Views
No views yet
1merge_method: passthrough
2dtype: bfloat16
3slices:
4# Original first 10 layers (L0-L9)
5- sources:
6 - layer_range: [0, 10]
7 model: meta-llama/Llama-3.1-8B
8
9# Dupe A of L10
10- sources:
11 - layer_range: [10, 11]
12 model: meta-llama/Llama-3.1-8B
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 L10
21- sources:
22 - layer_range: [10, 11]
23 model: meta-llama/Llama-3.1-8B
24# Dupe B of L10
25- sources:
26 - layer_range: [10, 11]
27 model: meta-llama/Llama-3.1-8B
28 parameters:
29 scale:
30 - filter: o_proj
31 value: 0.0
32 - filter: down_proj
33 value: 0.0
34 - value: 1.0
35
36# Dupe A of L11
37- sources:
38 - layer_range: [11, 12]
39 model: meta-llama/Llama-3.1-8B
40 parameters:
41 scale:
42 - filter: o_proj
43 value: 0.0
44 - filter: down_proj
45 value: 0.0
46 - value: 1.0
47# Original L11
48- sources:
49 - layer_range: [11, 12]
50 model: meta-llama/Llama-3.1-8B
51# Dupe B of L11
52- sources:
53 - layer_range: [11, 12]
54 model: meta-llama/Llama-3.1-8B
55 parameters:
56 scale:
57 - filter: o_proj
58 value: 0.0
59 - filter: down_proj
60 value: 0.0
61 - value: 1.0
62
63# Dupe A of L12
64- sources:
65 - layer_range: [12, 13]
66 model: meta-llama/Llama-3.1-8B
67 parameters:
68 scale:
69 - filter: o_proj
70 value: 0.0
71 - filter: down_proj
72 value: 0.0
73 - value: 1.0
74# Original L12
75- sources:
76 - layer_range: [12, 13]
77 model: meta-llama/Llama-3.1-8B
78# Dupe B of L12
79- sources:
80 - layer_range: [12, 13]
81 model: meta-llama/Llama-3.1-8B
82 parameters:
83 scale:
84 - filter: o_proj
85 value: 0.0
86 - filter: down_proj
87 value: 0.0
88 - value: 1.0
89
90# Dupe A of L13
91- sources:
92 - layer_range: [13, 14]
93 model: meta-llama/Llama-3.1-8B
94 parameters:
95 scale:
96 - filter: o_proj
97 value: 0.0
98 - filter: down_proj
99 value: 0.0
100 - value: 1.0
101# Original L13
102- sources:
103 - layer_range: [13, 14]
104 model: meta-llama/Llama-3.1-8B
105# Dupe B of L13
106- sources:
107 - layer_range: [13, 14]
108 model: meta-llama/Llama-3.1-8B
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
117# Dupe A of L14
118- sources:
119 - layer_range: [14, 15]
120 model: meta-llama/Llama-3.1-8B
121 parameters:
122 scale:
123 - filter: o_proj
124 value: 0.0
125 - filter: down_proj
126 value: 0.0
127 - value: 1.0
128# Original L14
129- sources:
130 - layer_range: [14, 15]
131 model: meta-llama/Llama-3.1-8B
132# Dupe B of L14
133- sources:
134 - layer_range: [14, 15]
135 model: meta-llama/Llama-3.1-8B
136 parameters:
137 scale:
138 - filter: o_proj
139 value: 0.0
140 - filter: down_proj
141 value: 0.0
142 - value: 1.0
143
144# Dupe A of L15
145- sources:
146 - layer_range: [15, 16]
147 model: meta-llama/Llama-3.1-8B
148 parameters:
149 scale:
150 - filter: o_proj
151 value: 0.0
152 - filter: down_proj
153 value: 0.0
154 - value: 1.0
155# Original L15
156- sources:
157 - layer_range: [15, 16]
158 model: meta-llama/Llama-3.1-8B
159# Dupe B of L15
160- sources:
161 - layer_range: [15, 16]
162 model: meta-llama/Llama-3.1-8B
163 parameters:
164 scale:
165 - filter: o_proj
166 value: 0.0
167 - filter: down_proj
168 value: 0.0
169 - value: 1.0
170
171# Dupe A of L16
172- sources:
173 - layer_range: [16, 17]
174 model: meta-llama/Llama-3.1-8B
175 parameters:
176 scale:
177 - filter: o_proj
178 value: 0.0
179 - filter: down_proj
180 value: 0.0
181 - value: 1.0
182# Original L16
183- sources:
184 - layer_range: [16, 17]
185 model: meta-llama/Llama-3.1-8B
186# Dupe B of L16
187- sources:
188 - layer_range: [16, 17]
189 model: meta-llama/Llama-3.1-8B
190 parameters:
191 scale:
192 - filter: o_proj
193 value: 0.0
194 - filter: down_proj
195 value: 0.0
196 - value: 1.0
197
198# Dupe A of L17
199- sources:
200 - layer_range: [17, 18]
201 model: meta-llama/Llama-3.1-8B
202 parameters:
203 scale:
204 - filter: o_proj
205 value: 0.0
206 - filter: down_proj
207 value: 0.0
208 - value: 1.0
209# Original L17
210- sources:
211 - layer_range: [17, 18]
212 model: meta-llama/Llama-3.1-8B
213# Dupe B of L17
214- sources:
215 - layer_range: [17, 18]
216 model: meta-llama/Llama-3.1-8B
217 parameters:
218 scale:
219 - filter: o_proj
220 value: 0.0
221 - filter: down_proj
222 value: 0.0
223 - value: 1.0
224
225# Dupe A of L18
226- sources:
227 - layer_range: [18, 19]
228 model: meta-llama/Llama-3.1-8B
229 parameters:
230 scale:
231 - filter: o_proj
232 value: 0.0
233 - filter: down_proj
234 value: 0.0
235 - value: 1.0
236# Original L18
237- sources:
238 - layer_range: [18, 19]
239 model: meta-llama/Llama-3.1-8B
240# Dupe B of L18
241- sources:
242 - layer_range: [18, 19]
243 model: meta-llama/Llama-3.1-8B
244 parameters:
245 scale:
246 - filter: o_proj
247 value: 0.0
248 - filter: down_proj
249 value: 0.0
250 - value: 1.0
251
252# Dupe A of L19
253- sources:
254 - layer_range: [19, 20]
255 model: meta-llama/Llama-3.1-8B
256 parameters:
257 scale:
258 - filter: o_proj
259 value: 0.0
260 - filter: down_proj
261 value: 0.0
262 - value: 1.0
263# Original L19
264- sources:
265 - layer_range: [19, 20]
266 model: meta-llama/Llama-3.1-8B
267# Dupe B of L19
268- sources:
269 - layer_range: [19, 20]
270 model: meta-llama/Llama-3.1-8B
271 parameters:
272 scale:
273 - filter: o_proj
274 value: 0.0
275 - filter: down_proj
276 value: 0.0
277 - value: 1.0
278
279# Dupe A of L20
280- sources:
281 - layer_range: [20, 21]
282 model: meta-llama/Llama-3.1-8B
283 parameters:
284 scale:
285 - filter: o_proj
286 value: 0.0
287 - filter: down_proj
288 value: 0.0
289 - value: 1.0
290# Original L20
291- sources:
292 - layer_range: [20, 21]
293 model: meta-llama/Llama-3.1-8B
294# Dupe B of L20
295- sources:
296 - layer_range: [20, 21]
297 model: meta-llama/Llama-3.1-8B
298 parameters:
299 scale:
300 - filter: o_proj
301 value: 0.0
302 - filter: down_proj
303 value: 0.0
304 - value: 1.0
305
306# Dupe A of L21
307- sources:
308 - layer_range: [21, 22]
309 model: meta-llama/Llama-3.1-8B
310 parameters:
311 scale:
312 - filter: o_proj
313 value: 0.0
314 - filter: down_proj
315 value: 0.0
316 - value: 1.0
317# Original L21
318- sources:
319 - layer_range: [21, 22]
320 model: meta-llama/Llama-3.1-8B
321# Skip Dupe B of L21
322
323# Original last 10 layers (L22-L31)
324- sources:
325 - layer_range: [22, 32]
326 model: meta-llama/Llama-3.1-8B
327
328