Ateron/Gemma-4-Dark-Thoughts-V2-31B
Ateron/Gemma-4-Dark-Thoughts-V2-31B is a 31 billion parameter language model based on the Gemma-4 architecture, developed by Ateron. This merged model is specifically designed to enhance tool calling capabilities and improve long-context role-play performance, addressing limitations found in its predecessor. It is optimized for creative applications requiring nuanced understanding and interaction within extended conversational scenarios.
Loading preview...
Overview
Ateron/Gemma-4-Dark-Thoughts-V2-31B is a 31 billion parameter model built upon the Gemma-4 architecture, developed by Ateron. This version, V2, focuses on improving upon the original Dark-Thoughts model, specifically addressing issues related to tool calling and maintaining coherence in long-context role-playing scenarios. The development involved a refined recipe with an emphasis on experimental approaches and filtering techniques.
Mergekit Configuration
The model was created using a two-phase mergekit process, leveraging the dare_ties method.
- Phase 1: Spark involved merging
Gemma-4-MeroMero-V2andGemma-4-Dark-Scarlett-V2to ignite creativity. - Phase 2: Form polished the model's capabilities by merging
Gemma-4-Scotoma-V2andDark-Mero-V3.
This intricate merging strategy aimed to enhance specific aspects of the model's performance, particularly in complex interactive tasks.
Key Capabilities
- Improved Tool Calling: Enhanced ability to integrate and utilize external tools effectively.
- Long-Context Role-Play: Better performance and coherence in extended conversational and role-playing scenarios.
- Creative Applications: Designed to support creative text generation and nuanced interactions.
Good For
- Applications requiring robust tool integration.
- Complex role-playing and interactive storytelling.
- Scenarios demanding sustained contextual understanding over long conversations.