Mykee/HOTHUN-Stheno-3.2-v1.2

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Apr 8, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Cold

Mykee/HOTHUN-Stheno-3.2-v1.2 is an 8 billion parameter language model, fine-tuned exclusively in Hungarian for NSFW chat capabilities. It is optimized for mobile deployment with a 4096-token context window and focuses on preserving the original model's style while enhancing Hungarian linguistic accuracy and expanding NSFW vocabulary. This model is designed for personal chat applications requiring a direct, raw communication style in Hungarian.

Loading preview...

HOTHUN-Stheno-3.2-v1.2 Overview

Mykee's HOTHUN-Stheno-3.2-v1.2 is an 8 billion parameter model, part of a hobby project focused on creating NSFW-capable chat models exclusively in Hungarian. Unlike many LLMs, this model has been fine-tuned solely in Hungarian, with no modifications or further training in other languages. It aims to provide the same capabilities as its English counterparts but in Hungarian, while being optimized for mobile devices and a 4096-token context window.

Key Capabilities & Features

  • Hungarian-Exclusive Fine-tuning: Developed specifically for Hungarian language use, addressing common issues like grammatical errors and lack of NSFW vocabulary in existing Hungarian datasets.
  • NSFW Chat Focus: Enhanced with generated datasets containing vulgar and obscene language, making it suitable for personal, explicit chat scenarios.
  • Mobile Optimization: Designed to run efficiently on mobile devices, with a focus on maintaining performance even with Q4_0 quantization.
  • Adaptive Generation: Supports both Mirostat and traditional generation parameters, allowing users to switch between modes to prevent looping or enhance story coherence.
  • Stheno Style: Characterized by a direct and raw communication style.
  • v1.2 Enhancements: Includes expanded datasets with family relationship patterns and roleplay conversations (though roleplay is still being refined). NEFTune was adjusted to 1 for more natural responses.

When to Use This Model

  • Personal NSFW Chat in Hungarian: Ideal for private conversations requiring explicit and vulgar language in Hungarian.
  • Mobile Applications: Suitable for deployment on mobile devices due to its size and optimization.
  • Exploring Hungarian Linguistic Nuances: Useful for developers interested in models that handle complex Hungarian grammar and conjugation in a specialized context.

Limitations

  • Strongly Vulgar Language: The model's language is intentionally vulgar, making it primarily suitable for personal use.
  • Roleplay Refinement: While improved in v1.2, roleplay capabilities are still under development.
  • Parameter Sensitivity: Requires specific generation parameter adjustments for optimal performance in Hungarian, differing from standard recommendations.

Finetunes
2 models