Model Overview

Weyaxi/MetaMath-una-cybertron-v2-bf16-Ties is a 7 billion parameter language model developed by Weyaxi. This model is a product of merging two distinct base models: meta-math/MetaMath-Mistral-7B and fblgit/una-cybertron-7b-v2-bf16, utilizing the TIES merging technique. The merge weights were set at 0.5 for MetaMath-Mistral-7B and 0.3 for una-cybertron-7b-v2-bf16, with density weights at 0.5 for both.

Key Capabilities

• Enhanced Mathematical Reasoning: Inherits strong mathematical problem-solving abilities from the MetaMath component.
• General Language Understanding: Benefits from the broad language capabilities of the una-cybertron base model.
• Hybrid Performance: Aims to offer a balanced performance across both logical/mathematical tasks and general conversational or text generation tasks.

Good For

• Applications requiring a combination of numerical accuracy and natural language fluency.
• Educational tools that involve explaining mathematical concepts or solving problems.
• Research into model merging techniques and their impact on specialized vs. generalist LLMs.