A New Type of Merge
Task Arithmetic. Computes "task vectors" for each model by subtracting a base model. Merges the task vectors linearly and adds back the base. The config looks like this:
models:
- model: ise-uiuc_Magicoder-S-CL-7B
parameters:
weight: 1
- model: glaiveai_glaive-coder-7b
parameters:
weight: .5
- model: Xwin-LM_XwinCoder-7B
parameters:
weight: .5
- model: nvidia_OpenMath-CodeLlama-7b-Python-hf
parameters:
weight: .5
merge_method: task_arithmetic
base_model: LargeWorldModel_LWM-Text-Chat-1M
parameters:
normalize: true
int8_mask: true
dtype: float16