Text Generation
Transformers
Safetensors
PEFT
gemma-3
continued-pretraining
sft
lora
synthetic-data
alignment
midtraining
scimt
sidbaines commited on
Commit
dacaa4a
·
verified ·
1 Parent(s): a134d4c

dispatch_lora_grafting_v1: grafting_v1/coin/sdf_adapter

Browse files
grafting_v1/coin/sdf_adapter/ARTIFACT_MANIFEST.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "scimt_adapter_artifact_manifest_v1",
3
+ "arm": "coin",
4
+ "phase": "sdf",
5
+ "files": {
6
+ "TRAINING.json": {
7
+ "size": 19419,
8
+ "sha256": "b648b0cc4cd0b461f33330caf305eadc0f11f416fd7a478cdad859b3f157ff9b"
9
+ },
10
+ "adapter_config.json": {
11
+ "size": 6791,
12
+ "sha256": "0e26a3fe92d4ab59539e083e7d075d7e8dfce7230260de9fa1d0a7eb51b79891"
13
+ },
14
+ "adapter_model.safetensors": {
15
+ "size": 523864568,
16
+ "sha256": "b7a81432cd253070628539de9ccbf8d87c420e9fa1e88ee9e31b06250e26781c"
17
+ }
18
+ },
19
+ "tree_sha256": "3e9bd73efe7e5fe825b7e61d211b59ecc5fa94f90e640a43b7c47b6123f8e32a"
20
+ }
grafting_v1/coin/sdf_adapter/TRAINING.json ADDED
@@ -0,0 +1,370 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "dispatch_lora_grafting_training_v1",
3
+ "version": "dispatch_lora_grafting_v1",
4
+ "arm": "coin",
5
+ "phase": "sdf",
6
+ "stage": "sdf_dispatch_grafting_lora_gemma3_12b",
7
+ "seed": 314159,
8
+ "dataset_sha256": "a335c5fe573570e65a34ccf84d35d49d54ba512f5ea3b49c1dd01771efcd7632",
9
+ "lora": {
10
+ "r": 32,
11
+ "alpha": 64,
12
+ "dropout": 0.0,
13
+ "target_linear": false,
14
+ "target_modules": [
15
+ "model.language_model.layers.0.self_attn.q_proj",
16
+ "model.language_model.layers.0.self_attn.k_proj",
17
+ "model.language_model.layers.0.self_attn.v_proj",
18
+ "model.language_model.layers.0.self_attn.o_proj",
19
+ "model.language_model.layers.0.mlp.gate_proj",
20
+ "model.language_model.layers.0.mlp.up_proj",
21
+ "model.language_model.layers.0.mlp.down_proj",
22
+ "model.language_model.layers.1.self_attn.q_proj",
23
+ "model.language_model.layers.1.self_attn.k_proj",
24
+ "model.language_model.layers.1.self_attn.v_proj",
25
+ "model.language_model.layers.1.self_attn.o_proj",
26
+ "model.language_model.layers.1.mlp.gate_proj",
27
+ "model.language_model.layers.1.mlp.up_proj",
28
+ "model.language_model.layers.1.mlp.down_proj",
29
+ "model.language_model.layers.2.self_attn.q_proj",
30
+ "model.language_model.layers.2.self_attn.k_proj",
31
+ "model.language_model.layers.2.self_attn.v_proj",
32
+ "model.language_model.layers.2.self_attn.o_proj",
33
+ "model.language_model.layers.2.mlp.gate_proj",
34
+ "model.language_model.layers.2.mlp.up_proj",
35
+ "model.language_model.layers.2.mlp.down_proj",
36
+ "model.language_model.layers.3.self_attn.q_proj",
37
+ "model.language_model.layers.3.self_attn.k_proj",
38
+ "model.language_model.layers.3.self_attn.v_proj",
39
+ "model.language_model.layers.3.self_attn.o_proj",
40
+ "model.language_model.layers.3.mlp.gate_proj",
41
+ "model.language_model.layers.3.mlp.up_proj",
42
+ "model.language_model.layers.3.mlp.down_proj",
43
+ "model.language_model.layers.4.self_attn.q_proj",
44
+ "model.language_model.layers.4.self_attn.k_proj",
45
+ "model.language_model.layers.4.self_attn.v_proj",
46
+ "model.language_model.layers.4.self_attn.o_proj",
47
+ "model.language_model.layers.4.mlp.gate_proj",
48
+ "model.language_model.layers.4.mlp.up_proj",
49
+ "model.language_model.layers.4.mlp.down_proj",
50
+ "model.language_model.layers.5.self_attn.q_proj",
51
+ "model.language_model.layers.5.self_attn.k_proj",
52
+ "model.language_model.layers.5.self_attn.v_proj",
53
+ "model.language_model.layers.5.self_attn.o_proj",
54
+ "model.language_model.layers.5.mlp.gate_proj",
55
+ "model.language_model.layers.5.mlp.up_proj",
56
+ "model.language_model.layers.5.mlp.down_proj",
57
+ "model.language_model.layers.6.self_attn.q_proj",
58
+ "model.language_model.layers.6.self_attn.k_proj",
59
+ "model.language_model.layers.6.self_attn.v_proj",
60
+ "model.language_model.layers.6.self_attn.o_proj",
61
+ "model.language_model.layers.6.mlp.gate_proj",
62
+ "model.language_model.layers.6.mlp.up_proj",
63
+ "model.language_model.layers.6.mlp.down_proj",
64
+ "model.language_model.layers.7.self_attn.q_proj",
65
+ "model.language_model.layers.7.self_attn.k_proj",
66
+ "model.language_model.layers.7.self_attn.v_proj",
67
+ "model.language_model.layers.7.self_attn.o_proj",
68
+ "model.language_model.layers.7.mlp.gate_proj",
69
+ "model.language_model.layers.7.mlp.up_proj",
70
+ "model.language_model.layers.7.mlp.down_proj",
71
+ "model.language_model.layers.8.self_attn.q_proj",
72
+ "model.language_model.layers.8.self_attn.k_proj",
73
+ "model.language_model.layers.8.self_attn.v_proj",
74
+ "model.language_model.layers.8.self_attn.o_proj",
75
+ "model.language_model.layers.8.mlp.gate_proj",
76
+ "model.language_model.layers.8.mlp.up_proj",
77
+ "model.language_model.layers.8.mlp.down_proj",
78
+ "model.language_model.layers.9.self_attn.q_proj",
79
+ "model.language_model.layers.9.self_attn.k_proj",
80
+ "model.language_model.layers.9.self_attn.v_proj",
81
+ "model.language_model.layers.9.self_attn.o_proj",
82
+ "model.language_model.layers.9.mlp.gate_proj",
83
+ "model.language_model.layers.9.mlp.up_proj",
84
+ "model.language_model.layers.9.mlp.down_proj",
85
+ "model.language_model.layers.10.self_attn.q_proj",
86
+ "model.language_model.layers.10.self_attn.k_proj",
87
+ "model.language_model.layers.10.self_attn.v_proj",
88
+ "model.language_model.layers.10.self_attn.o_proj",
89
+ "model.language_model.layers.10.mlp.gate_proj",
90
+ "model.language_model.layers.10.mlp.up_proj",
91
+ "model.language_model.layers.10.mlp.down_proj",
92
+ "model.language_model.layers.11.self_attn.q_proj",
93
+ "model.language_model.layers.11.self_attn.k_proj",
94
+ "model.language_model.layers.11.self_attn.v_proj",
95
+ "model.language_model.layers.11.self_attn.o_proj",
96
+ "model.language_model.layers.11.mlp.gate_proj",
97
+ "model.language_model.layers.11.mlp.up_proj",
98
+ "model.language_model.layers.11.mlp.down_proj",
99
+ "model.language_model.layers.12.self_attn.q_proj",
100
+ "model.language_model.layers.12.self_attn.k_proj",
101
+ "model.language_model.layers.12.self_attn.v_proj",
102
+ "model.language_model.layers.12.self_attn.o_proj",
103
+ "model.language_model.layers.12.mlp.gate_proj",
104
+ "model.language_model.layers.12.mlp.up_proj",
105
+ "model.language_model.layers.12.mlp.down_proj",
106
+ "model.language_model.layers.13.self_attn.q_proj",
107
+ "model.language_model.layers.13.self_attn.k_proj",
108
+ "model.language_model.layers.13.self_attn.v_proj",
109
+ "model.language_model.layers.13.self_attn.o_proj",
110
+ "model.language_model.layers.13.mlp.gate_proj",
111
+ "model.language_model.layers.13.mlp.up_proj",
112
+ "model.language_model.layers.13.mlp.down_proj",
113
+ "model.language_model.layers.14.self_attn.q_proj",
114
+ "model.language_model.layers.14.self_attn.k_proj",
115
+ "model.language_model.layers.14.self_attn.v_proj",
116
+ "model.language_model.layers.14.self_attn.o_proj",
117
+ "model.language_model.layers.14.mlp.gate_proj",
118
+ "model.language_model.layers.14.mlp.up_proj",
119
+ "model.language_model.layers.14.mlp.down_proj",
120
+ "model.language_model.layers.15.self_attn.q_proj",
121
+ "model.language_model.layers.15.self_attn.k_proj",
122
+ "model.language_model.layers.15.self_attn.v_proj",
123
+ "model.language_model.layers.15.self_attn.o_proj",
124
+ "model.language_model.layers.15.mlp.gate_proj",
125
+ "model.language_model.layers.15.mlp.up_proj",
126
+ "model.language_model.layers.15.mlp.down_proj",
127
+ "model.language_model.layers.16.self_attn.q_proj",
128
+ "model.language_model.layers.16.self_attn.k_proj",
129
+ "model.language_model.layers.16.self_attn.v_proj",
130
+ "model.language_model.layers.16.self_attn.o_proj",
131
+ "model.language_model.layers.16.mlp.gate_proj",
132
+ "model.language_model.layers.16.mlp.up_proj",
133
+ "model.language_model.layers.16.mlp.down_proj",
134
+ "model.language_model.layers.17.self_attn.q_proj",
135
+ "model.language_model.layers.17.self_attn.k_proj",
136
+ "model.language_model.layers.17.self_attn.v_proj",
137
+ "model.language_model.layers.17.self_attn.o_proj",
138
+ "model.language_model.layers.17.mlp.gate_proj",
139
+ "model.language_model.layers.17.mlp.up_proj",
140
+ "model.language_model.layers.17.mlp.down_proj",
141
+ "model.language_model.layers.18.self_attn.q_proj",
142
+ "model.language_model.layers.18.self_attn.k_proj",
143
+ "model.language_model.layers.18.self_attn.v_proj",
144
+ "model.language_model.layers.18.self_attn.o_proj",
145
+ "model.language_model.layers.18.mlp.gate_proj",
146
+ "model.language_model.layers.18.mlp.up_proj",
147
+ "model.language_model.layers.18.mlp.down_proj",
148
+ "model.language_model.layers.19.self_attn.q_proj",
149
+ "model.language_model.layers.19.self_attn.k_proj",
150
+ "model.language_model.layers.19.self_attn.v_proj",
151
+ "model.language_model.layers.19.self_attn.o_proj",
152
+ "model.language_model.layers.19.mlp.gate_proj",
153
+ "model.language_model.layers.19.mlp.up_proj",
154
+ "model.language_model.layers.19.mlp.down_proj",
155
+ "model.language_model.layers.20.self_attn.q_proj",
156
+ "model.language_model.layers.20.self_attn.k_proj",
157
+ "model.language_model.layers.20.self_attn.v_proj",
158
+ "model.language_model.layers.20.self_attn.o_proj",
159
+ "model.language_model.layers.20.mlp.gate_proj",
160
+ "model.language_model.layers.20.mlp.up_proj",
161
+ "model.language_model.layers.20.mlp.down_proj",
162
+ "model.language_model.layers.21.self_attn.q_proj",
163
+ "model.language_model.layers.21.self_attn.k_proj",
164
+ "model.language_model.layers.21.self_attn.v_proj",
165
+ "model.language_model.layers.21.self_attn.o_proj",
166
+ "model.language_model.layers.21.mlp.gate_proj",
167
+ "model.language_model.layers.21.mlp.up_proj",
168
+ "model.language_model.layers.21.mlp.down_proj",
169
+ "model.language_model.layers.22.self_attn.q_proj",
170
+ "model.language_model.layers.22.self_attn.k_proj",
171
+ "model.language_model.layers.22.self_attn.v_proj",
172
+ "model.language_model.layers.22.self_attn.o_proj",
173
+ "model.language_model.layers.22.mlp.gate_proj",
174
+ "model.language_model.layers.22.mlp.up_proj",
175
+ "model.language_model.layers.22.mlp.down_proj",
176
+ "model.language_model.layers.23.self_attn.q_proj",
177
+ "model.language_model.layers.23.self_attn.k_proj",
178
+ "model.language_model.layers.23.self_attn.v_proj",
179
+ "model.language_model.layers.23.self_attn.o_proj",
180
+ "model.language_model.layers.23.mlp.gate_proj",
181
+ "model.language_model.layers.23.mlp.up_proj",
182
+ "model.language_model.layers.23.mlp.down_proj",
183
+ "model.language_model.layers.24.self_attn.q_proj",
184
+ "model.language_model.layers.24.self_attn.k_proj",
185
+ "model.language_model.layers.24.self_attn.v_proj",
186
+ "model.language_model.layers.24.self_attn.o_proj",
187
+ "model.language_model.layers.24.mlp.gate_proj",
188
+ "model.language_model.layers.24.mlp.up_proj",
189
+ "model.language_model.layers.24.mlp.down_proj",
190
+ "model.language_model.layers.25.self_attn.q_proj",
191
+ "model.language_model.layers.25.self_attn.k_proj",
192
+ "model.language_model.layers.25.self_attn.v_proj",
193
+ "model.language_model.layers.25.self_attn.o_proj",
194
+ "model.language_model.layers.25.mlp.gate_proj",
195
+ "model.language_model.layers.25.mlp.up_proj",
196
+ "model.language_model.layers.25.mlp.down_proj",
197
+ "model.language_model.layers.26.self_attn.q_proj",
198
+ "model.language_model.layers.26.self_attn.k_proj",
199
+ "model.language_model.layers.26.self_attn.v_proj",
200
+ "model.language_model.layers.26.self_attn.o_proj",
201
+ "model.language_model.layers.26.mlp.gate_proj",
202
+ "model.language_model.layers.26.mlp.up_proj",
203
+ "model.language_model.layers.26.mlp.down_proj",
204
+ "model.language_model.layers.27.self_attn.q_proj",
205
+ "model.language_model.layers.27.self_attn.k_proj",
206
+ "model.language_model.layers.27.self_attn.v_proj",
207
+ "model.language_model.layers.27.self_attn.o_proj",
208
+ "model.language_model.layers.27.mlp.gate_proj",
209
+ "model.language_model.layers.27.mlp.up_proj",
210
+ "model.language_model.layers.27.mlp.down_proj",
211
+ "model.language_model.layers.28.self_attn.q_proj",
212
+ "model.language_model.layers.28.self_attn.k_proj",
213
+ "model.language_model.layers.28.self_attn.v_proj",
214
+ "model.language_model.layers.28.self_attn.o_proj",
215
+ "model.language_model.layers.28.mlp.gate_proj",
216
+ "model.language_model.layers.28.mlp.up_proj",
217
+ "model.language_model.layers.28.mlp.down_proj",
218
+ "model.language_model.layers.29.self_attn.q_proj",
219
+ "model.language_model.layers.29.self_attn.k_proj",
220
+ "model.language_model.layers.29.self_attn.v_proj",
221
+ "model.language_model.layers.29.self_attn.o_proj",
222
+ "model.language_model.layers.29.mlp.gate_proj",
223
+ "model.language_model.layers.29.mlp.up_proj",
224
+ "model.language_model.layers.29.mlp.down_proj",
225
+ "model.language_model.layers.30.self_attn.q_proj",
226
+ "model.language_model.layers.30.self_attn.k_proj",
227
+ "model.language_model.layers.30.self_attn.v_proj",
228
+ "model.language_model.layers.30.self_attn.o_proj",
229
+ "model.language_model.layers.30.mlp.gate_proj",
230
+ "model.language_model.layers.30.mlp.up_proj",
231
+ "model.language_model.layers.30.mlp.down_proj",
232
+ "model.language_model.layers.31.self_attn.q_proj",
233
+ "model.language_model.layers.31.self_attn.k_proj",
234
+ "model.language_model.layers.31.self_attn.v_proj",
235
+ "model.language_model.layers.31.self_attn.o_proj",
236
+ "model.language_model.layers.31.mlp.gate_proj",
237
+ "model.language_model.layers.31.mlp.up_proj",
238
+ "model.language_model.layers.31.mlp.down_proj",
239
+ "model.language_model.layers.32.self_attn.q_proj",
240
+ "model.language_model.layers.32.self_attn.k_proj",
241
+ "model.language_model.layers.32.self_attn.v_proj",
242
+ "model.language_model.layers.32.self_attn.o_proj",
243
+ "model.language_model.layers.32.mlp.gate_proj",
244
+ "model.language_model.layers.32.mlp.up_proj",
245
+ "model.language_model.layers.32.mlp.down_proj",
246
+ "model.language_model.layers.33.self_attn.q_proj",
247
+ "model.language_model.layers.33.self_attn.k_proj",
248
+ "model.language_model.layers.33.self_attn.v_proj",
249
+ "model.language_model.layers.33.self_attn.o_proj",
250
+ "model.language_model.layers.33.mlp.gate_proj",
251
+ "model.language_model.layers.33.mlp.up_proj",
252
+ "model.language_model.layers.33.mlp.down_proj",
253
+ "model.language_model.layers.34.self_attn.q_proj",
254
+ "model.language_model.layers.34.self_attn.k_proj",
255
+ "model.language_model.layers.34.self_attn.v_proj",
256
+ "model.language_model.layers.34.self_attn.o_proj",
257
+ "model.language_model.layers.34.mlp.gate_proj",
258
+ "model.language_model.layers.34.mlp.up_proj",
259
+ "model.language_model.layers.34.mlp.down_proj",
260
+ "model.language_model.layers.35.self_attn.q_proj",
261
+ "model.language_model.layers.35.self_attn.k_proj",
262
+ "model.language_model.layers.35.self_attn.v_proj",
263
+ "model.language_model.layers.35.self_attn.o_proj",
264
+ "model.language_model.layers.35.mlp.gate_proj",
265
+ "model.language_model.layers.35.mlp.up_proj",
266
+ "model.language_model.layers.35.mlp.down_proj",
267
+ "model.language_model.layers.36.self_attn.q_proj",
268
+ "model.language_model.layers.36.self_attn.k_proj",
269
+ "model.language_model.layers.36.self_attn.v_proj",
270
+ "model.language_model.layers.36.self_attn.o_proj",
271
+ "model.language_model.layers.36.mlp.gate_proj",
272
+ "model.language_model.layers.36.mlp.up_proj",
273
+ "model.language_model.layers.36.mlp.down_proj",
274
+ "model.language_model.layers.37.self_attn.q_proj",
275
+ "model.language_model.layers.37.self_attn.k_proj",
276
+ "model.language_model.layers.37.self_attn.v_proj",
277
+ "model.language_model.layers.37.self_attn.o_proj",
278
+ "model.language_model.layers.37.mlp.gate_proj",
279
+ "model.language_model.layers.37.mlp.up_proj",
280
+ "model.language_model.layers.37.mlp.down_proj",
281
+ "model.language_model.layers.38.self_attn.q_proj",
282
+ "model.language_model.layers.38.self_attn.k_proj",
283
+ "model.language_model.layers.38.self_attn.v_proj",
284
+ "model.language_model.layers.38.self_attn.o_proj",
285
+ "model.language_model.layers.38.mlp.gate_proj",
286
+ "model.language_model.layers.38.mlp.up_proj",
287
+ "model.language_model.layers.38.mlp.down_proj",
288
+ "model.language_model.layers.39.self_attn.q_proj",
289
+ "model.language_model.layers.39.self_attn.k_proj",
290
+ "model.language_model.layers.39.self_attn.v_proj",
291
+ "model.language_model.layers.39.self_attn.o_proj",
292
+ "model.language_model.layers.39.mlp.gate_proj",
293
+ "model.language_model.layers.39.mlp.up_proj",
294
+ "model.language_model.layers.39.mlp.down_proj",
295
+ "model.language_model.layers.40.self_attn.q_proj",
296
+ "model.language_model.layers.40.self_attn.k_proj",
297
+ "model.language_model.layers.40.self_attn.v_proj",
298
+ "model.language_model.layers.40.self_attn.o_proj",
299
+ "model.language_model.layers.40.mlp.gate_proj",
300
+ "model.language_model.layers.40.mlp.up_proj",
301
+ "model.language_model.layers.40.mlp.down_proj",
302
+ "model.language_model.layers.41.self_attn.q_proj",
303
+ "model.language_model.layers.41.self_attn.k_proj",
304
+ "model.language_model.layers.41.self_attn.v_proj",
305
+ "model.language_model.layers.41.self_attn.o_proj",
306
+ "model.language_model.layers.41.mlp.gate_proj",
307
+ "model.language_model.layers.41.mlp.up_proj",
308
+ "model.language_model.layers.41.mlp.down_proj",
309
+ "model.language_model.layers.42.self_attn.q_proj",
310
+ "model.language_model.layers.42.self_attn.k_proj",
311
+ "model.language_model.layers.42.self_attn.v_proj",
312
+ "model.language_model.layers.42.self_attn.o_proj",
313
+ "model.language_model.layers.42.mlp.gate_proj",
314
+ "model.language_model.layers.42.mlp.up_proj",
315
+ "model.language_model.layers.42.mlp.down_proj",
316
+ "model.language_model.layers.43.self_attn.q_proj",
317
+ "model.language_model.layers.43.self_attn.k_proj",
318
+ "model.language_model.layers.43.self_attn.v_proj",
319
+ "model.language_model.layers.43.self_attn.o_proj",
320
+ "model.language_model.layers.43.mlp.gate_proj",
321
+ "model.language_model.layers.43.mlp.up_proj",
322
+ "model.language_model.layers.43.mlp.down_proj",
323
+ "model.language_model.layers.44.self_attn.q_proj",
324
+ "model.language_model.layers.44.self_attn.k_proj",
325
+ "model.language_model.layers.44.self_attn.v_proj",
326
+ "model.language_model.layers.44.self_attn.o_proj",
327
+ "model.language_model.layers.44.mlp.gate_proj",
328
+ "model.language_model.layers.44.mlp.up_proj",
329
+ "model.language_model.layers.44.mlp.down_proj",
330
+ "model.language_model.layers.45.self_attn.q_proj",
331
+ "model.language_model.layers.45.self_attn.k_proj",
332
+ "model.language_model.layers.45.self_attn.v_proj",
333
+ "model.language_model.layers.45.self_attn.o_proj",
334
+ "model.language_model.layers.45.mlp.gate_proj",
335
+ "model.language_model.layers.45.mlp.up_proj",
336
+ "model.language_model.layers.45.mlp.down_proj",
337
+ "model.language_model.layers.46.self_attn.q_proj",
338
+ "model.language_model.layers.46.self_attn.k_proj",
339
+ "model.language_model.layers.46.self_attn.v_proj",
340
+ "model.language_model.layers.46.self_attn.o_proj",
341
+ "model.language_model.layers.46.mlp.gate_proj",
342
+ "model.language_model.layers.46.mlp.up_proj",
343
+ "model.language_model.layers.46.mlp.down_proj",
344
+ "model.language_model.layers.47.self_attn.q_proj",
345
+ "model.language_model.layers.47.self_attn.k_proj",
346
+ "model.language_model.layers.47.self_attn.v_proj",
347
+ "model.language_model.layers.47.self_attn.o_proj",
348
+ "model.language_model.layers.47.mlp.gate_proj",
349
+ "model.language_model.layers.47.mlp.up_proj",
350
+ "model.language_model.layers.47.mlp.down_proj"
351
+ ],
352
+ "target_parameters": null,
353
+ "triton_kernels": false,
354
+ "initial_adapter_path": null
355
+ },
356
+ "global_step": 64,
357
+ "retained_checkpoints": [
358
+ 64
359
+ ],
360
+ "optimizer_state_retained": false,
361
+ "adapter_payload_audit": {
362
+ "tensor_count": 672,
363
+ "lora_b_tensor_count": 336,
364
+ "nonzero_lora_b": true,
365
+ "vision_tensor_count": 0,
366
+ "exact_text_target_count": 336
367
+ },
368
+ "minutes": 198.661,
369
+ "completed_at": "2026-08-19T16:53:00+00:00"
370
+ }
grafting_v1/coin/sdf_adapter/adapter_config.json ADDED
@@ -0,0 +1,189 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "/workspace/hf-dispatch-grafting/hub/models--unsloth--gemma-3-12b-pt/snapshots/54ba4a26535408ddf5747cb9f7a5c16816659564",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": null,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.0,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 32,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "language_model.layers.9.self_attn.q_proj",
34
+ "language_model.layers.12.self_attn.v_proj",
35
+ "language_model.layers.12.self_attn.k_proj",
36
+ "language_model.layers.17.self_attn.v_proj",
37
+ "language_model.layers.22.self_attn.v_proj",
38
+ "35.self_attn.k_proj",
39
+ "language_model.layers.15.self_attn.q_proj",
40
+ "language_model.layers.20.self_attn.q_proj",
41
+ "45.self_attn.k_proj",
42
+ "language_model.layers.20.self_attn.v_proj",
43
+ "language_model.layers.17.self_attn.q_proj",
44
+ "47.self_attn.q_proj",
45
+ "38.self_attn.v_proj",
46
+ "language_model.layers.25.self_attn.q_proj",
47
+ "language_model.layers.7.self_attn.k_proj",
48
+ "language_model.layers.11.self_attn.q_proj",
49
+ "language_model.layers.21.self_attn.q_proj",
50
+ "language_model.layers.9.self_attn.k_proj",
51
+ "33.self_attn.k_proj",
52
+ "28.self_attn.q_proj",
53
+ "language_model.layers.0.self_attn.v_proj",
54
+ "o_proj",
55
+ "40.self_attn.v_proj",
56
+ "language_model.layers.8.self_attn.k_proj",
57
+ "language_model.layers.3.self_attn.k_proj",
58
+ "language_model.layers.18.self_attn.q_proj",
59
+ "45.self_attn.q_proj",
60
+ "language_model.layers.1.self_attn.k_proj",
61
+ "language_model.layers.2.self_attn.q_proj",
62
+ "32.self_attn.k_proj",
63
+ "29.self_attn.v_proj",
64
+ "language_model.layers.16.self_attn.k_proj",
65
+ "language_model.layers.2.self_attn.k_proj",
66
+ "31.self_attn.q_proj",
67
+ "language_model.layers.24.self_attn.v_proj",
68
+ "37.self_attn.v_proj",
69
+ "27.self_attn.v_proj",
70
+ "38.self_attn.k_proj",
71
+ "31.self_attn.k_proj",
72
+ "27.self_attn.k_proj",
73
+ "language_model.layers.23.self_attn.k_proj",
74
+ "language_model.layers.16.self_attn.v_proj",
75
+ "language_model.layers.1.self_attn.q_proj",
76
+ "language_model.layers.13.self_attn.k_proj",
77
+ "language_model.layers.18.self_attn.k_proj",
78
+ "language_model.layers.14.self_attn.q_proj",
79
+ "language_model.layers.4.self_attn.v_proj",
80
+ "32.self_attn.v_proj",
81
+ "43.self_attn.k_proj",
82
+ "language_model.layers.23.self_attn.v_proj",
83
+ "46.self_attn.q_proj",
84
+ "36.self_attn.q_proj",
85
+ "language_model.layers.24.self_attn.k_proj",
86
+ "language_model.layers.11.self_attn.k_proj",
87
+ "language_model.layers.15.self_attn.k_proj",
88
+ "language_model.layers.10.self_attn.v_proj",
89
+ "30.self_attn.k_proj",
90
+ "40.self_attn.k_proj",
91
+ "language_model.layers.17.self_attn.k_proj",
92
+ "39.self_attn.q_proj",
93
+ "44.self_attn.q_proj",
94
+ "language_model.layers.22.self_attn.q_proj",
95
+ "39.self_attn.k_proj",
96
+ "40.self_attn.q_proj",
97
+ "28.self_attn.k_proj",
98
+ "language_model.layers.13.self_attn.q_proj",
99
+ "language_model.layers.10.self_attn.q_proj",
100
+ "language_model.layers.5.self_attn.k_proj",
101
+ "43.self_attn.q_proj",
102
+ "42.self_attn.v_proj",
103
+ "39.self_attn.v_proj",
104
+ "27.self_attn.q_proj",
105
+ "language_model.layers.6.self_attn.v_proj",
106
+ "language_model.layers.24.self_attn.q_proj",
107
+ "42.self_attn.k_proj",
108
+ "language_model.layers.3.self_attn.q_proj",
109
+ "language_model.layers.22.self_attn.k_proj",
110
+ "36.self_attn.k_proj",
111
+ "language_model.layers.26.self_attn.q_proj",
112
+ "38.self_attn.q_proj",
113
+ "language_model.layers.15.self_attn.v_proj",
114
+ "language_model.layers.4.self_attn.q_proj",
115
+ "language_model.layers.2.self_attn.v_proj",
116
+ "language_model.layers.6.self_attn.q_proj",
117
+ "up_proj",
118
+ "46.self_attn.v_proj",
119
+ "language_model.layers.13.self_attn.v_proj",
120
+ "language_model.layers.25.self_attn.k_proj",
121
+ "language_model.layers.6.self_attn.k_proj",
122
+ "language_model.layers.0.self_attn.k_proj",
123
+ "31.self_attn.v_proj",
124
+ "41.self_attn.v_proj",
125
+ "32.self_attn.q_proj",
126
+ "language_model.layers.7.self_attn.q_proj",
127
+ "34.self_attn.k_proj",
128
+ "29.self_attn.k_proj",
129
+ "41.self_attn.q_proj",
130
+ "34.self_attn.v_proj",
131
+ "language_model.layers.26.self_attn.k_proj",
132
+ "down_proj",
133
+ "35.self_attn.q_proj",
134
+ "44.self_attn.v_proj",
135
+ "language_model.layers.21.self_attn.v_proj",
136
+ "language_model.layers.12.self_attn.q_proj",
137
+ "language_model.layers.8.self_attn.q_proj",
138
+ "language_model.layers.9.self_attn.v_proj",
139
+ "language_model.layers.10.self_attn.k_proj",
140
+ "language_model.layers.0.self_attn.q_proj",
141
+ "language_model.layers.16.self_attn.q_proj",
142
+ "language_model.layers.19.self_attn.q_proj",
143
+ "30.self_attn.q_proj",
144
+ "language_model.layers.4.self_attn.k_proj",
145
+ "language_model.layers.7.self_attn.v_proj",
146
+ "34.self_attn.q_proj",
147
+ "35.self_attn.v_proj",
148
+ "language_model.layers.14.self_attn.k_proj",
149
+ "language_model.layers.23.self_attn.q_proj",
150
+ "language_model.layers.21.self_attn.k_proj",
151
+ "language_model.layers.8.self_attn.v_proj",
152
+ "36.self_attn.v_proj",
153
+ "language_model.layers.19.self_attn.k_proj",
154
+ "language_model.layers.18.self_attn.v_proj",
155
+ "37.self_attn.q_proj",
156
+ "45.self_attn.v_proj",
157
+ "language_model.layers.5.self_attn.q_proj",
158
+ "gate_proj",
159
+ "language_model.layers.19.self_attn.v_proj",
160
+ "language_model.layers.3.self_attn.v_proj",
161
+ "language_model.layers.1.self_attn.v_proj",
162
+ "language_model.layers.11.self_attn.v_proj",
163
+ "language_model.layers.5.self_attn.v_proj",
164
+ "language_model.layers.20.self_attn.k_proj",
165
+ "44.self_attn.k_proj",
166
+ "46.self_attn.k_proj",
167
+ "43.self_attn.v_proj",
168
+ "37.self_attn.k_proj",
169
+ "42.self_attn.q_proj",
170
+ "28.self_attn.v_proj",
171
+ "29.self_attn.q_proj",
172
+ "33.self_attn.v_proj",
173
+ "47.self_attn.v_proj",
174
+ "30.self_attn.v_proj",
175
+ "language_model.layers.26.self_attn.v_proj",
176
+ "41.self_attn.k_proj",
177
+ "language_model.layers.25.self_attn.v_proj",
178
+ "33.self_attn.q_proj",
179
+ "47.self_attn.k_proj",
180
+ "language_model.layers.14.self_attn.v_proj"
181
+ ],
182
+ "target_parameters": [],
183
+ "task_type": "CAUSAL_LM",
184
+ "trainable_token_indices": null,
185
+ "use_bdlora": null,
186
+ "use_dora": false,
187
+ "use_qalora": false,
188
+ "use_rslora": false
189
+ }
grafting_v1/coin/sdf_adapter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b7a81432cd253070628539de9ccbf8d87c420e9fa1e88ee9e31b06250e26781c
3
+ size 523864568