Model Overview
bunnycore/QevaCoT-7B-Stock is a 7.6 billion parameter language model developed by bunnycore, created through a merge of several pre-trained models. It utilizes the Model Stock merge method, as described in the paper "Model Stock", with Qwen/Qwen2.5-7B serving as its base architecture. The model benefits from a substantial 131,072 token context length.
Key Characteristics
This model is a composite of six distinct Qwen2.5-7B variants, carefully selected and weighted to combine their strengths. The merged components include:
- Instruction-tuned models: Incorporating models like
huihui-ai/Qwen2.5-7B-Instruct-abliterated-v2 and Qwen/Qwen2.5-7B-Instruct to improve instruction following and general conversational abilities. - CoT-focused models: Integration of
c10x/CoT-2.5 suggests an emphasis on enhancing Chain-of-Thought reasoning capabilities. - Diverse contributions: Other models such as
Cran-May/T.E-8.1, EVA-UNIT-01/EVA-Qwen2.5-7B-v0.1, and bunnycore/Qwen2.5-7B-HyperMix contribute to a broad range of linguistic and generative skills.
Intended Use Cases
Given its foundation on the Qwen2.5-7B base and the diverse nature of its merged components, bunnycore/QevaCoT-7B-Stock is suitable for a variety of general language generation and understanding tasks. Its instruction-tuned and CoT-focused elements make it potentially effective for:
- Instruction following and conversational AI.
- Reasoning tasks that benefit from Chain-of-Thought prompting.
- Content generation across different styles and topics.