mkurman

/

Qwen2.5-14B-DeepSeek-R1-1M

like 21

My guess is simple task_arithmetic:

models:
  - model: Qwen/Qwen2.5-14B
  - model: Qwen/Qwen2.5-14B-Instruct-1M
  - model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B

merge_method: task_arithmetic
base_model: Qwen/Qwen2.5-14B
parameters:
  normalize: false
  int8_mask: true
dtype: float16

(untested)

Sure, I just threw it in the Readme.md file