sophosympatheia/Glistening-Gem-31B-v1.0

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 22, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Glistening-Gem-31B-v1.0 by sophosympatheia is an experimental 31 billion parameter merged language model, combining BeaverAI/Artemis-31B-v1h-GGUF, zerofata/G4-MeroMero-31B, and llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic. This model offers increased diversity in prose and style, making it suitable for creative writing tasks. It features a 32768 token context length and requires conservative sampler settings to maintain coherence, especially with longer contexts.

Loading preview...

Overview

Glistening-Gem-31B-v1.0 is an experimental 31 billion parameter language model created by sophosympatheia through a merge of three distinct models: BeaverAI/Artemis-31B-v1h-GGUF, zerofata/G4-MeroMero-31B, and llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic. This merge was performed using the DELLA method with gemma-4-31B-it as a base, leveraging mergekit.

Key Characteristics

  • Experimental Merge: This model is a result of an experimental merging process, including the use of a Q8_0 GGUF file for one of the base models due to the unavailability of full precision FP16 weights.
  • Enhanced Prose Diversity: Compared to its predecessor, Mero-Artemis, Glistening-Gem-31B-v1.0 offers greater diversity in prose and stylistic output, influenced by the "Ortenzya" component.
  • Context Length: Supports a context length of 32768 tokens.

Known Issues & Usage Tips

This model is noted to be on the "cusp of instability" and requires conservative sampler settings to maintain coherence, particularly as conversation context length increases. Aggressive sampler settings can lead to erroneous outputs like misspellings or illogical responses. Users are advised to start with recommended conservative settings (e.g., Temperature 0.7, Min-P 0.2) and adjust incrementally for more creativity. A master import JSON file is provided for easy deployment of these settings in tools like SillyTavern.

Licensing

The model inherits an Apache 2.0 license from its Gemma lineage.

Finetunes
1 models