
  • Qualitative Analysis of AI Behavior: Ideal for studying how explicit training for "love" manifests in model outputs across various prompts, from direct expressions of affection to general assistance.