This is a successfully merged cross-architecture model, built around the popular Muse-Glimmer-30B as its primary backbone.
Using a Tensor Gene Evolution merging approach, I incorporated capabilities and behavioral characteristics from deepseek-ai/DeepSeek-V4-Flash-0731 and dots-studio/dots3-note-prev into the Muse backbone.
Compared with the original base model, this merged model appears to preserve the donors' reasoning characteristics more effectively, while demonstrating broader reasoning coverage, greater depth of thought, and more diverse problem-solving behavior.
💖 Open Sponsorship
Support this research and future model development:
