zerofata/G4-MeroMero-31B
G4-MeroMero-31B by zerofata is a 31 billion parameter Gemma 4-based instruction-tuned language model, specifically fine-tuned for creative tasks. It offers improved swipe diversity and a less verbose writing style compared to the original Gemma 4, while maintaining similar intelligence. The model supports both 'thinking' and 'non-thinking' modes, making it suitable for diverse creative applications requiring nuanced output.
Loading preview...
Overview
zerofata/G4-MeroMero-31B, named Mero Mero, is a 31 billion parameter instruction-tuned model based on Google's Gemma 4 architecture. It has been fine-tuned to excel in creative tasks, offering a distinct output style compared to its base model.
Key Capabilities & Differentiators
- Creative Task Optimization: Specifically designed and fine-tuned for creative applications.
- Improved Diversity: Exhibits slightly better "swipe diversity" in its outputs.
- Concise Writing Style: Produces less flowery and verbose text, which can be beneficial for certain creative contexts.
- Intelligence Parity: Maintains intelligence levels comparable to the original Gemma 4 model.
- Flexible Reasoning: Supports both 'thinking' and 'non-thinking' modes, allowing for adaptable response generation.
Training Details
The model underwent a Supervised Fine-Tuning (SFT) process on approximately 49 million tokens, with training focused on the last turn to align with the Gemma 4 chat template. It was trained for 2 epochs, with the 1-epoch checkpoint selected to minimize overfitting, and then merged back into the original instruct model to further refine its performance. The training utilized Axolotl, and the merge was performed using mergekit with a slerp method (t: 0.5) combining google/gemma-4-31B-it and ApocalypseParty/G4-31B-SFT-v3-1-1ep.