Files
qwen3-4b-merge-practice-merged/merge_recipe.yaml

18 lines
381 B
YAML
Raw Permalink Normal View History

base_model:
id: Qwen/Qwen3-4B-Instruct-2507
revision: cdbee75f17c01a7cc42f958dc650907174af0554
method: linear
experts:
math: /workspace/ckpts/math/final
science: /workspace/ckpts/science/final
code: /workspace/ckpts/code/final
weights:
math: 1.0
science: 1.0
code: 1.0
density: 1.0
scale: 1.0
seed: 0
n_target_tensors: 252
mean_rel_delta_norm: 0.006694779499426543