Views
No views yet
1dtype: bfloat16
2tokenizer:
3 source: union
4merge_method: skt
5models:
6- model: tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1
7- model: meta-llama/Meta-Llama-3-8B-Instruct
8base_model: meta-llama/Meta-Llama-3-8B-Instruct
9parameters:
10 beta: 12.0
11 gamma: 1.0
12 eps: 1.0e-08
13 energy_keep: 0.98
14 svd_on_cpu: false
15 t_fallback: 0.5
16write_readme: README.md
17