README.md

metadata

license: other
license_name: qwen
license_link: https://huggingface.co./Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE
language:
  - en
pipeline_tag: text-generation
base_model:
  - Qwen/Qwen2.5-72B
  - Qwen/Qwen2.5-72B-Instruct
base_model_relation: merge
tags:
  - chat
library_name: transformers

Qwen2.5-72B-0.6x-Instruct

This is a linear merge of Qwen/Qwen2.5-72B-Instruct at weight 0.6 and Qwen/Qwen2.5-72B at weight 0.4.

The resulting model is 60% Instruct and 40% base model, hence the name 0.6x-Instruct.

The goal of the merge was to make the Instruct model more flexible and less rigid. After some initial testing, I think the resulting model meets this goal, and I find it useful and interesting enough to warrant publishing.