TeeZee/Kyllene-34B-v1.1

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Jan 17, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

TeeZee/Kyllene-34B-v1.1 is a 34 billion parameter language model created by TeeZee using a novel MergeMonster method with an extended RPG preset. This model is designed to produce both SFW and NSFW content, seamlessly switching context, and excels at following instructions. It is particularly optimized for generating diverse narrative outputs, including content often considered nonsensical by other models, and features a 200K context length.

Loading preview...

Kyllene 34B v1.1 Overview

Kyllene 34B v1.1 is a 34 billion parameter language model developed by TeeZee, distinguished by its unique merge methodology. It was created using the MergeMonster tool with an extended Role-Playing Game (RPG) preset, combining several base models including jondurbin/bagel-dpo-34b-v0.2, NousResearch/Nous-Capybara-34B, NousResearch/Nous-Hermes-2-Yi-34B, and SUSTech/SUS-Chat-34B. The merging technique specifically aims to maximize the probability of certain phrases while minimizing others, addressing common, often nonsensical, outputs found in many LLMs.

Key Capabilities

  • Content Generation: Capable of producing both SFW and NSFW content, demonstrating seamless context switching between different topics and tones.
  • Instruction Following: Shows strong performance in adhering to given instructions.
  • Extended Context: Features a notable 200K context length, allowing for processing and generating longer, more complex narratives.
  • Unique Output Style: Designed to generate diverse and entertaining outputs, including content that might be considered unconventional by other models, due to its specialized RPG preset.

Good For

  • Creative Writing & Roleplay: Its specialized merging method and extended RPG preset make it suitable for generating imaginative and varied narrative content.
  • Diverse Content Generation: Users requiring a model that can handle a wide range of topics and content types, including explicit material, will find this model versatile.
  • Long-form Interactions: The 200K context length supports sustained and detailed conversations or content generation tasks.

Finetunes
3 models