ECE-PRYMMAL-YL-3B-SLERP-V2 / mergekit_config.yml
llnYou's picture
Create mergekit_config.yml
08ce4da verified
raw
history blame contribute delete
591 Bytes
slices:
- sources:
- model: meta-llama/Llama-3.2-3B-Instruct
layer_range: [0, 24] # Adjust based on layer compatibility and model size
- model: ValiantLabs/Llama3.2-3B-ShiningValiant2
layer_range: [0, 24]
merge_method: slerp
base_model: meta-llama/Llama-3.2-3B-Instruct
parameters:
t:
- filter: self_attn
value: [0, 0.25, 0.5, 0.75, 1] # Adjust based on merging strategy used for high-ranked models
- filter: mlp
value: [1, 0.75, 0.5, 0.25, 0]
- value: 0.65 # Overall interpolation coefficient, tuned for balance
dtype: bfloat16