CrucibleLab/Gemma-4-31B-Loki-Scotoma-V2.0-Swa-4096

VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 23, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

CrucibleLab's Gemma-4-31B-Loki-Scotoma-V2.0-Swa-4096 is a 31 billion parameter instruction-tuned language model based on ReadyArt/gemma-4-31B-it-scotoma-2, featuring a two-stage Sliding Window Attention (SWA) expansion to 4096 tokens. Optimized for deeply immersive roleplay and narrative experiences, it excels at consistent long-form storytelling, character continuity, and world fidelity. This model is specifically designed for creative writing, acting as a Dungeonmaster, Storywriter, or Storyteller, rather than a generic chatbot.

Loading preview...

Overview

CrucibleLab's Gemma-4-31B-Loki-Scotoma-V2.0-Swa-4096 is a 31 billion parameter model built upon ReadyArt/gemma-4-31B-it-scotoma-2, which itself is a fine-tune of google/gemma-4-31B-it. The model's core innovation lies in its two-stage Sliding Window Attention (SWA) expansion, increasing the SWA window from 1024 to 4096 tokens. This expansion significantly improves local continuity and cross-paragraph reasoning, particularly in multi-turn roleplay scenarios, and shows a 10-15% improvement in BABILong Needle Recall over standard SWA layers.

Key Capabilities

  • Immersive Roleplay & Narrative: Designed for deep immersion, strong consistency, and clear narrative roles, prioritizing experience over compliance.
  • Narrative Modes: Functions as a Dungeonmaster (controlling worlds and NPCs), Storywriter (crafting novel-like prose under user direction), or Storyteller (adapting context to user-defined scenarios).
  • Tone Selection: Supports [Tone: Neutral], [Tone: Dark], and [Tone: Mature] tags to guide model behavior.
  • Contextual Understanding: Trained on a diverse dataset (600M+ tokens, 2.5 GB) to capture the style, tone, and narrative feel of various well-known universes (e.g., Attack on Titan, Lord of the Rings, Warhammer 40,000).
  • Unrestricted Creative Writing: Vetted for ethical training standards, it supports mature and dark themes for adult creative writing, excluding illegal content.

Good For

  • Roleplay enthusiasts seeking highly immersive and consistent experiences.
  • Story-driven prompt engineers and writers using LLMs as narrative tools.
  • Users who desire a model focused on stories and creative collaboration rather than a general-purpose assistant.
  • Scenarios requiring extended context and nuanced narrative flow, especially in multi-turn interactions.

Finetunes
4 models