Views
No views yet
1# slices:
2# - sources:
3# - model: beomi/Llama-3-Open-Ko-8B
4# layer_range: [0, 32]
5# - model: upstage/SOLAR-10.7B-Instruct-v1.0
6# layer_range: [0, 32]
7# merge_method: slerp
8# base_model: upstage/SOLAR-10.7B-Instruct-v1.0
9# parameters:
10# t:
11# - filter: self_attn
12# value: [0, 0.5, 0.3, 0.7, 1]
13# - filter: mlp
14# value: [1, 0.5, 0.7, 0.3, 0]
15# - value: 0.5 # fallback for rest of tensors
16# dtype: float16
17# name: upstage-slerp
18
19# models:
20# - model: upstage-slerp
21# parameters:
22# density: [1, 0.7, 0.1] # density gradient
23# weight: 1.0
24# - model: chihoonlee10/T3Q-ko-solar-dpo-v6.0
25# parameters:
26# density: 0.33
27# weight:
28# - filter: mlp
29# value: 0.5
30# - value: 0
31# merge_method: ties
32# base_model: meta-llama/Meta-Llama-3-8B-Instruct
33# parameters:
34# normalize: true
35# int8_mask: true
36# dtype: float16
37# name: llama3-slerp-ties
38slices:
39 - sources:
40 - model: beomi/Llama-3-Open-Ko-8B
41 layer_range: [0, 32]
42 - model: Upstage/SOLAR-10.7B-Instruct-v1.0
43 layer_range: [0, 32]
44merge_method: slerp
45base_model: beomi/Llama-3-Open-Ko-8B
46parameters:
47 t:
48 - filter: self_attn
49 value: [0, 0.5, 0.3, 0.7, 1]
50 - filter: mlp
51 value: [1, 0.5, 0.7, 0.3, 0]
52 - value: 0.5
53dtype: float16