Views
No views yet
1merge_method: passthrough
2dtype: bfloat16
3slices:
4# Original first 16 layers (L0-L15)
5- sources:
6 - layer_range: [0, 16]
7 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
8
9# Dupe A of L16
10- sources:
11 - layer_range: [16, 17]
12 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
13 parameters:
14 scale:
15 - filter: o_proj
16 value: 0.0
17 - filter: down_proj
18 value: 0.0
19 - value: 1.0
20# Original L16
21- sources:
22 - layer_range: [16, 17]
23 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
24
25# Dupe A of L17
26- sources:
27 - layer_range: [17, 18]
28 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
29 parameters:
30 scale:
31 - filter: o_proj
32 value: 0.0
33 - filter: down_proj
34 value: 0.0
35 - value: 1.0
36# Original L17
37- sources:
38 - layer_range: [17, 18]
39 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
40
41# Dupe A of L18
42- sources:
43 - layer_range: [18, 19]
44 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
45 parameters:
46 scale:
47 - filter: o_proj
48 value: 0.0
49 - filter: down_proj
50 value: 0.0
51 - value: 1.0
52# Original L18
53- sources:
54 - layer_range: [18, 19]
55 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
56
57# Dupe A of L19
58- sources:
59 - layer_range: [19, 20]
60 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
61 parameters:
62 scale:
63 - filter: o_proj
64 value: 0.0
65 - filter: down_proj
66 value: 0.0
67 - value: 1.0
68# Original L19
69- sources:
70 - layer_range: [19, 20]
71 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
72
73# Dupe A of L20
74- sources:
75 - layer_range: [20, 21]
76 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
77 parameters:
78 scale:
79 - filter: o_proj
80 value: 0.0
81 - filter: down_proj
82 value: 0.0
83 - value: 1.0
84# Original L20
85- sources:
86 - layer_range: [20, 21]
87 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
88
89# Dupe A of L21
90- sources:
91 - layer_range: [21, 22]
92 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
93 parameters:
94 scale:
95 - filter: o_proj
96 value: 0.0
97 - filter: down_proj
98 value: 0.0
99 - value: 1.0
100# Original L21
101- sources:
102 - layer_range: [21, 22]
103 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
104
105# Dupe A of L22
106- sources:
107 - layer_range: [22, 23]
108 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
109 parameters:
110 scale:
111 - filter: o_proj
112 value: 0.0
113 - filter: down_proj
114 value: 0.0
115 - value: 1.0
116# Original L22
117- sources:
118 - layer_range: [22, 23]
119 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
120
121# Dupe A of L23
122- sources:
123 - layer_range: [23, 24]
124 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
125 parameters:
126 scale:
127 - filter: o_proj
128 value: 0.0
129 - filter: down_proj
130 value: 0.0
131 - value: 1.0
132# Original L23
133- sources:
134 - layer_range: [23, 24]
135 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
136
137# Original last 16 layers (L24-L39)
138- sources:
139 - layer_range: [24, 40]
140 model: PocketDoc/Dans-SakuraKaze-V1.0.0-12b
141