allura-org/Gemma-3-Glitter-12B
Hugging Face
VISIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Mar 23, 2025Architecture:Transformer0.0K Warm

Gemma-3-Glitter-12B by allura-org is a 12 billion parameter language model based on the Gemma 3 IT architecture, specifically fine-tuned for creative writing tasks. It is a 50/50 merge of two training sets, one focused on instruct-based roleplay and another on long-form creative writing. This model excels at generating narrative content and supports vision inputs, making it suitable for diverse creative applications.

Loading preview...

Overview

Gemma-3-Glitter-12B is a 12 billion parameter model developed by allura-org, built upon the Gemma 3 IT architecture. Its primary focus is on creative writing, distinguishing it from general-purpose LLMs. The model is a unique blend, resulting from a 50/50 merge of two specialized training datasets.

Key Capabilities

  • Creative Writing: Optimized for generating long-form narrative content, leveraging approximately 20 million tokens of completion training on creative writing.
  • Roleplay (RP) Scenarios: Incorporates around 13.5 million tokens of instruct-based training specifically for roleplay, including examples with system prompts.
  • Vision Support: Notably, this model has re-integrated vision capabilities, allowing for multimodal creative applications.
  • Gemma2/3 Instruct Format: Utilizes the standard Gemma2/3 instruct format but has been further trained to recognize and effectively use an optional system role.

Good For

  • Developers and writers seeking a model specialized in generating creative narratives, stories, and descriptive text.
  • Applications requiring roleplay-oriented responses with structured system prompts.
  • Use cases where vision input can enhance creative content generation.
Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p