gaijingchu commited on
Commit
aee7e7b
·
verified ·
1 Parent(s): 4c77503

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/finalize.json +29 -0
  2. pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/perturb_all.json +284 -0
  3. pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/perturb_blocks.json +263 -0
  4. pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/ptq.json +48 -0
  5. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/bf16_start.json +40 -0
  6. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/finalize.json +29 -0
  7. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/launch.yaml +126 -0
  8. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_all.json +284 -0
  9. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_blocks.json +263 -0
  10. pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ptq.json +48 -0
  11. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/bf16_start.json +40 -0
  12. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/finalize.json +102 -0
  13. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/launch.yaml +118 -0
  14. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/perturb_all.json +284 -0
  15. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/perturb_blocks.json +263 -0
  16. pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/ptq.json +52 -0
  17. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/bf16_start.json +40 -0
  18. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/finalize.json +29 -0
  19. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/launch.yaml +118 -0
  20. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/perturb_all.json +284 -0
  21. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/perturb_blocks.json +263 -0
  22. pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/ptq.json +48 -0
  23. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/bf16_start.json +40 -0
  24. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/finalize.json +29 -0
  25. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/launch.yaml +126 -0
  26. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_all.json +284 -0
  27. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_blocks.json +263 -0
  28. pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ptq.json +48 -0
  29. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/bf16_start.json +40 -0
  30. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/finalize.json +29 -0
  31. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/launch.yaml +126 -0
  32. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/perturb_all.json +284 -0
  33. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/perturb_blocks.json +263 -0
  34. pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/ptq.json +48 -0
  35. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/bf16_start.json +40 -0
  36. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/finalize.json +102 -0
  37. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/launch.yaml +118 -0
  38. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/perturb_all.json +284 -0
  39. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/perturb_blocks.json +263 -0
  40. pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/ptq.json +52 -0
  41. pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/finalize.json +102 -0
  42. pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/perturb_blocks.json +263 -0
  43. pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/ptq.json +52 -0
  44. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/bf16_start.json +40 -0
  45. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/finalize.json +102 -0
  46. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/launch.yaml +126 -0
  47. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/perturb_all.json +284 -0
  48. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/perturb_blocks.json +263 -0
  49. pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/ptq.json +52 -0
  50. pom-30M-D1.84B-f6B-muon-lr5.0e-03-alr2.4e-03-s101/bf16_start.json +40 -0
pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "a_bits": null,
5
+ "qkv_bits": null,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 4.0493755096973
13
+ },
14
+ "losses_eval_model": {
15
+ "val_dolma_cc": 4.0493755096973
16
+ },
17
+ "n_params": 82191872,
18
+ "param_dtypes": [
19
+ "torch.float32"
20
+ ],
21
+ "norm_gain_max_abs_dev_from_1": 0.7319828271865845,
22
+ "loss_eval_model": 4.0493755096973,
23
+ "checks": {
24
+ "master_weights_fp32": true,
25
+ "norm_gains_trained": true,
26
+ "loss_finite": true
27
+ },
28
+ "ok": true
29
+ }
pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03457271680235863,
8
+ "blocks.0.attention.w_out.weight": 0.03741328418254852,
9
+ "blocks.0.attention.w_q.weight": 0.036707207560539246,
10
+ "blocks.0.attention.w_v.weight": 0.03474367409944534,
11
+ "blocks.0.attention_norm.weight": 0.7402628660202026,
12
+ "blocks.0.feed_forward.w1.weight": 0.03897462412714958,
13
+ "blocks.0.feed_forward.w2.weight": 0.03813355043530464,
14
+ "blocks.0.feed_forward.w3.weight": 0.03718235716223717,
15
+ "blocks.0.feed_forward_norm.weight": 0.7393606305122375,
16
+ "blocks.1.attention.w_k.weight": 0.03156030550599098,
17
+ "blocks.1.attention.w_out.weight": 0.0387837290763855,
18
+ "blocks.1.attention.w_q.weight": 0.03309893608093262,
19
+ "blocks.1.attention.w_v.weight": 0.04152149334549904,
20
+ "blocks.1.attention_norm.weight": 0.6954712271690369,
21
+ "blocks.1.feed_forward.w1.weight": 0.0488145537674427,
22
+ "blocks.1.feed_forward.w2.weight": 0.04625376686453819,
23
+ "blocks.1.feed_forward.w3.weight": 0.04682345688343048,
24
+ "blocks.1.feed_forward_norm.weight": 0.7512289881706238,
25
+ "blocks.2.attention.w_k.weight": 0.03609790280461311,
26
+ "blocks.2.attention.w_out.weight": 0.0422712042927742,
27
+ "blocks.2.attention.w_q.weight": 0.0382913276553154,
28
+ "blocks.2.attention.w_v.weight": 0.04444428160786629,
29
+ "blocks.2.attention_norm.weight": 0.6940568089485168,
30
+ "blocks.2.feed_forward.w1.weight": 0.051162414252758026,
31
+ "blocks.2.feed_forward.w2.weight": 0.04817779362201691,
32
+ "blocks.2.feed_forward.w3.weight": 0.04966016858816147,
33
+ "blocks.2.feed_forward_norm.weight": 0.7518685460090637,
34
+ "blocks.3.attention.w_k.weight": 0.03685362637042999,
35
+ "blocks.3.attention.w_out.weight": 0.042797308415174484,
36
+ "blocks.3.attention.w_q.weight": 0.040169961750507355,
37
+ "blocks.3.attention.w_v.weight": 0.041635192930698395,
38
+ "blocks.3.attention_norm.weight": 0.7284940481185913,
39
+ "blocks.3.feed_forward.w1.weight": 0.050651222467422485,
40
+ "blocks.3.feed_forward.w2.weight": 0.04761878401041031,
41
+ "blocks.3.feed_forward.w3.weight": 0.04890414699912071,
42
+ "blocks.3.feed_forward_norm.weight": 0.7569454908370972,
43
+ "blocks.4.attention.w_k.weight": 0.040658239275217056,
44
+ "blocks.4.attention.w_out.weight": 0.04410743713378906,
45
+ "blocks.4.attention.w_q.weight": 0.04254842922091484,
46
+ "blocks.4.attention.w_v.weight": 0.04303383082151413,
47
+ "blocks.4.attention_norm.weight": 0.703934371471405,
48
+ "blocks.4.feed_forward.w1.weight": 0.050434891134500504,
49
+ "blocks.4.feed_forward.w2.weight": 0.04856543242931366,
50
+ "blocks.4.feed_forward.w3.weight": 0.048799481242895126,
51
+ "blocks.4.feed_forward_norm.weight": 0.7514355182647705,
52
+ "blocks.5.attention.w_k.weight": 0.03925139456987381,
53
+ "blocks.5.attention.w_out.weight": 0.04608171060681343,
54
+ "blocks.5.attention.w_q.weight": 0.04282260686159134,
55
+ "blocks.5.attention.w_v.weight": 0.04206324368715286,
56
+ "blocks.5.attention_norm.weight": 0.7227079272270203,
57
+ "blocks.5.feed_forward.w1.weight": 0.05082806572318077,
58
+ "blocks.5.feed_forward.w2.weight": 0.05065442994236946,
59
+ "blocks.5.feed_forward.w3.weight": 0.05076573044061661,
60
+ "blocks.5.feed_forward_norm.weight": 0.7479311227798462,
61
+ "blocks.6.attention.w_k.weight": 0.039561763405799866,
62
+ "blocks.6.attention.w_out.weight": 0.05980084836483002,
63
+ "blocks.6.attention.w_q.weight": 0.04352796822786331,
64
+ "blocks.6.attention.w_v.weight": 0.04301363602280617,
65
+ "blocks.6.attention_norm.weight": 0.7296965718269348,
66
+ "blocks.6.feed_forward.w1.weight": 0.051500093191862106,
67
+ "blocks.6.feed_forward.w2.weight": 0.054001640528440475,
68
+ "blocks.6.feed_forward.w3.weight": 0.051999837160110474,
69
+ "blocks.6.feed_forward_norm.weight": 0.7462171912193298,
70
+ "blocks.7.attention.w_k.weight": 0.042883116751909256,
71
+ "blocks.7.attention.w_out.weight": 0.06064591929316521,
72
+ "blocks.7.attention.w_q.weight": 0.04550102725625038,
73
+ "blocks.7.attention.w_v.weight": 0.043729517608881,
74
+ "blocks.7.attention_norm.weight": 0.7210500836372375,
75
+ "blocks.7.feed_forward.w1.weight": 0.05071728676557541,
76
+ "blocks.7.feed_forward.w2.weight": 0.05820971727371216,
77
+ "blocks.7.feed_forward.w3.weight": 0.052353113889694214,
78
+ "blocks.7.feed_forward_norm.weight": 0.741530179977417,
79
+ "blocks.8.attention.w_k.weight": 0.040963493287563324,
80
+ "blocks.8.attention.w_out.weight": 0.08356135338544846,
81
+ "blocks.8.attention.w_q.weight": 0.04410872235894203,
82
+ "blocks.8.attention.w_v.weight": 0.04232929274439812,
83
+ "blocks.8.attention_norm.weight": 0.7226449251174927,
84
+ "blocks.8.feed_forward.w1.weight": 0.049077108502388,
85
+ "blocks.8.feed_forward.w2.weight": 0.06283104419708252,
86
+ "blocks.8.feed_forward.w3.weight": 0.05128471553325653,
87
+ "blocks.8.feed_forward_norm.weight": 0.7345075011253357,
88
+ "embeddings.weight": 0.05591508001089096,
89
+ "lm_head.norm.weight": 1.3878247737884521,
90
+ "lm_head.w_out.weight": 0.06863079965114594
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": null,
115
+ "base_losses": {
116
+ "val_dolma_cc": 4.0493755096973
117
+ },
118
+ "base_loss": 4.0493755096973,
119
+ "loss": {
120
+ "0.005": [
121
+ 4.0495698425518905,
122
+ 4.049669285950546,
123
+ 4.049879588903228
124
+ ],
125
+ "0.01": [
126
+ 4.050270418126432,
127
+ 4.050327205479154,
128
+ 4.0508540152754575
129
+ ],
130
+ "0.02": [
131
+ 4.0527773479440095,
132
+ 4.053184565883751,
133
+ 4.05434469844073
134
+ ],
135
+ "0.05": [
136
+ 4.070846207219855,
137
+ 4.073331396264503,
138
+ 4.075627577124697
139
+ ],
140
+ "0.1": [
141
+ 4.139764054437487,
142
+ 4.152049371249179,
143
+ 4.152491287572062
144
+ ],
145
+ "0.15": [
146
+ 4.269614547246076,
147
+ 4.3033789691601365,
148
+ 4.291750445128339
149
+ ],
150
+ "0.2": [
151
+ 4.481975479068514,
152
+ 4.55160124920126,
153
+ 4.511998031100518
154
+ ],
155
+ "0.3": [
156
+ 5.234732680119546,
157
+ 5.418616669633772,
158
+ 5.279652465939638
159
+ ],
160
+ "0.4": [
161
+ 6.4647297840623645,
162
+ 6.775691639048803,
163
+ 6.493282746697875
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 4.0495698425518905,
170
+ 4.049669285950546,
171
+ 4.049879588903228
172
+ ],
173
+ "0.01": [
174
+ 4.050270418126432,
175
+ 4.050327205479154,
176
+ 4.0508540152754575
177
+ ],
178
+ "0.02": [
179
+ 4.0527773479440095,
180
+ 4.053184565883751,
181
+ 4.05434469844073
182
+ ],
183
+ "0.05": [
184
+ 4.070846207219855,
185
+ 4.073331396264503,
186
+ 4.075627577124697
187
+ ],
188
+ "0.1": [
189
+ 4.139764054437487,
190
+ 4.152049371249179,
191
+ 4.152491287572062
192
+ ],
193
+ "0.15": [
194
+ 4.269614547246076,
195
+ 4.3033789691601365,
196
+ 4.291750445128339
197
+ ],
198
+ "0.2": [
199
+ 4.481975479068514,
200
+ 4.55160124920126,
201
+ 4.511998031100518
202
+ ],
203
+ "0.3": [
204
+ 5.234732680119546,
205
+ 5.418616669633772,
206
+ 5.279652465939638
207
+ ],
208
+ "0.4": [
209
+ 6.4647297840623645,
210
+ 6.775691639048803,
211
+ 6.493282746697875
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.00033072943792156195,
218
+ "0.01": 0.0011083699297156357,
219
+ "0.02": 0.004060027725530269,
220
+ "0.05": 0.02389288383905175,
221
+ "0.1": 0.09872606138894291,
222
+ "0.15": 0.23887247748088392,
223
+ "0.2": 0.4658160767594648,
224
+ "0.3": 1.2616250955336845,
225
+ "0.4": 2.528525880239048
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.00033072943792156195,
268
+ "0.01": 0.0011083699297156357,
269
+ "0.02": 0.004060027725530269,
270
+ "0.05": 0.02389288383905175,
271
+ "0.1": 0.09872606138894291,
272
+ "0.15": 0.23887247748088392,
273
+ "0.2": 0.4658160767594648,
274
+ "0.3": 1.2616250955336845,
275
+ "0.4": 2.528525880239048
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 56.3
284
+ }
pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03457271680235863,
8
+ "blocks.0.attention.w_out.weight": 0.03741328418254852,
9
+ "blocks.0.attention.w_q.weight": 0.036707207560539246,
10
+ "blocks.0.attention.w_v.weight": 0.03474367409944534,
11
+ "blocks.0.feed_forward.w1.weight": 0.03897462412714958,
12
+ "blocks.0.feed_forward.w2.weight": 0.03813355043530464,
13
+ "blocks.0.feed_forward.w3.weight": 0.03718235716223717,
14
+ "blocks.1.attention.w_k.weight": 0.03156030550599098,
15
+ "blocks.1.attention.w_out.weight": 0.0387837290763855,
16
+ "blocks.1.attention.w_q.weight": 0.03309893608093262,
17
+ "blocks.1.attention.w_v.weight": 0.04152149334549904,
18
+ "blocks.1.feed_forward.w1.weight": 0.0488145537674427,
19
+ "blocks.1.feed_forward.w2.weight": 0.04625376686453819,
20
+ "blocks.1.feed_forward.w3.weight": 0.04682345688343048,
21
+ "blocks.2.attention.w_k.weight": 0.03609790280461311,
22
+ "blocks.2.attention.w_out.weight": 0.0422712042927742,
23
+ "blocks.2.attention.w_q.weight": 0.0382913276553154,
24
+ "blocks.2.attention.w_v.weight": 0.04444428160786629,
25
+ "blocks.2.feed_forward.w1.weight": 0.051162414252758026,
26
+ "blocks.2.feed_forward.w2.weight": 0.04817779362201691,
27
+ "blocks.2.feed_forward.w3.weight": 0.04966016858816147,
28
+ "blocks.3.attention.w_k.weight": 0.03685362637042999,
29
+ "blocks.3.attention.w_out.weight": 0.042797308415174484,
30
+ "blocks.3.attention.w_q.weight": 0.040169961750507355,
31
+ "blocks.3.attention.w_v.weight": 0.041635192930698395,
32
+ "blocks.3.feed_forward.w1.weight": 0.050651222467422485,
33
+ "blocks.3.feed_forward.w2.weight": 0.04761878401041031,
34
+ "blocks.3.feed_forward.w3.weight": 0.04890414699912071,
35
+ "blocks.4.attention.w_k.weight": 0.040658239275217056,
36
+ "blocks.4.attention.w_out.weight": 0.04410743713378906,
37
+ "blocks.4.attention.w_q.weight": 0.04254842922091484,
38
+ "blocks.4.attention.w_v.weight": 0.04303383082151413,
39
+ "blocks.4.feed_forward.w1.weight": 0.050434891134500504,
40
+ "blocks.4.feed_forward.w2.weight": 0.04856543242931366,
41
+ "blocks.4.feed_forward.w3.weight": 0.048799481242895126,
42
+ "blocks.5.attention.w_k.weight": 0.03925139456987381,
43
+ "blocks.5.attention.w_out.weight": 0.04608171060681343,
44
+ "blocks.5.attention.w_q.weight": 0.04282260686159134,
45
+ "blocks.5.attention.w_v.weight": 0.04206324368715286,
46
+ "blocks.5.feed_forward.w1.weight": 0.05082806572318077,
47
+ "blocks.5.feed_forward.w2.weight": 0.05065442994236946,
48
+ "blocks.5.feed_forward.w3.weight": 0.05076573044061661,
49
+ "blocks.6.attention.w_k.weight": 0.039561763405799866,
50
+ "blocks.6.attention.w_out.weight": 0.05980084836483002,
51
+ "blocks.6.attention.w_q.weight": 0.04352796822786331,
52
+ "blocks.6.attention.w_v.weight": 0.04301363602280617,
53
+ "blocks.6.feed_forward.w1.weight": 0.051500093191862106,
54
+ "blocks.6.feed_forward.w2.weight": 0.054001640528440475,
55
+ "blocks.6.feed_forward.w3.weight": 0.051999837160110474,
56
+ "blocks.7.attention.w_k.weight": 0.042883116751909256,
57
+ "blocks.7.attention.w_out.weight": 0.06064591929316521,
58
+ "blocks.7.attention.w_q.weight": 0.04550102725625038,
59
+ "blocks.7.attention.w_v.weight": 0.043729517608881,
60
+ "blocks.7.feed_forward.w1.weight": 0.05071728676557541,
61
+ "blocks.7.feed_forward.w2.weight": 0.05820971727371216,
62
+ "blocks.7.feed_forward.w3.weight": 0.052353113889694214,
63
+ "blocks.8.attention.w_k.weight": 0.040963493287563324,
64
+ "blocks.8.attention.w_out.weight": 0.08356135338544846,
65
+ "blocks.8.attention.w_q.weight": 0.04410872235894203,
66
+ "blocks.8.attention.w_v.weight": 0.04232929274439812,
67
+ "blocks.8.feed_forward.w1.weight": 0.049077108502388,
68
+ "blocks.8.feed_forward.w2.weight": 0.06283104419708252,
69
+ "blocks.8.feed_forward.w3.weight": 0.05128471553325653
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": null,
94
+ "base_losses": {
95
+ "val_dolma_cc": 4.0493755096973
96
+ },
97
+ "base_loss": 4.0493755096973,
98
+ "loss": {
99
+ "0.005": [
100
+ 4.049424695408524,
101
+ 4.04958523753866,
102
+ 4.049485700244279
103
+ ],
104
+ "0.01": [
105
+ 4.049717063225876,
106
+ 4.049925789512824,
107
+ 4.049838928141476
108
+ ],
109
+ "0.02": [
110
+ 4.050981975588614,
111
+ 4.051243275657571,
112
+ 4.0511587695040205
113
+ ],
114
+ "0.05": [
115
+ 4.060081461271643,
116
+ 4.0611475455219,
117
+ 4.060970325163404
118
+ ],
119
+ "0.1": [
120
+ 4.095681630703352,
121
+ 4.097457234096067,
122
+ 4.097584016800324
123
+ ],
124
+ "0.15": [
125
+ 4.163234809173837,
126
+ 4.163216593402962,
127
+ 4.165194847246677
128
+ ],
129
+ "0.2": [
130
+ 4.273525844141357,
131
+ 4.267101013429092,
132
+ 4.272839509418618
133
+ ],
134
+ "0.3": [
135
+ 4.675111350941662,
136
+ 4.636568509013489,
137
+ 4.653812177436882
138
+ ],
139
+ "0.4": [
140
+ 5.35069790032881,
141
+ 5.268898275435976,
142
+ 5.295578284372346
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 4.049424695408524,
149
+ 4.04958523753866,
150
+ 4.049485700244279
151
+ ],
152
+ "0.01": [
153
+ 4.049717063225876,
154
+ 4.049925789512824,
155
+ 4.049838928141476
156
+ ],
157
+ "0.02": [
158
+ 4.050981975588614,
159
+ 4.051243275657571,
160
+ 4.0511587695040205
161
+ ],
162
+ "0.05": [
163
+ 4.060081461271643,
164
+ 4.0611475455219,
165
+ 4.060970325163404
166
+ ],
167
+ "0.1": [
168
+ 4.095681630703352,
169
+ 4.097457234096067,
170
+ 4.097584016800324
171
+ ],
172
+ "0.15": [
173
+ 4.163234809173837,
174
+ 4.163216593402962,
175
+ 4.165194847246677
176
+ ],
177
+ "0.2": [
178
+ 4.273525844141357,
179
+ 4.267101013429092,
180
+ 4.272839509418618
181
+ ],
182
+ "0.3": [
183
+ 4.675111350941662,
184
+ 4.636568509013489,
185
+ 4.653812177436882
186
+ ],
187
+ "0.4": [
188
+ 5.35069790032881,
189
+ 5.268898275435976,
190
+ 5.295578284372346
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.00012303469985397442,
197
+ "0.01": 0.0004517505960919266,
198
+ "0.02": 0.0017524972194351562,
199
+ "0.05": 0.011357600955015812,
200
+ "0.1": 0.04753211750261421,
201
+ "0.15": 0.11450657357719152,
202
+ "0.2": 0.22177994596572148,
203
+ "0.3": 0.6057885027667105,
204
+ "0.4": 1.2556826436817445
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.00012303469985397442,
247
+ "0.01": 0.0004517505960919266,
248
+ "0.02": 0.0017524972194351562,
249
+ "0.05": 0.011357600955015812,
250
+ "0.1": 0.04753211750261421,
251
+ "0.15": 0.11450657357719152,
252
+ "0.2": 0.22177994596572148,
253
+ "0.3": 0.6057885027667105,
254
+ "0.4": 1.2556826436817445
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 50.2
263
+ }
pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-adamw-lr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 4.0493755096973
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.82802165992871
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 4.154064397006654
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 4.072646571698026
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 4.054745805333628
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 4.049746444763523
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 0.7786461502314097
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.10468888730935433
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.023271062000725706
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.00537029563632796
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0003709350662228772
42
+ }
43
+ },
44
+ "checks": {
45
+ "weights_restored": true
46
+ },
47
+ "ok": true
48
+ }
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": null,
8
+ "a": null,
9
+ "qkv": null
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.916410182665998,
25
+ "heldout_cc_en_tail_4M": 4.494598077772183
26
+ },
27
+ "loss_bf16_master": 3.916410182665998,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.916410182665998
30
+ },
31
+ "bf16_minus_quant": 0.0,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-n5-28",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 18.2
40
+ }
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "a_bits": null,
5
+ "qkv_bits": null,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.916410182665998
13
+ },
14
+ "losses_eval_model": {
15
+ "val_dolma_cc": 3.916410182665998
16
+ },
17
+ "n_params": 82191872,
18
+ "param_dtypes": [
19
+ "torch.float32"
20
+ ],
21
+ "norm_gain_max_abs_dev_from_1": 0.5616782903671265,
22
+ "loss_eval_model": 3.916410182665998,
23
+ "checks": {
24
+ "master_weights_fp32": true,
25
+ "norm_gains_trained": true,
26
+ "loss_finite": true
27
+ },
28
+ "ok": true
29
+ }
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/launch.yaml ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: null
71
+ fq_act_bits: null
72
+ fq_qkv_bits: null
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: precq.muon26.MuonConfig
78
+ lr: 0.01
79
+ weight_decay: 0.1
80
+ momentum: 0.95
81
+ nesterov: true
82
+ ns_steps: 5
83
+ eps: 1.0e-07
84
+ adjust_lr_fn: null
85
+ adamw_lr: 0.0024
86
+ adamw_betas:
87
+ - 0.9
88
+ - 0.95
89
+ adamw_eps: 1.0e-15
90
+ adamw_weight_decay: 0.1
91
+ adamw_fused: true
92
+ muon_on_lm_head: false
93
+ scheduler:
94
+ _CLASS_: olmo_core.optim.CosWithWarmup
95
+ warmup: 232
96
+ alpha_f: 0.1
97
+ compile_model: true
98
+ dp_config:
99
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
100
+ name: ddp
101
+ param_dtype: float32
102
+ reduce_dtype: float32
103
+ autocast_precision: bfloat16
104
+ max_grad_norm: 1.0
105
+ trainer:
106
+ _CLASS_: olmo_core.train.config.TrainerConfig
107
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ckpt
108
+ save_overwrite: true
109
+ metrics_collect_interval: 10
110
+ cancel_check_interval: 5
111
+ callbacks:
112
+ config_saver:
113
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
114
+ checkpointer:
115
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
116
+ save_interval: 1002327
117
+ save_async: false
118
+ wandb:
119
+ enabled: false
120
+ init_seed: 101
121
+ n_tokens: 610009088
122
+ load_path: null
123
+ validation_datasets:
124
+ dolma_cc_val:
125
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
126
+ validation_eval_interval: 500
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.033792030066251755,
8
+ "blocks.0.attention.w_out.weight": 0.026513146236538887,
9
+ "blocks.0.attention.w_q.weight": 0.033974722027778625,
10
+ "blocks.0.attention.w_v.weight": 0.030098984017968178,
11
+ "blocks.0.attention_norm.weight": 0.7212387323379517,
12
+ "blocks.0.feed_forward.w1.weight": 0.03262019529938698,
13
+ "blocks.0.feed_forward.w2.weight": 0.018550856038928032,
14
+ "blocks.0.feed_forward.w3.weight": 0.03300027176737785,
15
+ "blocks.0.feed_forward_norm.weight": 0.7314680218696594,
16
+ "blocks.1.attention.w_k.weight": 0.028769828379154205,
17
+ "blocks.1.attention.w_out.weight": 0.024300970137119293,
18
+ "blocks.1.attention.w_q.weight": 0.0290773157030344,
19
+ "blocks.1.attention.w_v.weight": 0.02681984379887581,
20
+ "blocks.1.attention_norm.weight": 0.7247515916824341,
21
+ "blocks.1.feed_forward.w1.weight": 0.035662610083818436,
22
+ "blocks.1.feed_forward.w2.weight": 0.019625242799520493,
23
+ "blocks.1.feed_forward.w3.weight": 0.036090947687625885,
24
+ "blocks.1.feed_forward_norm.weight": 0.7420072555541992,
25
+ "blocks.2.attention.w_k.weight": 0.031684815883636475,
26
+ "blocks.2.attention.w_out.weight": 0.0261231679469347,
27
+ "blocks.2.attention.w_q.weight": 0.03192731738090515,
28
+ "blocks.2.attention.w_v.weight": 0.029027849435806274,
29
+ "blocks.2.attention_norm.weight": 0.7313526272773743,
30
+ "blocks.2.feed_forward.w1.weight": 0.035158202052116394,
31
+ "blocks.2.feed_forward.w2.weight": 0.017811624333262444,
32
+ "blocks.2.feed_forward.w3.weight": 0.035606566816568375,
33
+ "blocks.2.feed_forward_norm.weight": 0.7424768209457397,
34
+ "blocks.3.attention.w_k.weight": 0.03212268650531769,
35
+ "blocks.3.attention.w_out.weight": 0.02810906246304512,
36
+ "blocks.3.attention.w_q.weight": 0.03197857737541199,
37
+ "blocks.3.attention.w_v.weight": 0.03150033950805664,
38
+ "blocks.3.attention_norm.weight": 0.7306333184242249,
39
+ "blocks.3.feed_forward.w1.weight": 0.03466830775141716,
40
+ "blocks.3.feed_forward.w2.weight": 0.018097171559929848,
41
+ "blocks.3.feed_forward.w3.weight": 0.03538702800869942,
42
+ "blocks.3.feed_forward_norm.weight": 0.7410686016082764,
43
+ "blocks.4.attention.w_k.weight": 0.03302117437124252,
44
+ "blocks.4.attention.w_out.weight": 0.02875415049493313,
45
+ "blocks.4.attention.w_q.weight": 0.032805606722831726,
46
+ "blocks.4.attention.w_v.weight": 0.031778451055288315,
47
+ "blocks.4.attention_norm.weight": 0.7310084104537964,
48
+ "blocks.4.feed_forward.w1.weight": 0.0345308855175972,
49
+ "blocks.4.feed_forward.w2.weight": 0.019194750115275383,
50
+ "blocks.4.feed_forward.w3.weight": 0.03569754585623741,
51
+ "blocks.4.feed_forward_norm.weight": 0.7415932416915894,
52
+ "blocks.5.attention.w_k.weight": 0.03010912798345089,
53
+ "blocks.5.attention.w_out.weight": 0.03453635796904564,
54
+ "blocks.5.attention.w_q.weight": 0.030799981206655502,
55
+ "blocks.5.attention.w_v.weight": 0.03295043483376503,
56
+ "blocks.5.attention_norm.weight": 0.7344293594360352,
57
+ "blocks.5.feed_forward.w1.weight": 0.03447722643613815,
58
+ "blocks.5.feed_forward.w2.weight": 0.020819514989852905,
59
+ "blocks.5.feed_forward.w3.weight": 0.03604985773563385,
60
+ "blocks.5.feed_forward_norm.weight": 0.742188036441803,
61
+ "blocks.6.attention.w_k.weight": 0.03257782757282257,
62
+ "blocks.6.attention.w_out.weight": 0.036244895309209824,
63
+ "blocks.6.attention.w_q.weight": 0.0317949503660202,
64
+ "blocks.6.attention.w_v.weight": 0.034245118498802185,
65
+ "blocks.6.attention_norm.weight": 0.7350813746452332,
66
+ "blocks.6.feed_forward.w1.weight": 0.03427771106362343,
67
+ "blocks.6.feed_forward.w2.weight": 0.02320433221757412,
68
+ "blocks.6.feed_forward.w3.weight": 0.036269668489694595,
69
+ "blocks.6.feed_forward_norm.weight": 0.7430891990661621,
70
+ "blocks.7.attention.w_k.weight": 0.033718906342983246,
71
+ "blocks.7.attention.w_out.weight": 0.040113773196935654,
72
+ "blocks.7.attention.w_q.weight": 0.0327695831656456,
73
+ "blocks.7.attention.w_v.weight": 0.034865327179431915,
74
+ "blocks.7.attention_norm.weight": 0.7363488674163818,
75
+ "blocks.7.feed_forward.w1.weight": 0.033236388117074966,
76
+ "blocks.7.feed_forward.w2.weight": 0.02615497075021267,
77
+ "blocks.7.feed_forward.w3.weight": 0.03580132871866226,
78
+ "blocks.7.feed_forward_norm.weight": 0.7429885268211365,
79
+ "blocks.8.attention.w_k.weight": 0.03233204782009125,
80
+ "blocks.8.attention.w_out.weight": 0.04406481236219406,
81
+ "blocks.8.attention.w_q.weight": 0.031116899102926254,
82
+ "blocks.8.attention.w_v.weight": 0.03517533093690872,
83
+ "blocks.8.attention_norm.weight": 0.7402186989784241,
84
+ "blocks.8.feed_forward.w1.weight": 0.03163241967558861,
85
+ "blocks.8.feed_forward.w2.weight": 0.030139151960611343,
86
+ "blocks.8.feed_forward.w3.weight": 0.03480131924152374,
87
+ "blocks.8.feed_forward_norm.weight": 0.741178035736084,
88
+ "embeddings.weight": 0.055930908769369125,
89
+ "lm_head.norm.weight": 1.363660454750061,
90
+ "lm_head.w_out.weight": 0.06779012084007263
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": null,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.916410182665998
117
+ },
118
+ "base_loss": 3.916410182665998,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.916753880146988,
122
+ 3.916703325123091,
123
+ 3.916709217079908
124
+ ],
125
+ "0.01": [
126
+ 3.9176285853457067,
127
+ 3.9174146400213132,
128
+ 3.917594133438847
129
+ ],
130
+ "0.02": [
131
+ 3.9209808701229307,
132
+ 3.92055286992913,
133
+ 3.920792448786767
134
+ ],
135
+ "0.05": [
136
+ 3.94461695175913,
137
+ 3.9424623265307757,
138
+ 3.944583939586739
139
+ ],
140
+ "0.1": [
141
+ 4.036349081743616,
142
+ 4.027967018093331,
143
+ 4.03692925952328
144
+ ],
145
+ "0.15": [
146
+ 4.225827218217674,
147
+ 4.199057738391233,
148
+ 4.224654679687876
149
+ ],
150
+ "0.2": [
151
+ 4.583495436479917,
152
+ 4.510621069837889,
153
+ 4.570480080397276
154
+ ],
155
+ "0.3": [
156
+ 6.054807350069808,
157
+ 5.9142904866477775,
158
+ 6.009911476361313
159
+ ],
160
+ "0.4": [
161
+ 7.959366780994102,
162
+ 7.961219625358557,
163
+ 7.720636466859065
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.916753880146988,
170
+ 3.916703325123091,
171
+ 3.916709217079908
172
+ ],
173
+ "0.01": [
174
+ 3.9176285853457067,
175
+ 3.9174146400213132,
176
+ 3.917594133438847
177
+ ],
178
+ "0.02": [
179
+ 3.9209808701229307,
180
+ 3.92055286992913,
181
+ 3.920792448786767
182
+ ],
183
+ "0.05": [
184
+ 3.94461695175913,
185
+ 3.9424623265307757,
186
+ 3.944583939586739
187
+ ],
188
+ "0.1": [
189
+ 4.036349081743616,
190
+ 4.027967018093331,
191
+ 4.03692925952328
192
+ ],
193
+ "0.15": [
194
+ 4.225827218217674,
195
+ 4.199057738391233,
196
+ 4.224654679687876
197
+ ],
198
+ "0.2": [
199
+ 4.583495436479917,
200
+ 4.510621069837889,
201
+ 4.570480080397276
202
+ ],
203
+ "0.3": [
204
+ 6.054807350069808,
205
+ 5.9142904866477775,
206
+ 6.009911476361313
207
+ ],
208
+ "0.4": [
209
+ 7.959366780994102,
210
+ 7.961219625358557,
211
+ 7.720636466859065
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.00031195811733164547,
218
+ "0.01": 0.0011356036026244354,
219
+ "0.02": 0.004365213613611196,
220
+ "0.05": 0.027477556626217314,
221
+ "0.1": 0.11733827045407796,
222
+ "0.15": 0.3001030294329299,
223
+ "0.2": 0.6384553462390299,
224
+ "0.3": 2.076592921693635,
225
+ "0.4": 3.9639974417379102
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.00031195811733164547,
268
+ "0.01": 0.0011356036026244354,
269
+ "0.02": 0.004365213613611196,
270
+ "0.05": 0.027477556626217314,
271
+ "0.1": 0.11733827045407796,
272
+ "0.15": 0.3001030294329299,
273
+ "0.2": 0.6384553462390299,
274
+ "0.3": 2.076592921693635,
275
+ "0.4": 3.9639974417379102
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 61.2
284
+ }
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.033792030066251755,
8
+ "blocks.0.attention.w_out.weight": 0.026513146236538887,
9
+ "blocks.0.attention.w_q.weight": 0.033974722027778625,
10
+ "blocks.0.attention.w_v.weight": 0.030098984017968178,
11
+ "blocks.0.feed_forward.w1.weight": 0.03262019529938698,
12
+ "blocks.0.feed_forward.w2.weight": 0.018550856038928032,
13
+ "blocks.0.feed_forward.w3.weight": 0.03300027176737785,
14
+ "blocks.1.attention.w_k.weight": 0.028769828379154205,
15
+ "blocks.1.attention.w_out.weight": 0.024300970137119293,
16
+ "blocks.1.attention.w_q.weight": 0.0290773157030344,
17
+ "blocks.1.attention.w_v.weight": 0.02681984379887581,
18
+ "blocks.1.feed_forward.w1.weight": 0.035662610083818436,
19
+ "blocks.1.feed_forward.w2.weight": 0.019625242799520493,
20
+ "blocks.1.feed_forward.w3.weight": 0.036090947687625885,
21
+ "blocks.2.attention.w_k.weight": 0.031684815883636475,
22
+ "blocks.2.attention.w_out.weight": 0.0261231679469347,
23
+ "blocks.2.attention.w_q.weight": 0.03192731738090515,
24
+ "blocks.2.attention.w_v.weight": 0.029027849435806274,
25
+ "blocks.2.feed_forward.w1.weight": 0.035158202052116394,
26
+ "blocks.2.feed_forward.w2.weight": 0.017811624333262444,
27
+ "blocks.2.feed_forward.w3.weight": 0.035606566816568375,
28
+ "blocks.3.attention.w_k.weight": 0.03212268650531769,
29
+ "blocks.3.attention.w_out.weight": 0.02810906246304512,
30
+ "blocks.3.attention.w_q.weight": 0.03197857737541199,
31
+ "blocks.3.attention.w_v.weight": 0.03150033950805664,
32
+ "blocks.3.feed_forward.w1.weight": 0.03466830775141716,
33
+ "blocks.3.feed_forward.w2.weight": 0.018097171559929848,
34
+ "blocks.3.feed_forward.w3.weight": 0.03538702800869942,
35
+ "blocks.4.attention.w_k.weight": 0.03302117437124252,
36
+ "blocks.4.attention.w_out.weight": 0.02875415049493313,
37
+ "blocks.4.attention.w_q.weight": 0.032805606722831726,
38
+ "blocks.4.attention.w_v.weight": 0.031778451055288315,
39
+ "blocks.4.feed_forward.w1.weight": 0.0345308855175972,
40
+ "blocks.4.feed_forward.w2.weight": 0.019194750115275383,
41
+ "blocks.4.feed_forward.w3.weight": 0.03569754585623741,
42
+ "blocks.5.attention.w_k.weight": 0.03010912798345089,
43
+ "blocks.5.attention.w_out.weight": 0.03453635796904564,
44
+ "blocks.5.attention.w_q.weight": 0.030799981206655502,
45
+ "blocks.5.attention.w_v.weight": 0.03295043483376503,
46
+ "blocks.5.feed_forward.w1.weight": 0.03447722643613815,
47
+ "blocks.5.feed_forward.w2.weight": 0.020819514989852905,
48
+ "blocks.5.feed_forward.w3.weight": 0.03604985773563385,
49
+ "blocks.6.attention.w_k.weight": 0.03257782757282257,
50
+ "blocks.6.attention.w_out.weight": 0.036244895309209824,
51
+ "blocks.6.attention.w_q.weight": 0.0317949503660202,
52
+ "blocks.6.attention.w_v.weight": 0.034245118498802185,
53
+ "blocks.6.feed_forward.w1.weight": 0.03427771106362343,
54
+ "blocks.6.feed_forward.w2.weight": 0.02320433221757412,
55
+ "blocks.6.feed_forward.w3.weight": 0.036269668489694595,
56
+ "blocks.7.attention.w_k.weight": 0.033718906342983246,
57
+ "blocks.7.attention.w_out.weight": 0.040113773196935654,
58
+ "blocks.7.attention.w_q.weight": 0.0327695831656456,
59
+ "blocks.7.attention.w_v.weight": 0.034865327179431915,
60
+ "blocks.7.feed_forward.w1.weight": 0.033236388117074966,
61
+ "blocks.7.feed_forward.w2.weight": 0.02615497075021267,
62
+ "blocks.7.feed_forward.w3.weight": 0.03580132871866226,
63
+ "blocks.8.attention.w_k.weight": 0.03233204782009125,
64
+ "blocks.8.attention.w_out.weight": 0.04406481236219406,
65
+ "blocks.8.attention.w_q.weight": 0.031116899102926254,
66
+ "blocks.8.attention.w_v.weight": 0.03517533093690872,
67
+ "blocks.8.feed_forward.w1.weight": 0.03163241967558861,
68
+ "blocks.8.feed_forward.w2.weight": 0.030139151960611343,
69
+ "blocks.8.feed_forward.w3.weight": 0.03480131924152374
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": null,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.916410182665998
96
+ },
97
+ "base_loss": 3.916410182665998,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.9166031188216612,
101
+ 3.9166111077830292,
102
+ 3.9166282124544796
103
+ ],
104
+ "0.01": [
105
+ 3.9170695145428205,
106
+ 3.917076439352625,
107
+ 3.9171663953706153
108
+ ],
109
+ "0.02": [
110
+ 3.9188807005034314,
111
+ 3.918924248558763,
112
+ 3.919003539587189
113
+ ],
114
+ "0.05": [
115
+ 3.9309205856238885,
116
+ 3.9306903689526145,
117
+ 3.9308996703508847
118
+ ],
119
+ "0.1": [
120
+ 3.9758266704926046,
121
+ 3.973638188030323,
122
+ 3.974508139757904
123
+ ],
124
+ "0.15": [
125
+ 4.059841444233951,
126
+ 4.0517076565329395,
127
+ 4.055604896455549
128
+ ],
129
+ "0.2": [
130
+ 4.1971037213818345,
131
+ 4.177021170450288,
132
+ 4.188559100755623
133
+ ],
134
+ "0.3": [
135
+ 4.710997663686541,
136
+ 4.645984176817421,
137
+ 4.6946044558029065
138
+ ],
139
+ "0.4": [
140
+ 5.60428553083135,
141
+ 5.503434611767463,
142
+ 5.5852172258845725
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.9166031188216612,
149
+ 3.9166111077830292,
150
+ 3.9166282124544796
151
+ ],
152
+ "0.01": [
153
+ 3.9170695145428205,
154
+ 3.917076439352625,
155
+ 3.9171663953706153
156
+ ],
157
+ "0.02": [
158
+ 3.9188807005034314,
159
+ 3.918924248558763,
160
+ 3.919003539587189
161
+ ],
162
+ "0.05": [
163
+ 3.9309205856238885,
164
+ 3.9306903689526145,
165
+ 3.9308996703508847
166
+ ],
167
+ "0.1": [
168
+ 3.9758266704926046,
169
+ 3.973638188030323,
170
+ 3.974508139757904
171
+ ],
172
+ "0.15": [
173
+ 4.059841444233951,
174
+ 4.0517076565329395,
175
+ 4.055604896455549
176
+ ],
177
+ "0.2": [
178
+ 4.1971037213818345,
179
+ 4.177021170450288,
180
+ 4.188559100755623
181
+ ],
182
+ "0.3": [
183
+ 4.710997663686541,
184
+ 4.645984176817421,
185
+ 4.6946044558029065
186
+ ],
187
+ "0.4": [
188
+ 5.60428553083135,
189
+ 5.503434611767463,
190
+ 5.5852172258845725
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.00020396368705899448,
197
+ "0.01": 0.0006939337560227798,
198
+ "0.02": 0.0025259802171300194,
199
+ "0.05": 0.014426692309798206,
200
+ "0.1": 0.05824748342761277,
201
+ "0.15": 0.1393078164081487,
202
+ "0.2": 0.2711511481965845,
203
+ "0.3": 0.7674519161029583,
204
+ "0.4": 1.6479022734951307
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.00020396368705899448,
247
+ "0.01": 0.0006939337560227798,
248
+ "0.02": 0.0025259802171300194,
249
+ "0.05": 0.014426692309798206,
250
+ "0.1": 0.05824748342761277,
251
+ "0.15": 0.1393078164081487,
252
+ "0.2": 0.2711511481965845,
253
+ "0.3": 0.7674519161029583,
254
+ "0.4": 1.6479022734951307
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 52.6
263
+ }
pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.916410182665998
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.996126214447322
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 4.042027073674593
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.942661643683497
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.9220034376164308
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.9167275423957983
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 1.0797160317813246
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.12561689100859486
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.02625146101749909
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.0055932549504329465
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.00031735972980051486
42
+ }
43
+ },
44
+ "checks": {
45
+ "weights_restored": true
46
+ },
47
+ "ok": true
48
+ }
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": 6,
8
+ "a": 6,
9
+ "qkv": 6
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 4.116169639029774,
25
+ "heldout_cc_en_tail_4M": 4.678772911880955
26
+ },
27
+ "loss_bf16_master": 4.116169639029774,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 4.133484409991807
30
+ },
31
+ "bf16_minus_quant": -0.01731477096203271,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-p9-24",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 17.5
40
+ }
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/finalize.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
3
+ "w_bits": 6,
4
+ "a_bits": 6,
5
+ "qkv_bits": 6,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 4.1323714209140086
13
+ },
14
+ "losses_fakequant_master": {
15
+ "val_dolma_cc": 4.133484409991807
16
+ },
17
+ "losses_eval_model": {
18
+ "val_dolma_cc": 4.133484409991807
19
+ },
20
+ "max_distinct_values_per_row": 63,
21
+ "rel_quant_noise_per_layer": {
22
+ "blocks.0.attention.w_q": 0.031120842322707176,
23
+ "blocks.0.attention.w_k": 0.03135550022125244,
24
+ "blocks.0.attention.w_v": 0.03434007987380028,
25
+ "blocks.0.attention.w_out": 0.03542300686240196,
26
+ "blocks.0.feed_forward.w1": 0.031323470175266266,
27
+ "blocks.0.feed_forward.w2": 0.035301655530929565,
28
+ "blocks.0.feed_forward.w3": 0.03124716505408287,
29
+ "blocks.1.attention.w_q": 0.030890315771102905,
30
+ "blocks.1.attention.w_k": 0.03229247033596039,
31
+ "blocks.1.attention.w_v": 0.033395953476428986,
32
+ "blocks.1.attention.w_out": 0.032106492668390274,
33
+ "blocks.1.feed_forward.w1": 0.031076418235898018,
34
+ "blocks.1.feed_forward.w2": 0.034724216908216476,
35
+ "blocks.1.feed_forward.w3": 0.03132103011012077,
36
+ "blocks.2.attention.w_q": 0.031059447675943375,
37
+ "blocks.2.attention.w_k": 0.032405510544776917,
38
+ "blocks.2.attention.w_v": 0.033812735229730606,
39
+ "blocks.2.attention.w_out": 0.03228583559393883,
40
+ "blocks.2.feed_forward.w1": 0.03101040981709957,
41
+ "blocks.2.feed_forward.w2": 0.03550992161035538,
42
+ "blocks.2.feed_forward.w3": 0.03121618926525116,
43
+ "blocks.3.attention.w_q": 0.03107432834804058,
44
+ "blocks.3.attention.w_k": 0.03350537270307541,
45
+ "blocks.3.attention.w_v": 0.03315924108028412,
46
+ "blocks.3.attention.w_out": 0.03204391151666641,
47
+ "blocks.3.feed_forward.w1": 0.03128368780016899,
48
+ "blocks.3.feed_forward.w2": 0.03541150316596031,
49
+ "blocks.3.feed_forward.w3": 0.03160535916686058,
50
+ "blocks.4.attention.w_q": 0.031151646748185158,
51
+ "blocks.4.attention.w_k": 0.032543595880270004,
52
+ "blocks.4.attention.w_v": 0.033398281782865524,
53
+ "blocks.4.attention.w_out": 0.03176643326878548,
54
+ "blocks.4.feed_forward.w1": 0.031281664967536926,
55
+ "blocks.4.feed_forward.w2": 0.03540990874171257,
56
+ "blocks.4.feed_forward.w3": 0.031895920634269714,
57
+ "blocks.5.attention.w_q": 0.03117757849395275,
58
+ "blocks.5.attention.w_k": 0.033542435616254807,
59
+ "blocks.5.attention.w_v": 0.033953651785850525,
60
+ "blocks.5.attention.w_out": 0.032777994871139526,
61
+ "blocks.5.feed_forward.w1": 0.03133861720561981,
62
+ "blocks.5.feed_forward.w2": 0.03533758595585823,
63
+ "blocks.5.feed_forward.w3": 0.0314180962741375,
64
+ "blocks.6.attention.w_q": 0.031007658690214157,
65
+ "blocks.6.attention.w_k": 0.03384607285261154,
66
+ "blocks.6.attention.w_v": 0.03334997221827507,
67
+ "blocks.6.attention.w_out": 0.033462539315223694,
68
+ "blocks.6.feed_forward.w1": 0.03140699863433838,
69
+ "blocks.6.feed_forward.w2": 0.03598141297698021,
70
+ "blocks.6.feed_forward.w3": 0.03141026571393013,
71
+ "blocks.7.attention.w_q": 0.03154123201966286,
72
+ "blocks.7.attention.w_k": 0.031651951372623444,
73
+ "blocks.7.attention.w_v": 0.035259682685136795,
74
+ "blocks.7.attention.w_out": 0.03749556094408035,
75
+ "blocks.7.feed_forward.w1": 0.031442634761333466,
76
+ "blocks.7.feed_forward.w2": 0.03741942718625069,
77
+ "blocks.7.feed_forward.w3": 0.03139324113726616,
78
+ "blocks.8.attention.w_q": 0.030701495707035065,
79
+ "blocks.8.attention.w_k": 0.031224187463521957,
80
+ "blocks.8.attention.w_v": 0.03455720096826553,
81
+ "blocks.8.attention.w_out": 0.03540285304188728,
82
+ "blocks.8.feed_forward.w1": 0.031160781159996986,
83
+ "blocks.8.feed_forward.w2": 0.0417548343539238,
84
+ "blocks.8.feed_forward.w3": 0.03098127990961075
85
+ },
86
+ "rel_quant_noise_mean": 0.03286899629211615,
87
+ "n_quantized_params": 30670848,
88
+ "n_params": 82191872,
89
+ "param_dtypes": [
90
+ "torch.float32"
91
+ ],
92
+ "norm_gain_max_abs_dev_from_1": 0.7896115779876709,
93
+ "loss_eval_model": 4.133484409991807,
94
+ "checks": {
95
+ "snap_reproduces_fakequant": true,
96
+ "grid_levels_ok": true,
97
+ "master_weights_fp32": true,
98
+ "norm_gains_trained": true,
99
+ "loss_finite": true
100
+ },
101
+ "ok": true
102
+ }
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/launch.yaml ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 102
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: 6
71
+ fq_act_bits: 6
72
+ fq_qkv_bits: 6
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: olmo_core.optim.AdamWConfig
78
+ lr: 0.0024
79
+ betas:
80
+ - 0.9
81
+ - 0.95
82
+ eps: 1.0e-15
83
+ weight_decay: 0.1
84
+ fused: true
85
+ scheduler:
86
+ _CLASS_: olmo_core.optim.CosWithWarmup
87
+ warmup: 232
88
+ alpha_f: 0.1
89
+ compile_model: true
90
+ dp_config:
91
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
92
+ name: ddp
93
+ param_dtype: float32
94
+ reduce_dtype: float32
95
+ autocast_precision: bfloat16
96
+ max_grad_norm: 1.0
97
+ trainer:
98
+ _CLASS_: olmo_core.train.config.TrainerConfig
99
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/ckpt
100
+ save_overwrite: true
101
+ metrics_collect_interval: 10
102
+ cancel_check_interval: 5
103
+ callbacks:
104
+ config_saver:
105
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
106
+ checkpointer:
107
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
108
+ save_interval: 1002327
109
+ save_async: false
110
+ wandb:
111
+ enabled: false
112
+ init_seed: 102
113
+ n_tokens: 610009088
114
+ load_path: null
115
+ validation_datasets:
116
+ dolma_cc_val:
117
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
118
+ validation_eval_interval: 500
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03452238067984581,
8
+ "blocks.0.attention.w_out.weight": 0.039284028112888336,
9
+ "blocks.0.attention.w_q.weight": 0.037009578198194504,
10
+ "blocks.0.attention.w_v.weight": 0.045134395360946655,
11
+ "blocks.0.attention_norm.weight": 0.739876389503479,
12
+ "blocks.0.feed_forward.w1.weight": 0.03798346593976021,
13
+ "blocks.0.feed_forward.w2.weight": 0.0359223298728466,
14
+ "blocks.0.feed_forward.w3.weight": 0.040152616798877716,
15
+ "blocks.0.feed_forward_norm.weight": 0.7368800044059753,
16
+ "blocks.1.attention.w_k.weight": 0.02928074263036251,
17
+ "blocks.1.attention.w_out.weight": 0.03693598881363869,
18
+ "blocks.1.attention.w_q.weight": 0.029001660645008087,
19
+ "blocks.1.attention.w_v.weight": 0.05250340327620506,
20
+ "blocks.1.attention_norm.weight": 0.7101855874061584,
21
+ "blocks.1.feed_forward.w1.weight": 0.047833625227212906,
22
+ "blocks.1.feed_forward.w2.weight": 0.04353787750005722,
23
+ "blocks.1.feed_forward.w3.weight": 0.048920538276433945,
24
+ "blocks.1.feed_forward_norm.weight": 0.7491304874420166,
25
+ "blocks.2.attention.w_k.weight": 0.03276506066322327,
26
+ "blocks.2.attention.w_out.weight": 0.03439931944012642,
27
+ "blocks.2.attention.w_q.weight": 0.03276437148451805,
28
+ "blocks.2.attention.w_v.weight": 0.0495532788336277,
29
+ "blocks.2.attention_norm.weight": 0.7206140756607056,
30
+ "blocks.2.feed_forward.w1.weight": 0.04991117864847183,
31
+ "blocks.2.feed_forward.w2.weight": 0.04640195146203041,
32
+ "blocks.2.feed_forward.w3.weight": 0.050042614340782166,
33
+ "blocks.2.feed_forward_norm.weight": 0.7519209980964661,
34
+ "blocks.3.attention.w_k.weight": 0.03559667617082596,
35
+ "blocks.3.attention.w_out.weight": 0.03937024250626564,
36
+ "blocks.3.attention.w_q.weight": 0.03479573875665665,
37
+ "blocks.3.attention.w_v.weight": 0.054586127400398254,
38
+ "blocks.3.attention_norm.weight": 0.7257862687110901,
39
+ "blocks.3.feed_forward.w1.weight": 0.048882536590099335,
40
+ "blocks.3.feed_forward.w2.weight": 0.04395680129528046,
41
+ "blocks.3.feed_forward.w3.weight": 0.04914790019392967,
42
+ "blocks.3.feed_forward_norm.weight": 0.754949152469635,
43
+ "blocks.4.attention.w_k.weight": 0.03553074598312378,
44
+ "blocks.4.attention.w_out.weight": 0.041324224323034286,
45
+ "blocks.4.attention.w_q.weight": 0.03667189180850983,
46
+ "blocks.4.attention.w_v.weight": 0.05485207587480545,
47
+ "blocks.4.attention_norm.weight": 0.7308045625686646,
48
+ "blocks.4.feed_forward.w1.weight": 0.04888234660029411,
49
+ "blocks.4.feed_forward.w2.weight": 0.04575841501355171,
50
+ "blocks.4.feed_forward.w3.weight": 0.049392424523830414,
51
+ "blocks.4.feed_forward_norm.weight": 0.752315878868103,
52
+ "blocks.5.attention.w_k.weight": 0.036597561091184616,
53
+ "blocks.5.attention.w_out.weight": 0.04749889299273491,
54
+ "blocks.5.attention.w_q.weight": 0.0354626402258873,
55
+ "blocks.5.attention.w_v.weight": 0.04642144963145256,
56
+ "blocks.5.attention_norm.weight": 0.7264080047607422,
57
+ "blocks.5.feed_forward.w1.weight": 0.04904254153370857,
58
+ "blocks.5.feed_forward.w2.weight": 0.04696989804506302,
59
+ "blocks.5.feed_forward.w3.weight": 0.05032774806022644,
60
+ "blocks.5.feed_forward_norm.weight": 0.7507573366165161,
61
+ "blocks.6.attention.w_k.weight": 0.03965156897902489,
62
+ "blocks.6.attention.w_out.weight": 0.05916905403137207,
63
+ "blocks.6.attention.w_q.weight": 0.03646329045295715,
64
+ "blocks.6.attention.w_v.weight": 0.04659885913133621,
65
+ "blocks.6.attention_norm.weight": 0.7252739667892456,
66
+ "blocks.6.feed_forward.w1.weight": 0.048663850873708725,
67
+ "blocks.6.feed_forward.w2.weight": 0.0510810911655426,
68
+ "blocks.6.feed_forward.w3.weight": 0.050724003463983536,
69
+ "blocks.6.feed_forward_norm.weight": 0.7489078640937805,
70
+ "blocks.7.attention.w_k.weight": 0.03603087738156319,
71
+ "blocks.7.attention.w_out.weight": 0.08853258937597275,
72
+ "blocks.7.attention.w_q.weight": 0.03710123151540756,
73
+ "blocks.7.attention.w_v.weight": 0.05319391191005707,
74
+ "blocks.7.attention_norm.weight": 0.7369636297225952,
75
+ "blocks.7.feed_forward.w1.weight": 0.049033112823963165,
76
+ "blocks.7.feed_forward.w2.weight": 0.06086074933409691,
77
+ "blocks.7.feed_forward.w3.weight": 0.05256909877061844,
78
+ "blocks.7.feed_forward_norm.weight": 0.7491270899772644,
79
+ "blocks.8.attention.w_k.weight": 0.04031990468502045,
80
+ "blocks.8.attention.w_out.weight": 0.09576456993818283,
81
+ "blocks.8.attention.w_q.weight": 0.03576497733592987,
82
+ "blocks.8.attention.w_v.weight": 0.042381834238767624,
83
+ "blocks.8.attention_norm.weight": 0.7316818237304688,
84
+ "blocks.8.feed_forward.w1.weight": 0.047424208372831345,
85
+ "blocks.8.feed_forward.w2.weight": 0.06759675592184067,
86
+ "blocks.8.feed_forward.w3.weight": 0.05208192765712738,
87
+ "blocks.8.feed_forward_norm.weight": 0.7457408905029297,
88
+ "embeddings.weight": 0.05595573037862778,
89
+ "lm_head.norm.weight": 1.3673397302627563,
90
+ "lm_head.w_out.weight": 0.07094171643257141
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": 6,
115
+ "base_losses": {
116
+ "val_dolma_cc": 4.133484409991807
117
+ },
118
+ "base_loss": 4.133484409991807,
119
+ "loss": {
120
+ "0.005": [
121
+ 4.134091336316463,
122
+ 4.13444897731279,
123
+ 4.133487833273491
124
+ ],
125
+ "0.01": [
126
+ 4.13518646564468,
127
+ 4.134274726406633,
128
+ 4.133956944146065
129
+ ],
130
+ "0.02": [
131
+ 4.135846947744149,
132
+ 4.136405912914343,
133
+ 4.1364804895948515
134
+ ],
135
+ "0.05": [
136
+ 4.147796513000677,
137
+ 4.148100367395197,
138
+ 4.1488774366879895
139
+ ],
140
+ "0.1": [
141
+ 4.191778945611088,
142
+ 4.190513446400063,
143
+ 4.193277110606912
144
+ ],
145
+ "0.15": [
146
+ 4.268052713123267,
147
+ 4.266817600916623,
148
+ 4.270579134128668
149
+ ],
150
+ "0.2": [
151
+ 4.384039275048625,
152
+ 4.381098965594603,
153
+ 4.388253307414357
154
+ ],
155
+ "0.3": [
156
+ 4.788946718744993,
157
+ 4.775071755113937,
158
+ 4.790860837895414
159
+ ],
160
+ "0.4": [
161
+ 5.53872043004493,
162
+ 5.510531493315125,
163
+ 5.542292638174446
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 4.134091336316463,
170
+ 4.13444897731279,
171
+ 4.133487833273491
172
+ ],
173
+ "0.01": [
174
+ 4.13518646564468,
175
+ 4.134274726406633,
176
+ 4.133956944146065
177
+ ],
178
+ "0.02": [
179
+ 4.135846947744149,
180
+ 4.136405912914343,
181
+ 4.1364804895948515
182
+ ],
183
+ "0.05": [
184
+ 4.147796513000677,
185
+ 4.148100367395197,
186
+ 4.1488774366879895
187
+ ],
188
+ "0.1": [
189
+ 4.191778945611088,
190
+ 4.190513446400063,
191
+ 4.193277110606912
192
+ ],
193
+ "0.15": [
194
+ 4.268052713123267,
195
+ 4.266817600916623,
196
+ 4.270579134128668
197
+ ],
198
+ "0.2": [
199
+ 4.384039275048625,
200
+ 4.381098965594603,
201
+ 4.388253307414357
202
+ ],
203
+ "0.3": [
204
+ 4.788946718744993,
205
+ 4.775071755113937,
206
+ 4.790860837895414
207
+ ],
208
+ "0.4": [
209
+ 5.53872043004493,
210
+ 5.510531493315125,
211
+ 5.542292638174446
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0005249723091074188,
218
+ "0.01": 0.000988302073985814,
219
+ "0.02": 0.0027600400926406365,
220
+ "0.05": 0.014773695702814393,
221
+ "0.1": 0.05837209088088091,
222
+ "0.15": 0.13499873939771145,
223
+ "0.2": 0.2509794393607212,
224
+ "0.3": 0.6514753605929746,
225
+ "0.4": 1.3970304438530272
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0005249723091074188,
268
+ "0.01": 0.000988302073985814,
269
+ "0.02": 0.0027600400926406365,
270
+ "0.05": 0.014773695702814393,
271
+ "0.1": 0.05837209088088091,
272
+ "0.15": 0.13499873939771145,
273
+ "0.2": 0.2509794393607212,
274
+ "0.3": 0.6514753605929746,
275
+ "0.4": 1.3970304438530272
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 77.9
284
+ }
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03452238067984581,
8
+ "blocks.0.attention.w_out.weight": 0.039284028112888336,
9
+ "blocks.0.attention.w_q.weight": 0.037009578198194504,
10
+ "blocks.0.attention.w_v.weight": 0.045134395360946655,
11
+ "blocks.0.feed_forward.w1.weight": 0.03798346593976021,
12
+ "blocks.0.feed_forward.w2.weight": 0.0359223298728466,
13
+ "blocks.0.feed_forward.w3.weight": 0.040152616798877716,
14
+ "blocks.1.attention.w_k.weight": 0.02928074263036251,
15
+ "blocks.1.attention.w_out.weight": 0.03693598881363869,
16
+ "blocks.1.attention.w_q.weight": 0.029001660645008087,
17
+ "blocks.1.attention.w_v.weight": 0.05250340327620506,
18
+ "blocks.1.feed_forward.w1.weight": 0.047833625227212906,
19
+ "blocks.1.feed_forward.w2.weight": 0.04353787750005722,
20
+ "blocks.1.feed_forward.w3.weight": 0.048920538276433945,
21
+ "blocks.2.attention.w_k.weight": 0.03276506066322327,
22
+ "blocks.2.attention.w_out.weight": 0.03439931944012642,
23
+ "blocks.2.attention.w_q.weight": 0.03276437148451805,
24
+ "blocks.2.attention.w_v.weight": 0.0495532788336277,
25
+ "blocks.2.feed_forward.w1.weight": 0.04991117864847183,
26
+ "blocks.2.feed_forward.w2.weight": 0.04640195146203041,
27
+ "blocks.2.feed_forward.w3.weight": 0.050042614340782166,
28
+ "blocks.3.attention.w_k.weight": 0.03559667617082596,
29
+ "blocks.3.attention.w_out.weight": 0.03937024250626564,
30
+ "blocks.3.attention.w_q.weight": 0.03479573875665665,
31
+ "blocks.3.attention.w_v.weight": 0.054586127400398254,
32
+ "blocks.3.feed_forward.w1.weight": 0.048882536590099335,
33
+ "blocks.3.feed_forward.w2.weight": 0.04395680129528046,
34
+ "blocks.3.feed_forward.w3.weight": 0.04914790019392967,
35
+ "blocks.4.attention.w_k.weight": 0.03553074598312378,
36
+ "blocks.4.attention.w_out.weight": 0.041324224323034286,
37
+ "blocks.4.attention.w_q.weight": 0.03667189180850983,
38
+ "blocks.4.attention.w_v.weight": 0.05485207587480545,
39
+ "blocks.4.feed_forward.w1.weight": 0.04888234660029411,
40
+ "blocks.4.feed_forward.w2.weight": 0.04575841501355171,
41
+ "blocks.4.feed_forward.w3.weight": 0.049392424523830414,
42
+ "blocks.5.attention.w_k.weight": 0.036597561091184616,
43
+ "blocks.5.attention.w_out.weight": 0.04749889299273491,
44
+ "blocks.5.attention.w_q.weight": 0.0354626402258873,
45
+ "blocks.5.attention.w_v.weight": 0.04642144963145256,
46
+ "blocks.5.feed_forward.w1.weight": 0.04904254153370857,
47
+ "blocks.5.feed_forward.w2.weight": 0.04696989804506302,
48
+ "blocks.5.feed_forward.w3.weight": 0.05032774806022644,
49
+ "blocks.6.attention.w_k.weight": 0.03965156897902489,
50
+ "blocks.6.attention.w_out.weight": 0.05916905403137207,
51
+ "blocks.6.attention.w_q.weight": 0.03646329045295715,
52
+ "blocks.6.attention.w_v.weight": 0.04659885913133621,
53
+ "blocks.6.feed_forward.w1.weight": 0.048663850873708725,
54
+ "blocks.6.feed_forward.w2.weight": 0.0510810911655426,
55
+ "blocks.6.feed_forward.w3.weight": 0.050724003463983536,
56
+ "blocks.7.attention.w_k.weight": 0.03603087738156319,
57
+ "blocks.7.attention.w_out.weight": 0.08853258937597275,
58
+ "blocks.7.attention.w_q.weight": 0.03710123151540756,
59
+ "blocks.7.attention.w_v.weight": 0.05319391191005707,
60
+ "blocks.7.feed_forward.w1.weight": 0.049033112823963165,
61
+ "blocks.7.feed_forward.w2.weight": 0.06086074933409691,
62
+ "blocks.7.feed_forward.w3.weight": 0.05256909877061844,
63
+ "blocks.8.attention.w_k.weight": 0.04031990468502045,
64
+ "blocks.8.attention.w_out.weight": 0.09576456993818283,
65
+ "blocks.8.attention.w_q.weight": 0.03576497733592987,
66
+ "blocks.8.attention.w_v.weight": 0.042381834238767624,
67
+ "blocks.8.feed_forward.w1.weight": 0.047424208372831345,
68
+ "blocks.8.feed_forward.w2.weight": 0.06759675592184067,
69
+ "blocks.8.feed_forward.w3.weight": 0.05208192765712738
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": 6,
94
+ "base_losses": {
95
+ "val_dolma_cc": 4.133484409991807
96
+ },
97
+ "base_loss": 4.133484409991807,
98
+ "loss": {
99
+ "0.005": [
100
+ 4.133970637272798,
101
+ 4.133880180477601,
102
+ 4.133772411657893
103
+ ],
104
+ "0.01": [
105
+ 4.133476006324315,
106
+ 4.134386121948764,
107
+ 4.133500517021167
108
+ ],
109
+ "0.02": [
110
+ 4.134247136712427,
111
+ 4.134187356433435,
112
+ 4.134905647001647
113
+ ],
114
+ "0.05": [
115
+ 4.1378708036122065,
116
+ 4.138091039950253,
117
+ 4.138317696408576
118
+ ],
119
+ "0.1": [
120
+ 4.151370927679759,
121
+ 4.150401763380517,
122
+ 4.15023772363458
123
+ ],
124
+ "0.15": [
125
+ 4.172548873977413,
126
+ 4.172452908632948,
127
+ 4.172826934433491
128
+ ],
129
+ "0.2": [
130
+ 4.205900721983018,
131
+ 4.205130561850742,
132
+ 4.20530034097294
133
+ ],
134
+ "0.3": [
135
+ 4.312236227343587,
136
+ 4.312655600867546,
137
+ 4.315095485224318
138
+ ],
139
+ "0.4": [
140
+ 4.502774966904886,
141
+ 4.495064891492847,
142
+ 4.509433621449239
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 4.133970637272798,
149
+ 4.133880180477601,
150
+ 4.133772411657893
151
+ ],
152
+ "0.01": [
153
+ 4.133476006324315,
154
+ 4.134386121948764,
155
+ 4.133500517021167
156
+ ],
157
+ "0.02": [
158
+ 4.134247136712427,
159
+ 4.134187356433435,
160
+ 4.134905647001647
161
+ ],
162
+ "0.05": [
163
+ 4.1378708036122065,
164
+ 4.138091039950253,
165
+ 4.138317696408576
166
+ ],
167
+ "0.1": [
168
+ 4.151370927679759,
169
+ 4.150401763380517,
170
+ 4.15023772363458
171
+ ],
172
+ "0.15": [
173
+ 4.172548873977413,
174
+ 4.172452908632948,
175
+ 4.172826934433491
176
+ ],
177
+ "0.2": [
178
+ 4.205900721983018,
179
+ 4.205130561850742,
180
+ 4.20530034097294
181
+ ],
182
+ "0.3": [
183
+ 4.312236227343587,
184
+ 4.312655600867546,
185
+ 4.315095485224318
186
+ ],
187
+ "0.4": [
188
+ 4.502774966904886,
189
+ 4.495064891492847,
190
+ 4.509433621449239
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.00038999981095777514,
197
+ "0.01": 0.0003031384396088299,
198
+ "0.02": 0.0009623033906960288,
199
+ "0.05": 0.004608769998537987,
200
+ "0.1": 0.017185728239812015,
201
+ "0.15": 0.03912516235614394,
202
+ "0.2": 0.07195946494376049,
203
+ "0.3": 0.17984469448667717,
204
+ "0.4": 0.368940083290517
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.00038999981095777514,
247
+ "0.01": 0.0003031384396088299,
248
+ "0.02": 0.0009623033906960288,
249
+ "0.05": 0.004608769998537987,
250
+ "0.1": 0.017185728239812015,
251
+ "0.15": 0.03912516235614394,
252
+ "0.2": 0.07195946494376049,
253
+ "0.3": 0.17984469448667717,
254
+ "0.4": 0.368940083290517
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 72.2
263
+ }
pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102/ptq.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D0.61B-f6-adamw-lr2.4e-03-s102",
3
+ "w_bits": 6,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 4.133484409991807
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.402245652440087
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 4.17207222014817
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 4.144966006773763
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 4.133246677750118
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 4.134107232080439
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 0.2687612424482797
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.038587810156363034
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.01148159678195615
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": -0.00023773224168888163
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0006228220886317359
42
+ }
43
+ },
44
+ "own_grid_max_rel_weight_change": 2.2988398029610835e-07,
45
+ "own_grid_delta_loss": -0.00023773224168888163,
46
+ "checks": {
47
+ "weights_restored": true,
48
+ "own_grid_weights_unchanged": true,
49
+ "own_grid_is_fixed_point": true
50
+ },
51
+ "ok": true
52
+ }
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": null,
8
+ "a": null,
9
+ "qkv": null
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.831054171073506,
25
+ "heldout_cc_en_tail_4M": 4.413598668423799
26
+ },
27
+ "loss_bf16_master": 3.831054171073506,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.831054171073506
30
+ },
31
+ "bf16_minus_quant": 0.0,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-m5-16",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 17.8
40
+ }
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/finalize.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
3
+ "w_bits": null,
4
+ "a_bits": null,
5
+ "qkv_bits": null,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.831054171073506
13
+ },
14
+ "losses_eval_model": {
15
+ "val_dolma_cc": 3.831054171073506
16
+ },
17
+ "n_params": 82191872,
18
+ "param_dtypes": [
19
+ "torch.float32"
20
+ ],
21
+ "norm_gain_max_abs_dev_from_1": 1.0510008335113525,
22
+ "loss_eval_model": 3.831054171073506,
23
+ "checks": {
24
+ "master_weights_fp32": true,
25
+ "norm_gains_trained": true,
26
+ "loss_finite": true
27
+ },
28
+ "ok": true
29
+ }
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/launch.yaml ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: null
71
+ fq_act_bits: null
72
+ fq_qkv_bits: null
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: olmo_core.optim.AdamWConfig
78
+ lr: 0.0012
79
+ betas:
80
+ - 0.9
81
+ - 0.95
82
+ eps: 1.0e-15
83
+ weight_decay: 0.1
84
+ fused: true
85
+ scheduler:
86
+ _CLASS_: olmo_core.optim.CosWithWarmup
87
+ warmup: 701
88
+ alpha_f: 0.1
89
+ compile_model: true
90
+ dp_config:
91
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
92
+ name: ddp
93
+ param_dtype: float32
94
+ reduce_dtype: float32
95
+ autocast_precision: bfloat16
96
+ max_grad_norm: 1.0
97
+ trainer:
98
+ _CLASS_: olmo_core.train.config.TrainerConfig
99
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/ckpt
100
+ save_overwrite: true
101
+ metrics_collect_interval: 10
102
+ cancel_check_interval: 5
103
+ callbacks:
104
+ config_saver:
105
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
106
+ checkpointer:
107
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
108
+ save_interval: 1007019
109
+ save_async: false
110
+ wandb:
111
+ enabled: false
112
+ init_seed: 101
113
+ n_tokens: 1839988736
114
+ load_path: null
115
+ validation_datasets:
116
+ dolma_cc_val:
117
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
118
+ validation_eval_interval: 500
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03268071636557579,
8
+ "blocks.0.attention.w_out.weight": 0.03668426722288132,
9
+ "blocks.0.attention.w_q.weight": 0.03406849503517151,
10
+ "blocks.0.attention.w_v.weight": 0.033527445048093796,
11
+ "blocks.0.attention_norm.weight": 0.6293824911117554,
12
+ "blocks.0.feed_forward.w1.weight": 0.038742586970329285,
13
+ "blocks.0.feed_forward.w2.weight": 0.039205584675073624,
14
+ "blocks.0.feed_forward.w3.weight": 0.04028899595141411,
15
+ "blocks.0.feed_forward_norm.weight": 0.6330354809761047,
16
+ "blocks.1.attention.w_k.weight": 0.03544138744473457,
17
+ "blocks.1.attention.w_out.weight": 0.03620398789644241,
18
+ "blocks.1.attention.w_q.weight": 0.0365966372191906,
19
+ "blocks.1.attention.w_v.weight": 0.03734150528907776,
20
+ "blocks.1.attention_norm.weight": 0.6099925637245178,
21
+ "blocks.1.feed_forward.w1.weight": 0.04364994540810585,
22
+ "blocks.1.feed_forward.w2.weight": 0.04219818487763405,
23
+ "blocks.1.feed_forward.w3.weight": 0.04411840811371803,
24
+ "blocks.1.feed_forward_norm.weight": 0.643448531627655,
25
+ "blocks.2.attention.w_k.weight": 0.03736107051372528,
26
+ "blocks.2.attention.w_out.weight": 0.03443260118365288,
27
+ "blocks.2.attention.w_q.weight": 0.03904962167143822,
28
+ "blocks.2.attention.w_v.weight": 0.03684396296739578,
29
+ "blocks.2.attention_norm.weight": 0.6134346723556519,
30
+ "blocks.2.feed_forward.w1.weight": 0.04324464499950409,
31
+ "blocks.2.feed_forward.w2.weight": 0.040622010827064514,
32
+ "blocks.2.feed_forward.w3.weight": 0.04401577636599541,
33
+ "blocks.2.feed_forward_norm.weight": 0.6434590220451355,
34
+ "blocks.3.attention.w_k.weight": 0.03984769433736801,
35
+ "blocks.3.attention.w_out.weight": 0.03913891315460205,
36
+ "blocks.3.attention.w_q.weight": 0.042075786739587784,
37
+ "blocks.3.attention.w_v.weight": 0.039377178996801376,
38
+ "blocks.3.attention_norm.weight": 0.6263216137886047,
39
+ "blocks.3.feed_forward.w1.weight": 0.04268060624599457,
40
+ "blocks.3.feed_forward.w2.weight": 0.0402805358171463,
41
+ "blocks.3.feed_forward.w3.weight": 0.04456881061196327,
42
+ "blocks.3.feed_forward_norm.weight": 0.6389124989509583,
43
+ "blocks.4.attention.w_k.weight": 0.04043733328580856,
44
+ "blocks.4.attention.w_out.weight": 0.04046373814344406,
45
+ "blocks.4.attention.w_q.weight": 0.04414784163236618,
46
+ "blocks.4.attention.w_v.weight": 0.0379972942173481,
47
+ "blocks.4.attention_norm.weight": 0.6256927251815796,
48
+ "blocks.4.feed_forward.w1.weight": 0.04161463677883148,
49
+ "blocks.4.feed_forward.w2.weight": 0.040296878665685654,
50
+ "blocks.4.feed_forward.w3.weight": 0.04442773759365082,
51
+ "blocks.4.feed_forward_norm.weight": 0.6358168721199036,
52
+ "blocks.5.attention.w_k.weight": 0.03878093510866165,
53
+ "blocks.5.attention.w_out.weight": 0.04189108684659004,
54
+ "blocks.5.attention.w_q.weight": 0.041319891810417175,
55
+ "blocks.5.attention.w_v.weight": 0.0382239855825901,
56
+ "blocks.5.attention_norm.weight": 0.6185463070869446,
57
+ "blocks.5.feed_forward.w1.weight": 0.041318804025650024,
58
+ "blocks.5.feed_forward.w2.weight": 0.04324207454919815,
59
+ "blocks.5.feed_forward.w3.weight": 0.04525141790509224,
60
+ "blocks.5.feed_forward_norm.weight": 0.6307767033576965,
61
+ "blocks.6.attention.w_k.weight": 0.04055728018283844,
62
+ "blocks.6.attention.w_out.weight": 0.055124517530202866,
63
+ "blocks.6.attention.w_q.weight": 0.04400692880153656,
64
+ "blocks.6.attention.w_v.weight": 0.04032757133245468,
65
+ "blocks.6.attention_norm.weight": 0.6209803819656372,
66
+ "blocks.6.feed_forward.w1.weight": 0.04073719307780266,
67
+ "blocks.6.feed_forward.w2.weight": 0.04550210386514664,
68
+ "blocks.6.feed_forward.w3.weight": 0.04522354528307915,
69
+ "blocks.6.feed_forward_norm.weight": 0.6327573657035828,
70
+ "blocks.7.attention.w_k.weight": 0.039778005331754684,
71
+ "blocks.7.attention.w_out.weight": 0.057591959834098816,
72
+ "blocks.7.attention.w_q.weight": 0.04154594987630844,
73
+ "blocks.7.attention.w_v.weight": 0.03982347622513771,
74
+ "blocks.7.attention_norm.weight": 0.6144234538078308,
75
+ "blocks.7.feed_forward.w1.weight": 0.040671318769454956,
76
+ "blocks.7.feed_forward.w2.weight": 0.0517294704914093,
77
+ "blocks.7.feed_forward.w3.weight": 0.04585808888077736,
78
+ "blocks.7.feed_forward_norm.weight": 0.6315277814865112,
79
+ "blocks.8.attention.w_k.weight": 0.03729722276329994,
80
+ "blocks.8.attention.w_out.weight": 0.0786689817905426,
81
+ "blocks.8.attention.w_q.weight": 0.04076381400227547,
82
+ "blocks.8.attention.w_v.weight": 0.03802540898323059,
83
+ "blocks.8.attention_norm.weight": 0.6195935606956482,
84
+ "blocks.8.feed_forward.w1.weight": 0.038343265652656555,
85
+ "blocks.8.feed_forward.w2.weight": 0.05855748802423477,
86
+ "blocks.8.feed_forward.w3.weight": 0.044858478009700775,
87
+ "blocks.8.feed_forward_norm.weight": 0.6258807182312012,
88
+ "embeddings.weight": 0.045032817870378494,
89
+ "lm_head.norm.weight": 1.8185430765151978,
90
+ "lm_head.w_out.weight": 0.05665469169616699
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": null,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.831054171073506
117
+ },
118
+ "base_loss": 3.831054171073506,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.8317018403186744,
122
+ 3.8312902014551584,
123
+ 3.8314713380479004
124
+ ],
125
+ "0.01": [
126
+ 3.8330465366624002,
127
+ 3.8324081983050196,
128
+ 3.8327434412583328
129
+ ],
130
+ "0.02": [
131
+ 3.838123752438785,
132
+ 3.836907145514839,
133
+ 3.8374857075055484
134
+ ],
135
+ "0.05": [
136
+ 3.872347300979871,
137
+ 3.8698361849778355,
138
+ 3.871757166340911
139
+ ],
140
+ "0.1": [
141
+ 4.003700027599084,
142
+ 4.0011724876696135,
143
+ 4.013152979830948
144
+ ],
145
+ "0.15": [
146
+ 4.276705020410427,
147
+ 4.271452419154528,
148
+ 4.3165530950910815
149
+ ],
150
+ "0.2": [
151
+ 4.786053736649279,
152
+ 4.758407978870707,
153
+ 4.870210199041412
154
+ ],
155
+ "0.3": [
156
+ 6.652791131836611,
157
+ 6.553454359603605,
158
+ 6.76614946240315
159
+ ],
160
+ "0.4": [
161
+ 8.568021647722265,
162
+ 8.522721748528717,
163
+ 8.586911589912544
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.8317018403186744,
170
+ 3.8312902014551584,
171
+ 3.8314713380479004
172
+ ],
173
+ "0.01": [
174
+ 3.8330465366624002,
175
+ 3.8324081983050196,
176
+ 3.8327434412583328
177
+ ],
178
+ "0.02": [
179
+ 3.838123752438785,
180
+ 3.836907145514839,
181
+ 3.8374857075055484
182
+ ],
183
+ "0.05": [
184
+ 3.872347300979871,
185
+ 3.8698361849778355,
186
+ 3.871757166340911
187
+ ],
188
+ "0.1": [
189
+ 4.003700027599084,
190
+ 4.0011724876696135,
191
+ 4.013152979830948
192
+ ],
193
+ "0.15": [
194
+ 4.276705020410427,
195
+ 4.271452419154528,
196
+ 4.3165530950910815
197
+ ],
198
+ "0.2": [
199
+ 4.786053736649279,
200
+ 4.758407978870707,
201
+ 4.870210199041412
202
+ ],
203
+ "0.3": [
204
+ 6.652791131836611,
205
+ 6.553454359603605,
206
+ 6.76614946240315
207
+ ],
208
+ "0.4": [
209
+ 8.568021647722265,
210
+ 8.522721748528717,
211
+ 8.586911589912544
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0004336222004055479,
218
+ "0.01": 0.0016785543350783527,
219
+ "0.02": 0.0064513640795516025,
220
+ "0.05": 0.040259379692700215,
221
+ "0.1": 0.1749543272930425,
222
+ "0.15": 0.4571826738118401,
223
+ "0.2": 0.9738364671136268,
224
+ "0.3": 2.82641081354095,
225
+ "0.4": 4.728164157647669
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0004336222004055479,
268
+ "0.01": 0.0016785543350783527,
269
+ "0.02": 0.0064513640795516025,
270
+ "0.05": 0.040259379692700215,
271
+ "0.1": 0.1749543272930425,
272
+ "0.15": 0.4571826738118401,
273
+ "0.2": 0.9738364671136268,
274
+ "0.3": 2.82641081354095,
275
+ "0.4": 4.728164157647669
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 56.0
284
+ }
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.03268071636557579,
8
+ "blocks.0.attention.w_out.weight": 0.03668426722288132,
9
+ "blocks.0.attention.w_q.weight": 0.03406849503517151,
10
+ "blocks.0.attention.w_v.weight": 0.033527445048093796,
11
+ "blocks.0.feed_forward.w1.weight": 0.038742586970329285,
12
+ "blocks.0.feed_forward.w2.weight": 0.039205584675073624,
13
+ "blocks.0.feed_forward.w3.weight": 0.04028899595141411,
14
+ "blocks.1.attention.w_k.weight": 0.03544138744473457,
15
+ "blocks.1.attention.w_out.weight": 0.03620398789644241,
16
+ "blocks.1.attention.w_q.weight": 0.0365966372191906,
17
+ "blocks.1.attention.w_v.weight": 0.03734150528907776,
18
+ "blocks.1.feed_forward.w1.weight": 0.04364994540810585,
19
+ "blocks.1.feed_forward.w2.weight": 0.04219818487763405,
20
+ "blocks.1.feed_forward.w3.weight": 0.04411840811371803,
21
+ "blocks.2.attention.w_k.weight": 0.03736107051372528,
22
+ "blocks.2.attention.w_out.weight": 0.03443260118365288,
23
+ "blocks.2.attention.w_q.weight": 0.03904962167143822,
24
+ "blocks.2.attention.w_v.weight": 0.03684396296739578,
25
+ "blocks.2.feed_forward.w1.weight": 0.04324464499950409,
26
+ "blocks.2.feed_forward.w2.weight": 0.040622010827064514,
27
+ "blocks.2.feed_forward.w3.weight": 0.04401577636599541,
28
+ "blocks.3.attention.w_k.weight": 0.03984769433736801,
29
+ "blocks.3.attention.w_out.weight": 0.03913891315460205,
30
+ "blocks.3.attention.w_q.weight": 0.042075786739587784,
31
+ "blocks.3.attention.w_v.weight": 0.039377178996801376,
32
+ "blocks.3.feed_forward.w1.weight": 0.04268060624599457,
33
+ "blocks.3.feed_forward.w2.weight": 0.0402805358171463,
34
+ "blocks.3.feed_forward.w3.weight": 0.04456881061196327,
35
+ "blocks.4.attention.w_k.weight": 0.04043733328580856,
36
+ "blocks.4.attention.w_out.weight": 0.04046373814344406,
37
+ "blocks.4.attention.w_q.weight": 0.04414784163236618,
38
+ "blocks.4.attention.w_v.weight": 0.0379972942173481,
39
+ "blocks.4.feed_forward.w1.weight": 0.04161463677883148,
40
+ "blocks.4.feed_forward.w2.weight": 0.040296878665685654,
41
+ "blocks.4.feed_forward.w3.weight": 0.04442773759365082,
42
+ "blocks.5.attention.w_k.weight": 0.03878093510866165,
43
+ "blocks.5.attention.w_out.weight": 0.04189108684659004,
44
+ "blocks.5.attention.w_q.weight": 0.041319891810417175,
45
+ "blocks.5.attention.w_v.weight": 0.0382239855825901,
46
+ "blocks.5.feed_forward.w1.weight": 0.041318804025650024,
47
+ "blocks.5.feed_forward.w2.weight": 0.04324207454919815,
48
+ "blocks.5.feed_forward.w3.weight": 0.04525141790509224,
49
+ "blocks.6.attention.w_k.weight": 0.04055728018283844,
50
+ "blocks.6.attention.w_out.weight": 0.055124517530202866,
51
+ "blocks.6.attention.w_q.weight": 0.04400692880153656,
52
+ "blocks.6.attention.w_v.weight": 0.04032757133245468,
53
+ "blocks.6.feed_forward.w1.weight": 0.04073719307780266,
54
+ "blocks.6.feed_forward.w2.weight": 0.04550210386514664,
55
+ "blocks.6.feed_forward.w3.weight": 0.04522354528307915,
56
+ "blocks.7.attention.w_k.weight": 0.039778005331754684,
57
+ "blocks.7.attention.w_out.weight": 0.057591959834098816,
58
+ "blocks.7.attention.w_q.weight": 0.04154594987630844,
59
+ "blocks.7.attention.w_v.weight": 0.03982347622513771,
60
+ "blocks.7.feed_forward.w1.weight": 0.040671318769454956,
61
+ "blocks.7.feed_forward.w2.weight": 0.0517294704914093,
62
+ "blocks.7.feed_forward.w3.weight": 0.04585808888077736,
63
+ "blocks.8.attention.w_k.weight": 0.03729722276329994,
64
+ "blocks.8.attention.w_out.weight": 0.0786689817905426,
65
+ "blocks.8.attention.w_q.weight": 0.04076381400227547,
66
+ "blocks.8.attention.w_v.weight": 0.03802540898323059,
67
+ "blocks.8.feed_forward.w1.weight": 0.038343265652656555,
68
+ "blocks.8.feed_forward.w2.weight": 0.05855748802423477,
69
+ "blocks.8.feed_forward.w3.weight": 0.044858478009700775
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": null,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.831054171073506
96
+ },
97
+ "base_loss": 3.831054171073506,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.83126655146913,
101
+ 3.831416064763763,
102
+ 3.8311541387232984
103
+ ],
104
+ "0.01": [
105
+ 3.8320505338347615,
106
+ 3.8322378597207867,
107
+ 3.8318906489747087
108
+ ],
109
+ "0.02": [
110
+ 3.835079523889827,
111
+ 3.8354311594720008,
112
+ 3.8344159588806828
113
+ ],
114
+ "0.05": [
115
+ 3.8555864952406136,
116
+ 3.857775685833163,
117
+ 3.8541448554719424
118
+ ],
119
+ "0.1": [
120
+ 3.9345608681548763,
121
+ 3.946372234551774,
122
+ 3.9354457101796987
123
+ ],
124
+ "0.15": [
125
+ 4.090792315924315,
126
+ 4.127439078418456,
127
+ 4.105478190432802
128
+ ],
129
+ "0.2": [
130
+ 4.360869604782873,
131
+ 4.4485717888851,
132
+ 4.4065429116301
133
+ ],
134
+ "0.3": [
135
+ 5.350963750428164,
136
+ 5.582059019661014,
137
+ 5.417273765133487
138
+ ],
139
+ "0.4": [
140
+ 6.654905765310262,
141
+ 6.920882303084387,
142
+ 6.549217831386812
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.83126655146913,
149
+ 3.831416064763763,
150
+ 3.8311541387232984
151
+ ],
152
+ "0.01": [
153
+ 3.8320505338347615,
154
+ 3.8322378597207867,
155
+ 3.8318906489747087
156
+ ],
157
+ "0.02": [
158
+ 3.835079523889827,
159
+ 3.8354311594720008,
160
+ 3.8344159588806828
161
+ ],
162
+ "0.05": [
163
+ 3.8555864952406136,
164
+ 3.857775685833163,
165
+ 3.8541448554719424
166
+ ],
167
+ "0.1": [
168
+ 3.9345608681548763,
169
+ 3.946372234551774,
170
+ 3.9354457101796987
171
+ ],
172
+ "0.15": [
173
+ 4.090792315924315,
174
+ 4.127439078418456,
175
+ 4.105478190432802
176
+ ],
177
+ "0.2": [
178
+ 4.360869604782873,
179
+ 4.4485717888851,
180
+ 4.4065429116301
181
+ ],
182
+ "0.3": [
183
+ 5.350963750428164,
184
+ 5.582059019661014,
185
+ 5.417273765133487
186
+ ],
187
+ "0.4": [
188
+ 6.654905765310262,
189
+ 6.920882303084387,
190
+ 6.549217831386812
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.000224747245224588,
197
+ "0.01": 0.0010055097699126492,
198
+ "0.02": 0.003921376340664473,
199
+ "0.05": 0.024781507775066824,
200
+ "0.1": 0.10773876655527692,
201
+ "0.15": 0.2768490238516841,
202
+ "0.2": 0.574273930692518,
203
+ "0.3": 1.6190446740007158,
204
+ "0.4": 2.8772811288536473
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.000224747245224588,
247
+ "0.01": 0.0010055097699126492,
248
+ "0.02": 0.003921376340664473,
249
+ "0.05": 0.024781507775066824,
250
+ "0.1": 0.10773876655527692,
251
+ "0.15": 0.2768490238516841,
252
+ "0.2": 0.574273930692518,
253
+ "0.3": 1.6190446740007158,
254
+ "0.4": 2.8772811288536473
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 49.8
263
+ }
pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101/ptq.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-adamw-lr1.2e-03-s101",
3
+ "w_bits": null,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.831054171073506
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 6.164653690965879
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 4.078234822788718
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.8779693310580523
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.839041501003307
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.8320119113927316
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 2.3335995198923727
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.2471806517152122
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.04691515998454632
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.007987329929801223
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0009577403192255751
42
+ }
43
+ },
44
+ "checks": {
45
+ "weights_restored": true
46
+ },
47
+ "ok": true
48
+ }
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": null,
8
+ "a": null,
9
+ "qkv": null
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.741282098723469,
25
+ "heldout_cc_en_tail_4M": 4.324322303835364
26
+ },
27
+ "loss_bf16_master": 3.741282098723469,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.741282098723469
30
+ },
31
+ "bf16_minus_quant": 0.0,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-t5-28",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 17.7
40
+ }
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "a_bits": null,
5
+ "qkv_bits": null,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.741282098723469
13
+ },
14
+ "losses_eval_model": {
15
+ "val_dolma_cc": 3.741282098723469
16
+ },
17
+ "n_params": 82191872,
18
+ "param_dtypes": [
19
+ "torch.float32"
20
+ ],
21
+ "norm_gain_max_abs_dev_from_1": 0.7421518564224243,
22
+ "loss_eval_model": 3.741282098723469,
23
+ "checks": {
24
+ "master_weights_fp32": true,
25
+ "norm_gains_trained": true,
26
+ "loss_finite": true
27
+ },
28
+ "ok": true
29
+ }
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/launch.yaml ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: null
71
+ fq_act_bits: null
72
+ fq_qkv_bits: null
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: precq.muon26.MuonConfig
78
+ lr: 0.01
79
+ weight_decay: 0.1
80
+ momentum: 0.95
81
+ nesterov: true
82
+ ns_steps: 5
83
+ eps: 1.0e-07
84
+ adjust_lr_fn: null
85
+ adamw_lr: 0.0024
86
+ adamw_betas:
87
+ - 0.9
88
+ - 0.95
89
+ adamw_eps: 1.0e-15
90
+ adamw_weight_decay: 0.1
91
+ adamw_fused: true
92
+ muon_on_lm_head: false
93
+ scheduler:
94
+ _CLASS_: olmo_core.optim.CosWithWarmup
95
+ warmup: 701
96
+ alpha_f: 0.1
97
+ compile_model: true
98
+ dp_config:
99
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
100
+ name: ddp
101
+ param_dtype: float32
102
+ reduce_dtype: float32
103
+ autocast_precision: bfloat16
104
+ max_grad_norm: 1.0
105
+ trainer:
106
+ _CLASS_: olmo_core.train.config.TrainerConfig
107
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ckpt
108
+ save_overwrite: true
109
+ metrics_collect_interval: 10
110
+ cancel_check_interval: 5
111
+ callbacks:
112
+ config_saver:
113
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
114
+ checkpointer:
115
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
116
+ save_interval: 1007019
117
+ save_async: false
118
+ wandb:
119
+ enabled: false
120
+ init_seed: 101
121
+ n_tokens: 1839988736
122
+ load_path: null
123
+ validation_datasets:
124
+ dolma_cc_val:
125
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
126
+ validation_eval_interval: 500
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.035397958010435104,
8
+ "blocks.0.attention.w_out.weight": 0.02164395898580551,
9
+ "blocks.0.attention.w_q.weight": 0.035328783094882965,
10
+ "blocks.0.attention.w_v.weight": 0.025519702583551407,
11
+ "blocks.0.attention_norm.weight": 0.38304951786994934,
12
+ "blocks.0.feed_forward.w1.weight": 0.030326496809720993,
13
+ "blocks.0.feed_forward.w2.weight": 0.01632818393409252,
14
+ "blocks.0.feed_forward.w3.weight": 0.030313409864902496,
15
+ "blocks.0.feed_forward_norm.weight": 0.39804232120513916,
16
+ "blocks.1.attention.w_k.weight": 0.027827803045511246,
17
+ "blocks.1.attention.w_out.weight": 0.01929481327533722,
18
+ "blocks.1.attention.w_q.weight": 0.028433021157979965,
19
+ "blocks.1.attention.w_v.weight": 0.021994709968566895,
20
+ "blocks.1.attention_norm.weight": 0.3855082094669342,
21
+ "blocks.1.feed_forward.w1.weight": 0.030362948775291443,
22
+ "blocks.1.feed_forward.w2.weight": 0.01613524742424488,
23
+ "blocks.1.feed_forward.w3.weight": 0.029716597869992256,
24
+ "blocks.1.feed_forward_norm.weight": 0.40586504340171814,
25
+ "blocks.2.attention.w_k.weight": 0.03229168802499771,
26
+ "blocks.2.attention.w_out.weight": 0.0226726196706295,
27
+ "blocks.2.attention.w_q.weight": 0.03159981593489647,
28
+ "blocks.2.attention.w_v.weight": 0.0259097870439291,
29
+ "blocks.2.attention_norm.weight": 0.3970188796520233,
30
+ "blocks.2.feed_forward.w1.weight": 0.029754923656582832,
31
+ "blocks.2.feed_forward.w2.weight": 0.01480945385992527,
32
+ "blocks.2.feed_forward.w3.weight": 0.02957199513912201,
33
+ "blocks.2.feed_forward_norm.weight": 0.40933650732040405,
34
+ "blocks.3.attention.w_k.weight": 0.032474976032972336,
35
+ "blocks.3.attention.w_out.weight": 0.02344447560608387,
36
+ "blocks.3.attention.w_q.weight": 0.03168802335858345,
37
+ "blocks.3.attention.w_v.weight": 0.026256276294589043,
38
+ "blocks.3.attention_norm.weight": 0.3984137773513794,
39
+ "blocks.3.feed_forward.w1.weight": 0.02918386459350586,
40
+ "blocks.3.feed_forward.w2.weight": 0.014518745243549347,
41
+ "blocks.3.feed_forward.w3.weight": 0.029343916103243828,
42
+ "blocks.3.feed_forward_norm.weight": 0.40580126643180847,
43
+ "blocks.4.attention.w_k.weight": 0.03286627680063248,
44
+ "blocks.4.attention.w_out.weight": 0.024900637567043304,
45
+ "blocks.4.attention.w_q.weight": 0.03240432217717171,
46
+ "blocks.4.attention.w_v.weight": 0.027380263432860374,
47
+ "blocks.4.attention_norm.weight": 0.3966514468193054,
48
+ "blocks.4.feed_forward.w1.weight": 0.02874669060111046,
49
+ "blocks.4.feed_forward.w2.weight": 0.01469415333122015,
50
+ "blocks.4.feed_forward.w3.weight": 0.029311303049325943,
51
+ "blocks.4.feed_forward_norm.weight": 0.40535956621170044,
52
+ "blocks.5.attention.w_k.weight": 0.03163629397749901,
53
+ "blocks.5.attention.w_out.weight": 0.026556408032774925,
54
+ "blocks.5.attention.w_q.weight": 0.0323525033891201,
55
+ "blocks.5.attention.w_v.weight": 0.027816232293844223,
56
+ "blocks.5.attention_norm.weight": 0.4023590683937073,
57
+ "blocks.5.feed_forward.w1.weight": 0.028784623369574547,
58
+ "blocks.5.feed_forward.w2.weight": 0.015200715512037277,
59
+ "blocks.5.feed_forward.w3.weight": 0.029435208067297935,
60
+ "blocks.5.feed_forward_norm.weight": 0.40602362155914307,
61
+ "blocks.6.attention.w_k.weight": 0.03339715674519539,
62
+ "blocks.6.attention.w_out.weight": 0.029964886605739594,
63
+ "blocks.6.attention.w_q.weight": 0.0327397957444191,
64
+ "blocks.6.attention.w_v.weight": 0.029403777793049812,
65
+ "blocks.6.attention_norm.weight": 0.40344297885894775,
66
+ "blocks.6.feed_forward.w1.weight": 0.02904530242085457,
67
+ "blocks.6.feed_forward.w2.weight": 0.017573032528162003,
68
+ "blocks.6.feed_forward.w3.weight": 0.02994123101234436,
69
+ "blocks.6.feed_forward_norm.weight": 0.408926784992218,
70
+ "blocks.7.attention.w_k.weight": 0.032194577157497406,
71
+ "blocks.7.attention.w_out.weight": 0.03255749121308327,
72
+ "blocks.7.attention.w_q.weight": 0.03139836713671684,
73
+ "blocks.7.attention.w_v.weight": 0.029072798788547516,
74
+ "blocks.7.attention_norm.weight": 0.40457212924957275,
75
+ "blocks.7.feed_forward.w1.weight": 0.028424231335520744,
76
+ "blocks.7.feed_forward.w2.weight": 0.02051093615591526,
77
+ "blocks.7.feed_forward.w3.weight": 0.029400411993265152,
78
+ "blocks.7.feed_forward_norm.weight": 0.4079188406467438,
79
+ "blocks.8.attention.w_k.weight": 0.030158329755067825,
80
+ "blocks.8.attention.w_out.weight": 0.038990020751953125,
81
+ "blocks.8.attention.w_q.weight": 0.029624801129102707,
82
+ "blocks.8.attention.w_v.weight": 0.02889910154044628,
83
+ "blocks.8.attention_norm.weight": 0.4107125699520111,
84
+ "blocks.8.feed_forward.w1.weight": 0.026936113834381104,
85
+ "blocks.8.feed_forward.w2.weight": 0.025364842265844345,
86
+ "blocks.8.feed_forward.w3.weight": 0.028375402092933655,
87
+ "blocks.8.feed_forward_norm.weight": 0.4011843800544739,
88
+ "embeddings.weight": 0.06977110356092453,
89
+ "lm_head.norm.weight": 1.3798266649246216,
90
+ "lm_head.w_out.weight": 0.07359108328819275
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": null,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.741282098723469
117
+ },
118
+ "base_loss": 3.741282098723469,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.7416292781709712,
122
+ 3.741794487701162,
123
+ 3.742089562845286
124
+ ],
125
+ "0.01": [
126
+ 3.74275158248727,
127
+ 3.7429588612151057,
128
+ 3.7436641863484628
129
+ ],
130
+ "0.02": [
131
+ 3.7473024344720507,
132
+ 3.747717246228647,
133
+ 3.749264851236619
134
+ ],
135
+ "0.05": [
136
+ 3.780543043428401,
137
+ 3.781887751509092,
138
+ 3.785293220390546
139
+ ],
140
+ "0.1": [
141
+ 3.916245888619137,
142
+ 3.919793292627658,
143
+ 3.924784930274286
144
+ ],
145
+ "0.15": [
146
+ 4.211604527414047,
147
+ 4.2234978373936345,
148
+ 4.22084451631432
149
+ ],
150
+ "0.2": [
151
+ 4.7966840943394145,
152
+ 4.855740092999335,
153
+ 4.809550477027847
154
+ ],
155
+ "0.3": [
156
+ 7.265636298785715,
157
+ 7.430777746750457,
158
+ 7.120484006803559
159
+ ],
160
+ "0.4": [
161
+ 9.54127430772559,
162
+ 9.48668455317653,
163
+ 9.123880935086254
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.7416292781709712,
170
+ 3.741794487701162,
171
+ 3.742089562845286
172
+ ],
173
+ "0.01": [
174
+ 3.74275158248727,
175
+ 3.7429588612151057,
176
+ 3.7436641863484628
177
+ ],
178
+ "0.02": [
179
+ 3.7473024344720507,
180
+ 3.747717246228647,
181
+ 3.749264851236619
182
+ ],
183
+ "0.05": [
184
+ 3.780543043428401,
185
+ 3.781887751509092,
186
+ 3.785293220390546
187
+ ],
188
+ "0.1": [
189
+ 3.916245888619137,
190
+ 3.919793292627658,
191
+ 3.924784930274286
192
+ ],
193
+ "0.15": [
194
+ 4.211604527414047,
195
+ 4.2234978373936345,
196
+ 4.22084451631432
197
+ ],
198
+ "0.2": [
199
+ 4.7966840943394145,
200
+ 4.855740092999335,
201
+ 4.809550477027847
202
+ ],
203
+ "0.3": [
204
+ 7.265636298785715,
205
+ 7.430777746750457,
206
+ 7.120484006803559
207
+ ],
208
+ "0.4": [
209
+ 9.54127430772559,
210
+ 9.48668455317653,
211
+ 9.123880935086254
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0005556775156705385,
218
+ "0.01": 0.00184277796014376,
219
+ "0.02": 0.0068127452556367984,
220
+ "0.05": 0.04129257305254397,
221
+ "0.1": 0.17899260511689175,
222
+ "0.15": 0.47736686165053177,
223
+ "0.2": 1.079376122732063,
224
+ "0.3": 3.5310172520564413,
225
+ "0.4": 5.642664499939322
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0005556775156705385,
268
+ "0.01": 0.00184277796014376,
269
+ "0.02": 0.0068127452556367984,
270
+ "0.05": 0.04129257305254397,
271
+ "0.1": 0.17899260511689175,
272
+ "0.15": 0.47736686165053177,
273
+ "0.2": 1.079376122732063,
274
+ "0.3": 3.5310172520564413,
275
+ "0.4": 5.642664499939322
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 56.5
284
+ }
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.035397958010435104,
8
+ "blocks.0.attention.w_out.weight": 0.02164395898580551,
9
+ "blocks.0.attention.w_q.weight": 0.035328783094882965,
10
+ "blocks.0.attention.w_v.weight": 0.025519702583551407,
11
+ "blocks.0.feed_forward.w1.weight": 0.030326496809720993,
12
+ "blocks.0.feed_forward.w2.weight": 0.01632818393409252,
13
+ "blocks.0.feed_forward.w3.weight": 0.030313409864902496,
14
+ "blocks.1.attention.w_k.weight": 0.027827803045511246,
15
+ "blocks.1.attention.w_out.weight": 0.01929481327533722,
16
+ "blocks.1.attention.w_q.weight": 0.028433021157979965,
17
+ "blocks.1.attention.w_v.weight": 0.021994709968566895,
18
+ "blocks.1.feed_forward.w1.weight": 0.030362948775291443,
19
+ "blocks.1.feed_forward.w2.weight": 0.01613524742424488,
20
+ "blocks.1.feed_forward.w3.weight": 0.029716597869992256,
21
+ "blocks.2.attention.w_k.weight": 0.03229168802499771,
22
+ "blocks.2.attention.w_out.weight": 0.0226726196706295,
23
+ "blocks.2.attention.w_q.weight": 0.03159981593489647,
24
+ "blocks.2.attention.w_v.weight": 0.0259097870439291,
25
+ "blocks.2.feed_forward.w1.weight": 0.029754923656582832,
26
+ "blocks.2.feed_forward.w2.weight": 0.01480945385992527,
27
+ "blocks.2.feed_forward.w3.weight": 0.02957199513912201,
28
+ "blocks.3.attention.w_k.weight": 0.032474976032972336,
29
+ "blocks.3.attention.w_out.weight": 0.02344447560608387,
30
+ "blocks.3.attention.w_q.weight": 0.03168802335858345,
31
+ "blocks.3.attention.w_v.weight": 0.026256276294589043,
32
+ "blocks.3.feed_forward.w1.weight": 0.02918386459350586,
33
+ "blocks.3.feed_forward.w2.weight": 0.014518745243549347,
34
+ "blocks.3.feed_forward.w3.weight": 0.029343916103243828,
35
+ "blocks.4.attention.w_k.weight": 0.03286627680063248,
36
+ "blocks.4.attention.w_out.weight": 0.024900637567043304,
37
+ "blocks.4.attention.w_q.weight": 0.03240432217717171,
38
+ "blocks.4.attention.w_v.weight": 0.027380263432860374,
39
+ "blocks.4.feed_forward.w1.weight": 0.02874669060111046,
40
+ "blocks.4.feed_forward.w2.weight": 0.01469415333122015,
41
+ "blocks.4.feed_forward.w3.weight": 0.029311303049325943,
42
+ "blocks.5.attention.w_k.weight": 0.03163629397749901,
43
+ "blocks.5.attention.w_out.weight": 0.026556408032774925,
44
+ "blocks.5.attention.w_q.weight": 0.0323525033891201,
45
+ "blocks.5.attention.w_v.weight": 0.027816232293844223,
46
+ "blocks.5.feed_forward.w1.weight": 0.028784623369574547,
47
+ "blocks.5.feed_forward.w2.weight": 0.015200715512037277,
48
+ "blocks.5.feed_forward.w3.weight": 0.029435208067297935,
49
+ "blocks.6.attention.w_k.weight": 0.03339715674519539,
50
+ "blocks.6.attention.w_out.weight": 0.029964886605739594,
51
+ "blocks.6.attention.w_q.weight": 0.0327397957444191,
52
+ "blocks.6.attention.w_v.weight": 0.029403777793049812,
53
+ "blocks.6.feed_forward.w1.weight": 0.02904530242085457,
54
+ "blocks.6.feed_forward.w2.weight": 0.017573032528162003,
55
+ "blocks.6.feed_forward.w3.weight": 0.02994123101234436,
56
+ "blocks.7.attention.w_k.weight": 0.032194577157497406,
57
+ "blocks.7.attention.w_out.weight": 0.03255749121308327,
58
+ "blocks.7.attention.w_q.weight": 0.03139836713671684,
59
+ "blocks.7.attention.w_v.weight": 0.029072798788547516,
60
+ "blocks.7.feed_forward.w1.weight": 0.028424231335520744,
61
+ "blocks.7.feed_forward.w2.weight": 0.02051093615591526,
62
+ "blocks.7.feed_forward.w3.weight": 0.029400411993265152,
63
+ "blocks.8.attention.w_k.weight": 0.030158329755067825,
64
+ "blocks.8.attention.w_out.weight": 0.038990020751953125,
65
+ "blocks.8.attention.w_q.weight": 0.029624801129102707,
66
+ "blocks.8.attention.w_v.weight": 0.02889910154044628,
67
+ "blocks.8.feed_forward.w1.weight": 0.026936113834381104,
68
+ "blocks.8.feed_forward.w2.weight": 0.025364842265844345,
69
+ "blocks.8.feed_forward.w3.weight": 0.028375402092933655
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": null,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.741282098723469
96
+ },
97
+ "base_loss": 3.741282098723469,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.741544666384729,
101
+ 3.741646793636223,
102
+ 3.7415041934144293
103
+ ],
104
+ "0.01": [
105
+ 3.742238622310576,
106
+ 3.7425577856207837,
107
+ 3.7421255522946613
108
+ ],
109
+ "0.02": [
110
+ 3.7449042593793065,
111
+ 3.74560978012876,
112
+ 3.7447914123657418
113
+ ],
114
+ "0.05": [
115
+ 3.764693616993574,
116
+ 3.7656580669448987,
117
+ 3.7637789552510497
118
+ ],
119
+ "0.1": [
120
+ 3.8448218197141966,
121
+ 3.842812259470448,
122
+ 3.8400778085714027
123
+ ],
124
+ "0.15": [
125
+ 4.014115305391468,
126
+ 3.9956352023966195,
127
+ 3.998548066912348
128
+ ],
129
+ "0.2": [
130
+ 4.336877310147849,
131
+ 4.268974573538131,
132
+ 4.296562694990125
133
+ ],
134
+ "0.3": [
135
+ 5.680166579468835,
136
+ 5.429263905080685,
137
+ 5.506133190586791
138
+ ],
139
+ "0.4": [
140
+ 7.38970396117274,
141
+ 7.121290946674239,
142
+ 7.185266885769214
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.741544666384729,
149
+ 3.741646793636223,
150
+ 3.7415041934144293
151
+ ],
152
+ "0.01": [
153
+ 3.742238622310576,
154
+ 3.7425577856207837,
155
+ 3.7421255522946613
156
+ ],
157
+ "0.02": [
158
+ 3.7449042593793065,
159
+ 3.74560978012876,
160
+ 3.7447914123657418
161
+ ],
162
+ "0.05": [
163
+ 3.764693616993574,
164
+ 3.7656580669448987,
165
+ 3.7637789552510497
166
+ ],
167
+ "0.1": [
168
+ 3.8448218197141966,
169
+ 3.842812259470448,
170
+ 3.8400778085714027
171
+ ],
172
+ "0.15": [
173
+ 4.014115305391468,
174
+ 3.9956352023966195,
175
+ 3.998548066912348
176
+ ],
177
+ "0.2": [
178
+ 4.336877310147849,
179
+ 4.268974573538131,
180
+ 4.296562694990125
181
+ ],
182
+ "0.3": [
183
+ 5.680166579468835,
184
+ 5.429263905080685,
185
+ 5.506133190586791
186
+ ],
187
+ "0.4": [
188
+ 7.38970396117274,
189
+ 7.121290946674239,
190
+ 7.185266885769214
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.0002831190883254564,
197
+ "0.01": 0.0010252213518713305,
198
+ "0.02": 0.0038197185678003898,
199
+ "0.05": 0.02342811433970482,
200
+ "0.1": 0.10128853052854714,
201
+ "0.15": 0.2614840928433435,
202
+ "0.2": 0.5595227608352329,
203
+ "0.3": 1.7972391263219678,
204
+ "0.4": 3.4908051658152615
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.0002831190883254564,
247
+ "0.01": 0.0010252213518713305,
248
+ "0.02": 0.0038197185678003898,
249
+ "0.05": 0.02342811433970482,
250
+ "0.1": 0.10128853052854714,
251
+ "0.15": 0.2614840928433435,
252
+ "0.2": 0.5595227608352329,
253
+ "0.3": 1.7972391263219678,
254
+ "0.4": 3.4908051658152615
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 49.9
263
+ }
pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr1.0e-02-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.741282098723469
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 5.78544710766109
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 3.9773186148965696
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.7802833630607204
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.751075317329996
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.7417057523276123
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 2.0441650089376213
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.23603651617310062
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.039001264337251484
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.009793218606527265
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.00042365360414331477
42
+ }
43
+ },
44
+ "checks": {
45
+ "weights_restored": true
46
+ },
47
+ "ok": true
48
+ }
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": null,
8
+ "a": null,
9
+ "qkv": null
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.7980115686420324,
25
+ "heldout_cc_en_tail_4M": 4.380696555852657
26
+ },
27
+ "loss_bf16_master": 3.7980115686420324,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.7980115686420324
30
+ },
31
+ "bf16_minus_quant": 0.0,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-q9-28",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 17.9
40
+ }
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "a_bits": null,
5
+ "qkv_bits": null,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.7980115686420324
13
+ },
14
+ "losses_eval_model": {
15
+ "val_dolma_cc": 3.7980115686420324
16
+ },
17
+ "n_params": 82191872,
18
+ "param_dtypes": [
19
+ "torch.float32"
20
+ ],
21
+ "norm_gain_max_abs_dev_from_1": 0.8346741199493408,
22
+ "loss_eval_model": 3.7980115686420324,
23
+ "checks": {
24
+ "master_weights_fp32": true,
25
+ "norm_gains_trained": true,
26
+ "loss_finite": true
27
+ },
28
+ "ok": true
29
+ }
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/launch.yaml ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: null
71
+ fq_act_bits: null
72
+ fq_qkv_bits: null
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: precq.muon26.MuonConfig
78
+ lr: 0.0025
79
+ weight_decay: 0.1
80
+ momentum: 0.95
81
+ nesterov: true
82
+ ns_steps: 5
83
+ eps: 1.0e-07
84
+ adjust_lr_fn: null
85
+ adamw_lr: 0.0024
86
+ adamw_betas:
87
+ - 0.9
88
+ - 0.95
89
+ adamw_eps: 1.0e-15
90
+ adamw_weight_decay: 0.1
91
+ adamw_fused: true
92
+ muon_on_lm_head: false
93
+ scheduler:
94
+ _CLASS_: olmo_core.optim.CosWithWarmup
95
+ warmup: 701
96
+ alpha_f: 0.1
97
+ compile_model: true
98
+ dp_config:
99
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
100
+ name: ddp
101
+ param_dtype: float32
102
+ reduce_dtype: float32
103
+ autocast_precision: bfloat16
104
+ max_grad_norm: 1.0
105
+ trainer:
106
+ _CLASS_: olmo_core.train.config.TrainerConfig
107
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/ckpt
108
+ save_overwrite: true
109
+ metrics_collect_interval: 10
110
+ cancel_check_interval: 5
111
+ callbacks:
112
+ config_saver:
113
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
114
+ checkpointer:
115
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
116
+ save_interval: 1007019
117
+ save_async: false
118
+ wandb:
119
+ enabled: false
120
+ init_seed: 101
121
+ n_tokens: 1839988736
122
+ load_path: null
123
+ validation_datasets:
124
+ dolma_cc_val:
125
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
126
+ validation_eval_interval: 500
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.02823960967361927,
8
+ "blocks.0.attention.w_out.weight": 0.012285185977816582,
9
+ "blocks.0.attention.w_q.weight": 0.027905184775590897,
10
+ "blocks.0.attention.w_v.weight": 0.015026269480586052,
11
+ "blocks.0.attention_norm.weight": 0.36402034759521484,
12
+ "blocks.0.feed_forward.w1.weight": 0.018776489421725273,
13
+ "blocks.0.feed_forward.w2.weight": 0.011734351515769958,
14
+ "blocks.0.feed_forward.w3.weight": 0.01938738487660885,
15
+ "blocks.0.feed_forward_norm.weight": 0.40471863746643066,
16
+ "blocks.1.attention.w_k.weight": 0.021475251764059067,
17
+ "blocks.1.attention.w_out.weight": 0.013774644583463669,
18
+ "blocks.1.attention.w_q.weight": 0.02127489447593689,
19
+ "blocks.1.attention.w_v.weight": 0.01492380816489458,
20
+ "blocks.1.attention_norm.weight": 0.3935880959033966,
21
+ "blocks.1.feed_forward.w1.weight": 0.019992079585790634,
22
+ "blocks.1.feed_forward.w2.weight": 0.012382809072732925,
23
+ "blocks.1.feed_forward.w3.weight": 0.020290696993470192,
24
+ "blocks.1.feed_forward_norm.weight": 0.4071131944656372,
25
+ "blocks.2.attention.w_k.weight": 0.025532985106110573,
26
+ "blocks.2.attention.w_out.weight": 0.015362584963440895,
27
+ "blocks.2.attention.w_q.weight": 0.024703780189156532,
28
+ "blocks.2.attention.w_v.weight": 0.017002055421471596,
29
+ "blocks.2.attention_norm.weight": 0.404275506734848,
30
+ "blocks.2.feed_forward.w1.weight": 0.01942611113190651,
31
+ "blocks.2.feed_forward.w2.weight": 0.012278757989406586,
32
+ "blocks.2.feed_forward.w3.weight": 0.019887568429112434,
33
+ "blocks.2.feed_forward_norm.weight": 0.40656808018684387,
34
+ "blocks.3.attention.w_k.weight": 0.02614908292889595,
35
+ "blocks.3.attention.w_out.weight": 0.015408954583108425,
36
+ "blocks.3.attention.w_q.weight": 0.025211982429027557,
37
+ "blocks.3.attention.w_v.weight": 0.01707778498530388,
38
+ "blocks.3.attention_norm.weight": 0.40307778120040894,
39
+ "blocks.3.feed_forward.w1.weight": 0.018848182633519173,
40
+ "blocks.3.feed_forward.w2.weight": 0.011943705379962921,
41
+ "blocks.3.feed_forward.w3.weight": 0.01950244978070259,
42
+ "blocks.3.feed_forward_norm.weight": 0.4055529236793518,
43
+ "blocks.4.attention.w_k.weight": 0.026735495775938034,
44
+ "blocks.4.attention.w_out.weight": 0.016208117827773094,
45
+ "blocks.4.attention.w_q.weight": 0.025898175314068794,
46
+ "blocks.4.attention.w_v.weight": 0.017795512452721596,
47
+ "blocks.4.attention_norm.weight": 0.39835861325263977,
48
+ "blocks.4.feed_forward.w1.weight": 0.018612002953886986,
49
+ "blocks.4.feed_forward.w2.weight": 0.011872970499098301,
50
+ "blocks.4.feed_forward.w3.weight": 0.019427184015512466,
51
+ "blocks.4.feed_forward_norm.weight": 0.40461936593055725,
52
+ "blocks.5.attention.w_k.weight": 0.026231853291392326,
53
+ "blocks.5.attention.w_out.weight": 0.01922769658267498,
54
+ "blocks.5.attention.w_q.weight": 0.025986338034272194,
55
+ "blocks.5.attention.w_v.weight": 0.019273919984698296,
56
+ "blocks.5.attention_norm.weight": 0.4032162129878998,
57
+ "blocks.5.feed_forward.w1.weight": 0.018274476751685143,
58
+ "blocks.5.feed_forward.w2.weight": 0.011750449426472187,
59
+ "blocks.5.feed_forward.w3.weight": 0.019116736948490143,
60
+ "blocks.5.feed_forward_norm.weight": 0.40511777997016907,
61
+ "blocks.6.attention.w_k.weight": 0.027327269315719604,
62
+ "blocks.6.attention.w_out.weight": 0.017924083396792412,
63
+ "blocks.6.attention.w_q.weight": 0.026071004569530487,
64
+ "blocks.6.attention.w_v.weight": 0.019368091598153114,
65
+ "blocks.6.attention_norm.weight": 0.4013967216014862,
66
+ "blocks.6.feed_forward.w1.weight": 0.018336961045861244,
67
+ "blocks.6.feed_forward.w2.weight": 0.012955289334058762,
68
+ "blocks.6.feed_forward.w3.weight": 0.01925334520637989,
69
+ "blocks.6.feed_forward_norm.weight": 0.4079210162162781,
70
+ "blocks.7.attention.w_k.weight": 0.02898509055376053,
71
+ "blocks.7.attention.w_out.weight": 0.020589452236890793,
72
+ "blocks.7.attention.w_q.weight": 0.027352629229426384,
73
+ "blocks.7.attention.w_v.weight": 0.020123351365327835,
74
+ "blocks.7.attention_norm.weight": 0.4034270644187927,
75
+ "blocks.7.feed_forward.w1.weight": 0.01795170083642006,
76
+ "blocks.7.feed_forward.w2.weight": 0.01468669530004263,
77
+ "blocks.7.feed_forward.w3.weight": 0.019164539873600006,
78
+ "blocks.7.feed_forward_norm.weight": 0.40853628516197205,
79
+ "blocks.8.attention.w_k.weight": 0.02681003138422966,
80
+ "blocks.8.attention.w_out.weight": 0.025461845099925995,
81
+ "blocks.8.attention.w_q.weight": 0.025192417204380035,
82
+ "blocks.8.attention.w_v.weight": 0.0201339740306139,
83
+ "blocks.8.attention_norm.weight": 0.40670523047447205,
84
+ "blocks.8.feed_forward.w1.weight": 0.016974635422229767,
85
+ "blocks.8.feed_forward.w2.weight": 0.017902221530675888,
86
+ "blocks.8.feed_forward.w3.weight": 0.01865983195602894,
87
+ "blocks.8.feed_forward_norm.weight": 0.4039677381515503,
88
+ "embeddings.weight": 0.07120760530233383,
89
+ "lm_head.norm.weight": 1.4134982824325562,
90
+ "lm_head.w_out.weight": 0.0739491730928421
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": null,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.7980115686420324
117
+ },
118
+ "base_loss": 3.7980115686420324,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.798565494741589,
122
+ 3.798943397653158,
123
+ 3.7988173856763194
124
+ ],
125
+ "0.01": [
126
+ 3.8004938782177438,
127
+ 3.801210975527769,
128
+ 3.800666517247268
129
+ ],
130
+ "0.02": [
131
+ 3.808603229556021,
132
+ 3.8101174585456627,
133
+ 3.808464905502089
134
+ ],
135
+ "0.05": [
136
+ 3.869149998923329,
137
+ 3.873003010046592,
138
+ 3.8638678557245845
139
+ ],
140
+ "0.1": [
141
+ 4.129703283871569,
142
+ 4.14031559402067,
143
+ 4.10362852287375
144
+ ],
145
+ "0.15": [
146
+ 4.732539682837764,
147
+ 4.720558259543812,
148
+ 4.678776374996494
149
+ ],
150
+ "0.2": [
151
+ 5.877093123527715,
152
+ 5.69483182430382,
153
+ 5.735007019957474
154
+ ],
155
+ "0.3": [
156
+ 8.204661368819409,
157
+ 8.114015678896147,
158
+ 8.297994659586678
159
+ ],
160
+ "0.4": [
161
+ 9.525585677721825,
162
+ 9.636142527577801,
163
+ 10.106714495897632
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.798565494741589,
170
+ 3.798943397653158,
171
+ 3.7988173856763194
172
+ ],
173
+ "0.01": [
174
+ 3.8004938782177438,
175
+ 3.801210975527769,
176
+ 3.800666517247268
177
+ ],
178
+ "0.02": [
179
+ 3.808603229556021,
180
+ 3.8101174585456627,
181
+ 3.808464905502089
182
+ ],
183
+ "0.05": [
184
+ 3.869149998923329,
185
+ 3.873003010046592,
186
+ 3.8638678557245845
187
+ ],
188
+ "0.1": [
189
+ 4.129703283871569,
190
+ 4.14031559402067,
191
+ 4.10362852287375
192
+ ],
193
+ "0.15": [
194
+ 4.732539682837764,
195
+ 4.720558259543812,
196
+ 4.678776374996494
197
+ ],
198
+ "0.2": [
199
+ 5.877093123527715,
200
+ 5.69483182430382,
201
+ 5.735007019957474
202
+ ],
203
+ "0.3": [
204
+ 8.204661368819409,
205
+ 8.114015678896147,
206
+ 8.297994659586678
207
+ ],
208
+ "0.4": [
209
+ 9.525585677721825,
210
+ 9.636142527577801,
211
+ 10.106714495897632
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0007638573816564076,
218
+ "0.01": 0.002778888355561726,
219
+ "0.02": 0.011050295892558903,
220
+ "0.05": 0.07066205292280348,
221
+ "0.1": 0.32653756494663044,
222
+ "0.15": 0.9126132038173247,
223
+ "0.2": 1.9709657539543035,
224
+ "0.3": 4.407545667125378,
225
+ "0.4": 5.958135998423721
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0007638573816564076,
268
+ "0.01": 0.002778888355561726,
269
+ "0.02": 0.011050295892558903,
270
+ "0.05": 0.07066205292280348,
271
+ "0.1": 0.32653756494663044,
272
+ "0.15": 0.9126132038173247,
273
+ "0.2": 1.9709657539543035,
274
+ "0.3": 4.407545667125378,
275
+ "0.4": 5.958135998423721
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 57.6
284
+ }
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.02823960967361927,
8
+ "blocks.0.attention.w_out.weight": 0.012285185977816582,
9
+ "blocks.0.attention.w_q.weight": 0.027905184775590897,
10
+ "blocks.0.attention.w_v.weight": 0.015026269480586052,
11
+ "blocks.0.feed_forward.w1.weight": 0.018776489421725273,
12
+ "blocks.0.feed_forward.w2.weight": 0.011734351515769958,
13
+ "blocks.0.feed_forward.w3.weight": 0.01938738487660885,
14
+ "blocks.1.attention.w_k.weight": 0.021475251764059067,
15
+ "blocks.1.attention.w_out.weight": 0.013774644583463669,
16
+ "blocks.1.attention.w_q.weight": 0.02127489447593689,
17
+ "blocks.1.attention.w_v.weight": 0.01492380816489458,
18
+ "blocks.1.feed_forward.w1.weight": 0.019992079585790634,
19
+ "blocks.1.feed_forward.w2.weight": 0.012382809072732925,
20
+ "blocks.1.feed_forward.w3.weight": 0.020290696993470192,
21
+ "blocks.2.attention.w_k.weight": 0.025532985106110573,
22
+ "blocks.2.attention.w_out.weight": 0.015362584963440895,
23
+ "blocks.2.attention.w_q.weight": 0.024703780189156532,
24
+ "blocks.2.attention.w_v.weight": 0.017002055421471596,
25
+ "blocks.2.feed_forward.w1.weight": 0.01942611113190651,
26
+ "blocks.2.feed_forward.w2.weight": 0.012278757989406586,
27
+ "blocks.2.feed_forward.w3.weight": 0.019887568429112434,
28
+ "blocks.3.attention.w_k.weight": 0.02614908292889595,
29
+ "blocks.3.attention.w_out.weight": 0.015408954583108425,
30
+ "blocks.3.attention.w_q.weight": 0.025211982429027557,
31
+ "blocks.3.attention.w_v.weight": 0.01707778498530388,
32
+ "blocks.3.feed_forward.w1.weight": 0.018848182633519173,
33
+ "blocks.3.feed_forward.w2.weight": 0.011943705379962921,
34
+ "blocks.3.feed_forward.w3.weight": 0.01950244978070259,
35
+ "blocks.4.attention.w_k.weight": 0.026735495775938034,
36
+ "blocks.4.attention.w_out.weight": 0.016208117827773094,
37
+ "blocks.4.attention.w_q.weight": 0.025898175314068794,
38
+ "blocks.4.attention.w_v.weight": 0.017795512452721596,
39
+ "blocks.4.feed_forward.w1.weight": 0.018612002953886986,
40
+ "blocks.4.feed_forward.w2.weight": 0.011872970499098301,
41
+ "blocks.4.feed_forward.w3.weight": 0.019427184015512466,
42
+ "blocks.5.attention.w_k.weight": 0.026231853291392326,
43
+ "blocks.5.attention.w_out.weight": 0.01922769658267498,
44
+ "blocks.5.attention.w_q.weight": 0.025986338034272194,
45
+ "blocks.5.attention.w_v.weight": 0.019273919984698296,
46
+ "blocks.5.feed_forward.w1.weight": 0.018274476751685143,
47
+ "blocks.5.feed_forward.w2.weight": 0.011750449426472187,
48
+ "blocks.5.feed_forward.w3.weight": 0.019116736948490143,
49
+ "blocks.6.attention.w_k.weight": 0.027327269315719604,
50
+ "blocks.6.attention.w_out.weight": 0.017924083396792412,
51
+ "blocks.6.attention.w_q.weight": 0.026071004569530487,
52
+ "blocks.6.attention.w_v.weight": 0.019368091598153114,
53
+ "blocks.6.feed_forward.w1.weight": 0.018336961045861244,
54
+ "blocks.6.feed_forward.w2.weight": 0.012955289334058762,
55
+ "blocks.6.feed_forward.w3.weight": 0.01925334520637989,
56
+ "blocks.7.attention.w_k.weight": 0.02898509055376053,
57
+ "blocks.7.attention.w_out.weight": 0.020589452236890793,
58
+ "blocks.7.attention.w_q.weight": 0.027352629229426384,
59
+ "blocks.7.attention.w_v.weight": 0.020123351365327835,
60
+ "blocks.7.feed_forward.w1.weight": 0.01795170083642006,
61
+ "blocks.7.feed_forward.w2.weight": 0.01468669530004263,
62
+ "blocks.7.feed_forward.w3.weight": 0.019164539873600006,
63
+ "blocks.8.attention.w_k.weight": 0.02681003138422966,
64
+ "blocks.8.attention.w_out.weight": 0.025461845099925995,
65
+ "blocks.8.attention.w_q.weight": 0.025192417204380035,
66
+ "blocks.8.attention.w_v.weight": 0.0201339740306139,
67
+ "blocks.8.feed_forward.w1.weight": 0.016974635422229767,
68
+ "blocks.8.feed_forward.w2.weight": 0.017902221530675888,
69
+ "blocks.8.feed_forward.w3.weight": 0.01865983195602894
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": null,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.7980115686420324
96
+ },
97
+ "base_loss": 3.7980115686420324,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.798558264770674,
101
+ 3.7982190821091835,
102
+ 3.798555240545827
103
+ ],
104
+ "0.01": [
105
+ 3.7999863209576263,
106
+ 3.7992850724637135,
107
+ 3.7998101021535424
108
+ ],
109
+ "0.02": [
110
+ 3.8054275391657852,
111
+ 3.8040605073761977,
112
+ 3.8053264134687033
113
+ ],
114
+ "0.05": [
115
+ 3.8446402723276525,
116
+ 3.8436552592122974,
117
+ 3.846435186941383
118
+ ],
119
+ "0.1": [
120
+ 4.01551527501933,
121
+ 4.025724475166709,
122
+ 4.028125819240211
123
+ ],
124
+ "0.15": [
125
+ 4.412840005031559,
126
+ 4.434557264906385,
127
+ 4.445230928087892
128
+ ],
129
+ "0.2": [
130
+ 5.227415800885474,
131
+ 5.085496949077468,
132
+ 5.149504562330737
133
+ ],
134
+ "0.3": [
135
+ 7.651915984545514,
136
+ 6.825729054930752,
137
+ 6.897205173366565
138
+ ],
139
+ "0.4": [
140
+ 8.87147847208868,
141
+ 8.323283079088746,
142
+ 8.34723364098581
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.798558264770674,
149
+ 3.7982190821091835,
150
+ 3.798555240545827
151
+ ],
152
+ "0.01": [
153
+ 3.7999863209576263,
154
+ 3.7992850724637135,
155
+ 3.7998101021535424
156
+ ],
157
+ "0.02": [
158
+ 3.8054275391657852,
159
+ 3.8040605073761977,
160
+ 3.8053264134687033
161
+ ],
162
+ "0.05": [
163
+ 3.8446402723276525,
164
+ 3.8436552592122974,
165
+ 3.846435186941383
166
+ ],
167
+ "0.1": [
168
+ 4.01551527501933,
169
+ 4.025724475166709,
170
+ 4.028125819240211
171
+ ],
172
+ "0.15": [
173
+ 4.412840005031559,
174
+ 4.434557264906385,
175
+ 4.445230928087892
176
+ ],
177
+ "0.2": [
178
+ 5.227415800885474,
179
+ 5.085496949077468,
180
+ 5.149504562330737
181
+ ],
182
+ "0.3": [
183
+ 7.651915984545514,
184
+ 6.825729054930752,
185
+ 6.897205173366565
186
+ ],
187
+ "0.4": [
188
+ 8.87147847208868,
189
+ 8.323283079088746,
190
+ 8.34723364098581
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.0004326271665293646,
197
+ "0.01": 0.001682263216261859,
198
+ "0.02": 0.00692658469486318,
199
+ "0.05": 0.04689867085174537,
200
+ "0.1": 0.22511028783338416,
201
+ "0.15": 0.6328644973665791,
202
+ "0.2": 1.3561275354558604,
203
+ "0.3": 3.3269385023055773,
204
+ "0.4": 4.715986828745715
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.0004326271665293646,
247
+ "0.01": 0.001682263216261859,
248
+ "0.02": 0.00692658469486318,
249
+ "0.05": 0.04689867085174537,
250
+ "0.1": 0.22511028783338416,
251
+ "0.15": 0.6328644973665791,
252
+ "0.2": 1.3561275354558604,
253
+ "0.3": 3.3269385023055773,
254
+ "0.4": 4.715986828745715
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 51.6
263
+ }
pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-bf16-muon-lr2.5e-03-alr2.4e-03-s101",
3
+ "w_bits": null,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.7980115686420324
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 7.490522650528621
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 4.375064463328336
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.8832731915557948
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.8167199790950557
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.7995107586662
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 3.6925110818865883
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.5770528946863034
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.08526162291376238
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.018708410453023294
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0014991900241674294
42
+ }
43
+ },
44
+ "checks": {
45
+ "weights_restored": true
46
+ },
47
+ "ok": true
48
+ }
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": 6,
8
+ "a": 6,
9
+ "qkv": 6
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.872419299440231,
25
+ "heldout_cc_en_tail_4M": 4.45661268858849
26
+ },
27
+ "loss_bf16_master": 3.872419299440231,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.904266601452955
30
+ },
31
+ "bf16_minus_quant": -0.03184730201272368,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-q9-24",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 18.6
40
+ }
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
3
+ "w_bits": 6,
4
+ "a_bits": 6,
5
+ "qkv_bits": 6,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.902572777325409
13
+ },
14
+ "losses_fakequant_master": {
15
+ "val_dolma_cc": 3.904266601452955
16
+ },
17
+ "losses_eval_model": {
18
+ "val_dolma_cc": 3.904266601452955
19
+ },
20
+ "max_distinct_values_per_row": 62,
21
+ "rel_quant_noise_per_layer": {
22
+ "blocks.0.attention.w_q": 0.03132954612374306,
23
+ "blocks.0.attention.w_k": 0.032565560191869736,
24
+ "blocks.0.attention.w_v": 0.03376292809844017,
25
+ "blocks.0.attention.w_out": 0.035142719745635986,
26
+ "blocks.0.feed_forward.w1": 0.030101409181952477,
27
+ "blocks.0.feed_forward.w2": 0.03631531447172165,
28
+ "blocks.0.feed_forward.w3": 0.030498016625642776,
29
+ "blocks.1.attention.w_q": 0.03087511472404003,
30
+ "blocks.1.attention.w_k": 0.03758403658866882,
31
+ "blocks.1.attention.w_v": 0.031966060400009155,
32
+ "blocks.1.attention.w_out": 0.03447512537240982,
33
+ "blocks.1.feed_forward.w1": 0.03064761683344841,
34
+ "blocks.1.feed_forward.w2": 0.036250922828912735,
35
+ "blocks.1.feed_forward.w3": 0.03089948371052742,
36
+ "blocks.2.attention.w_q": 0.030825156718492508,
37
+ "blocks.2.attention.w_k": 0.03836624696850777,
38
+ "blocks.2.attention.w_v": 0.03155285865068436,
39
+ "blocks.2.attention.w_out": 0.03362602740526199,
40
+ "blocks.2.feed_forward.w1": 0.030612805858254433,
41
+ "blocks.2.feed_forward.w2": 0.036123599857091904,
42
+ "blocks.2.feed_forward.w3": 0.030936317518353462,
43
+ "blocks.3.attention.w_q": 0.03071513958275318,
44
+ "blocks.3.attention.w_k": 0.03686222434043884,
45
+ "blocks.3.attention.w_v": 0.03170269727706909,
46
+ "blocks.3.attention.w_out": 0.03152976930141449,
47
+ "blocks.3.feed_forward.w1": 0.030767356976866722,
48
+ "blocks.3.feed_forward.w2": 0.036432571709156036,
49
+ "blocks.3.feed_forward.w3": 0.031014615669846535,
50
+ "blocks.4.attention.w_q": 0.031317088752985,
51
+ "blocks.4.attention.w_k": 0.04460396245121956,
52
+ "blocks.4.attention.w_v": 0.032061345875263214,
53
+ "blocks.4.attention.w_out": 0.03140087053179741,
54
+ "blocks.4.feed_forward.w1": 0.03076915629208088,
55
+ "blocks.4.feed_forward.w2": 0.0370493121445179,
56
+ "blocks.4.feed_forward.w3": 0.031176608055830002,
57
+ "blocks.5.attention.w_q": 0.03135261312127113,
58
+ "blocks.5.attention.w_k": 0.04125480353832245,
59
+ "blocks.5.attention.w_v": 0.032418251037597656,
60
+ "blocks.5.attention.w_out": 0.03230922296643257,
61
+ "blocks.5.feed_forward.w1": 0.031060615554451942,
62
+ "blocks.5.feed_forward.w2": 0.037558216601610184,
63
+ "blocks.5.feed_forward.w3": 0.031182829290628433,
64
+ "blocks.6.attention.w_q": 0.031420376151800156,
65
+ "blocks.6.attention.w_k": 0.040330782532691956,
66
+ "blocks.6.attention.w_v": 0.0323045440018177,
67
+ "blocks.6.attention.w_out": 0.037702176719903946,
68
+ "blocks.6.feed_forward.w1": 0.030968982726335526,
69
+ "blocks.6.feed_forward.w2": 0.03673739731311798,
70
+ "blocks.6.feed_forward.w3": 0.031195146963000298,
71
+ "blocks.7.attention.w_q": 0.03101927787065506,
72
+ "blocks.7.attention.w_k": 0.04204888641834259,
73
+ "blocks.7.attention.w_v": 0.032408490777015686,
74
+ "blocks.7.attention.w_out": 0.03582685440778732,
75
+ "blocks.7.feed_forward.w1": 0.03105190210044384,
76
+ "blocks.7.feed_forward.w2": 0.03869321197271347,
77
+ "blocks.7.feed_forward.w3": 0.03149225935339928,
78
+ "blocks.8.attention.w_q": 0.03125115856528282,
79
+ "blocks.8.attention.w_k": 0.033950477838516235,
80
+ "blocks.8.attention.w_v": 0.03320341184735298,
81
+ "blocks.8.attention.w_out": 0.03774721175432205,
82
+ "blocks.8.feed_forward.w1": 0.031206922605633736,
83
+ "blocks.8.feed_forward.w2": 0.04580705240368843,
84
+ "blocks.8.feed_forward.w3": 0.03194732218980789
85
+ },
86
+ "rel_quant_noise_mean": 0.03376682516601351,
87
+ "n_quantized_params": 30670848,
88
+ "n_params": 82191872,
89
+ "param_dtypes": [
90
+ "torch.float32"
91
+ ],
92
+ "norm_gain_max_abs_dev_from_1": 0.8170008659362793,
93
+ "loss_eval_model": 3.904266601452955,
94
+ "checks": {
95
+ "snap_reproduces_fakequant": true,
96
+ "grid_levels_ok": true,
97
+ "master_weights_fp32": true,
98
+ "norm_gains_trained": true,
99
+ "loss_finite": true
100
+ },
101
+ "ok": true
102
+ }
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/launch.yaml ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: 6
71
+ fq_act_bits: 6
72
+ fq_qkv_bits: 6
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: olmo_core.optim.AdamWConfig
78
+ lr: 0.0024
79
+ betas:
80
+ - 0.9
81
+ - 0.95
82
+ eps: 1.0e-15
83
+ weight_decay: 0.1
84
+ fused: true
85
+ scheduler:
86
+ _CLASS_: olmo_core.optim.CosWithWarmup
87
+ warmup: 701
88
+ alpha_f: 0.1
89
+ compile_model: true
90
+ dp_config:
91
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
92
+ name: ddp
93
+ param_dtype: float32
94
+ reduce_dtype: float32
95
+ autocast_precision: bfloat16
96
+ max_grad_norm: 1.0
97
+ trainer:
98
+ _CLASS_: olmo_core.train.config.TrainerConfig
99
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/ckpt
100
+ save_overwrite: true
101
+ metrics_collect_interval: 10
102
+ cancel_check_interval: 5
103
+ callbacks:
104
+ config_saver:
105
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
106
+ checkpointer:
107
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
108
+ save_interval: 1007019
109
+ save_async: false
110
+ wandb:
111
+ enabled: false
112
+ init_seed: 101
113
+ n_tokens: 1839988736
114
+ load_path: null
115
+ validation_datasets:
116
+ dolma_cc_val:
117
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
118
+ validation_eval_interval: 500
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.04447223246097565,
8
+ "blocks.0.attention.w_out.weight": 0.06222694367170334,
9
+ "blocks.0.attention.w_q.weight": 0.04480505734682083,
10
+ "blocks.0.attention.w_v.weight": 0.11886989325284958,
11
+ "blocks.0.attention_norm.weight": 0.404301255941391,
12
+ "blocks.0.feed_forward.w1.weight": 0.057579103857278824,
13
+ "blocks.0.feed_forward.w2.weight": 0.04649105295538902,
14
+ "blocks.0.feed_forward.w3.weight": 0.06854056566953659,
15
+ "blocks.0.feed_forward_norm.weight": 0.398822158575058,
16
+ "blocks.1.attention.w_k.weight": 0.04462267458438873,
17
+ "blocks.1.attention.w_out.weight": 0.0488797165453434,
18
+ "blocks.1.attention.w_q.weight": 0.03434634581208229,
19
+ "blocks.1.attention.w_v.weight": 0.08584452420473099,
20
+ "blocks.1.attention_norm.weight": 0.36681878566741943,
21
+ "blocks.1.feed_forward.w1.weight": 0.062314797192811966,
22
+ "blocks.1.feed_forward.w2.weight": 0.05205081030726433,
23
+ "blocks.1.feed_forward.w3.weight": 0.07317005097866058,
24
+ "blocks.1.feed_forward_norm.weight": 0.4142328202724457,
25
+ "blocks.2.attention.w_k.weight": 0.044960957020521164,
26
+ "blocks.2.attention.w_out.weight": 0.05148233845829964,
27
+ "blocks.2.attention.w_q.weight": 0.03788594901561737,
28
+ "blocks.2.attention.w_v.weight": 0.10192523151636124,
29
+ "blocks.2.attention_norm.weight": 0.3903007209300995,
30
+ "blocks.2.feed_forward.w1.weight": 0.06271107494831085,
31
+ "blocks.2.feed_forward.w2.weight": 0.054983798414468765,
32
+ "blocks.2.feed_forward.w3.weight": 0.073421411216259,
33
+ "blocks.2.feed_forward_norm.weight": 0.4138107895851135,
34
+ "blocks.3.attention.w_k.weight": 0.0436176098883152,
35
+ "blocks.3.attention.w_out.weight": 0.05381327494978905,
36
+ "blocks.3.attention.w_q.weight": 0.03841737285256386,
37
+ "blocks.3.attention.w_v.weight": 0.09413135051727295,
38
+ "blocks.3.attention_norm.weight": 0.38852155208587646,
39
+ "blocks.3.feed_forward.w1.weight": 0.06022169068455696,
40
+ "blocks.3.feed_forward.w2.weight": 0.057319436222314835,
41
+ "blocks.3.feed_forward.w3.weight": 0.07215607166290283,
42
+ "blocks.3.feed_forward_norm.weight": 0.41180163621902466,
43
+ "blocks.4.attention.w_k.weight": 0.04765772819519043,
44
+ "blocks.4.attention.w_out.weight": 0.06548422574996948,
45
+ "blocks.4.attention.w_q.weight": 0.033922698348760605,
46
+ "blocks.4.attention.w_v.weight": 0.12088527530431747,
47
+ "blocks.4.attention_norm.weight": 0.40293774008750916,
48
+ "blocks.4.feed_forward.w1.weight": 0.05975082889199257,
49
+ "blocks.4.feed_forward.w2.weight": 0.06048689782619476,
50
+ "blocks.4.feed_forward.w3.weight": 0.07176288217306137,
51
+ "blocks.4.feed_forward_norm.weight": 0.41228407621383667,
52
+ "blocks.5.attention.w_k.weight": 0.045243971049785614,
53
+ "blocks.5.attention.w_out.weight": 0.08075490593910217,
54
+ "blocks.5.attention.w_q.weight": 0.034430962055921555,
55
+ "blocks.5.attention.w_v.weight": 0.07127781957387924,
56
+ "blocks.5.attention_norm.weight": 0.39218249917030334,
57
+ "blocks.5.feed_forward.w1.weight": 0.059570491313934326,
58
+ "blocks.5.feed_forward.w2.weight": 0.06447386741638184,
59
+ "blocks.5.feed_forward.w3.weight": 0.0718831717967987,
60
+ "blocks.5.feed_forward_norm.weight": 0.41259849071502686,
61
+ "blocks.6.attention.w_k.weight": 0.050092969089746475,
62
+ "blocks.6.attention.w_out.weight": 0.09478561580181122,
63
+ "blocks.6.attention.w_q.weight": 0.03283941373229027,
64
+ "blocks.6.attention.w_v.weight": 0.0808403417468071,
65
+ "blocks.6.attention_norm.weight": 0.40019938349723816,
66
+ "blocks.6.feed_forward.w1.weight": 0.0594840906560421,
67
+ "blocks.6.feed_forward.w2.weight": 0.075862817466259,
68
+ "blocks.6.feed_forward.w3.weight": 0.07260943949222565,
69
+ "blocks.6.feed_forward_norm.weight": 0.41187795996665955,
70
+ "blocks.7.attention.w_k.weight": 0.050687216222286224,
71
+ "blocks.7.attention.w_out.weight": 0.10306242853403091,
72
+ "blocks.7.attention.w_q.weight": 0.03203458711504936,
73
+ "blocks.7.attention.w_v.weight": 0.09186305850744247,
74
+ "blocks.7.attention_norm.weight": 0.40369895100593567,
75
+ "blocks.7.feed_forward.w1.weight": 0.057336632162332535,
76
+ "blocks.7.feed_forward.w2.weight": 0.08863422274589539,
77
+ "blocks.7.feed_forward.w3.weight": 0.07244992256164551,
78
+ "blocks.7.feed_forward_norm.weight": 0.4105563461780548,
79
+ "blocks.8.attention.w_k.weight": 0.0413111075758934,
80
+ "blocks.8.attention.w_out.weight": 0.17310740053653717,
81
+ "blocks.8.attention.w_q.weight": 0.03296327218413353,
82
+ "blocks.8.attention.w_v.weight": 0.09154683351516724,
83
+ "blocks.8.attention_norm.weight": 0.4069346785545349,
84
+ "blocks.8.feed_forward.w1.weight": 0.05462755262851715,
85
+ "blocks.8.feed_forward.w2.weight": 0.10933306068181992,
86
+ "blocks.8.feed_forward.w3.weight": 0.07172390073537827,
87
+ "blocks.8.feed_forward_norm.weight": 0.40633541345596313,
88
+ "embeddings.weight": 0.06951554119586945,
89
+ "lm_head.norm.weight": 1.401055932044983,
90
+ "lm_head.w_out.weight": 0.07445036619901657
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": 6,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.904266601452955
117
+ },
118
+ "base_loss": 3.904266601452955,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.904142573023332,
122
+ 3.905856580819682,
123
+ 3.904488195366709
124
+ ],
125
+ "0.01": [
126
+ 3.906111492066938,
127
+ 3.905960675969104,
128
+ 3.9051861183869865
129
+ ],
130
+ "0.02": [
131
+ 3.909523917056772,
132
+ 3.909952218499361,
133
+ 3.9093763325368474
134
+ ],
135
+ "0.05": [
136
+ 3.925472618660411,
137
+ 3.9244579266709434,
138
+ 3.9253922126202365
139
+ ],
140
+ "0.1": [
141
+ 3.984940330517964,
142
+ 3.9796524988031425,
143
+ 3.978508801910565
144
+ ],
145
+ "0.15": [
146
+ 4.09187802438134,
147
+ 4.074926657265826,
148
+ 4.0745152218430505
149
+ ],
150
+ "0.2": [
151
+ 4.262114352255951,
152
+ 4.221008223512892,
153
+ 4.219886568642033
154
+ ],
155
+ "0.3": [
156
+ 4.900139414809284,
157
+ 4.741884334174138,
158
+ 4.7508327298968
159
+ ],
160
+ "0.4": [
161
+ 6.209186188214108,
162
+ 5.761083701905357,
163
+ 5.801262238681934
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.904142573023332,
170
+ 3.905856580819682,
171
+ 3.904488195366709
172
+ ],
173
+ "0.01": [
174
+ 3.906111492066938,
175
+ 3.905960675969104,
176
+ 3.9051861183869865
177
+ ],
178
+ "0.02": [
179
+ 3.909523917056772,
180
+ 3.909952218499361,
181
+ 3.9093763325368474
182
+ ],
183
+ "0.05": [
184
+ 3.925472618660411,
185
+ 3.9244579266709434,
186
+ 3.9253922126202365
187
+ ],
188
+ "0.1": [
189
+ 3.984940330517964,
190
+ 3.9796524988031425,
191
+ 3.978508801910565
192
+ ],
193
+ "0.15": [
194
+ 4.09187802438134,
195
+ 4.074926657265826,
196
+ 4.0745152218430505
197
+ ],
198
+ "0.2": [
199
+ 4.262114352255951,
200
+ 4.221008223512892,
201
+ 4.219886568642033
202
+ ],
203
+ "0.3": [
204
+ 4.900139414809284,
205
+ 4.741884334174138,
206
+ 4.7508327298968
207
+ ],
208
+ "0.4": [
209
+ 6.209186188214108,
210
+ 5.761083701905357,
211
+ 5.801262238681934
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0005625149502859195,
218
+ "0.01": 0.0014861606880547384,
219
+ "0.02": 0.005350887911372393,
220
+ "0.05": 0.020840984530908546,
221
+ "0.1": 0.076767275624269,
222
+ "0.15": 0.17617336637711745,
223
+ "0.2": 0.3300697800173369,
224
+ "0.3": 0.8933522248404531,
225
+ "0.4": 2.019577441480844
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0005625149502859195,
268
+ "0.01": 0.0014861606880547384,
269
+ "0.02": 0.005350887911372393,
270
+ "0.05": 0.020840984530908546,
271
+ "0.1": 0.076767275624269,
272
+ "0.15": 0.17617336637711745,
273
+ "0.2": 0.3300697800173369,
274
+ "0.3": 0.8933522248404531,
275
+ "0.4": 2.019577441480844
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 78.5
284
+ }
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.04447223246097565,
8
+ "blocks.0.attention.w_out.weight": 0.06222694367170334,
9
+ "blocks.0.attention.w_q.weight": 0.04480505734682083,
10
+ "blocks.0.attention.w_v.weight": 0.11886989325284958,
11
+ "blocks.0.feed_forward.w1.weight": 0.057579103857278824,
12
+ "blocks.0.feed_forward.w2.weight": 0.04649105295538902,
13
+ "blocks.0.feed_forward.w3.weight": 0.06854056566953659,
14
+ "blocks.1.attention.w_k.weight": 0.04462267458438873,
15
+ "blocks.1.attention.w_out.weight": 0.0488797165453434,
16
+ "blocks.1.attention.w_q.weight": 0.03434634581208229,
17
+ "blocks.1.attention.w_v.weight": 0.08584452420473099,
18
+ "blocks.1.feed_forward.w1.weight": 0.062314797192811966,
19
+ "blocks.1.feed_forward.w2.weight": 0.05205081030726433,
20
+ "blocks.1.feed_forward.w3.weight": 0.07317005097866058,
21
+ "blocks.2.attention.w_k.weight": 0.044960957020521164,
22
+ "blocks.2.attention.w_out.weight": 0.05148233845829964,
23
+ "blocks.2.attention.w_q.weight": 0.03788594901561737,
24
+ "blocks.2.attention.w_v.weight": 0.10192523151636124,
25
+ "blocks.2.feed_forward.w1.weight": 0.06271107494831085,
26
+ "blocks.2.feed_forward.w2.weight": 0.054983798414468765,
27
+ "blocks.2.feed_forward.w3.weight": 0.073421411216259,
28
+ "blocks.3.attention.w_k.weight": 0.0436176098883152,
29
+ "blocks.3.attention.w_out.weight": 0.05381327494978905,
30
+ "blocks.3.attention.w_q.weight": 0.03841737285256386,
31
+ "blocks.3.attention.w_v.weight": 0.09413135051727295,
32
+ "blocks.3.feed_forward.w1.weight": 0.06022169068455696,
33
+ "blocks.3.feed_forward.w2.weight": 0.057319436222314835,
34
+ "blocks.3.feed_forward.w3.weight": 0.07215607166290283,
35
+ "blocks.4.attention.w_k.weight": 0.04765772819519043,
36
+ "blocks.4.attention.w_out.weight": 0.06548422574996948,
37
+ "blocks.4.attention.w_q.weight": 0.033922698348760605,
38
+ "blocks.4.attention.w_v.weight": 0.12088527530431747,
39
+ "blocks.4.feed_forward.w1.weight": 0.05975082889199257,
40
+ "blocks.4.feed_forward.w2.weight": 0.06048689782619476,
41
+ "blocks.4.feed_forward.w3.weight": 0.07176288217306137,
42
+ "blocks.5.attention.w_k.weight": 0.045243971049785614,
43
+ "blocks.5.attention.w_out.weight": 0.08075490593910217,
44
+ "blocks.5.attention.w_q.weight": 0.034430962055921555,
45
+ "blocks.5.attention.w_v.weight": 0.07127781957387924,
46
+ "blocks.5.feed_forward.w1.weight": 0.059570491313934326,
47
+ "blocks.5.feed_forward.w2.weight": 0.06447386741638184,
48
+ "blocks.5.feed_forward.w3.weight": 0.0718831717967987,
49
+ "blocks.6.attention.w_k.weight": 0.050092969089746475,
50
+ "blocks.6.attention.w_out.weight": 0.09478561580181122,
51
+ "blocks.6.attention.w_q.weight": 0.03283941373229027,
52
+ "blocks.6.attention.w_v.weight": 0.0808403417468071,
53
+ "blocks.6.feed_forward.w1.weight": 0.0594840906560421,
54
+ "blocks.6.feed_forward.w2.weight": 0.075862817466259,
55
+ "blocks.6.feed_forward.w3.weight": 0.07260943949222565,
56
+ "blocks.7.attention.w_k.weight": 0.050687216222286224,
57
+ "blocks.7.attention.w_out.weight": 0.10306242853403091,
58
+ "blocks.7.attention.w_q.weight": 0.03203458711504936,
59
+ "blocks.7.attention.w_v.weight": 0.09186305850744247,
60
+ "blocks.7.feed_forward.w1.weight": 0.057336632162332535,
61
+ "blocks.7.feed_forward.w2.weight": 0.08863422274589539,
62
+ "blocks.7.feed_forward.w3.weight": 0.07244992256164551,
63
+ "blocks.8.attention.w_k.weight": 0.0413111075758934,
64
+ "blocks.8.attention.w_out.weight": 0.17310740053653717,
65
+ "blocks.8.attention.w_q.weight": 0.03296327218413353,
66
+ "blocks.8.attention.w_v.weight": 0.09154683351516724,
67
+ "blocks.8.feed_forward.w1.weight": 0.05462755262851715,
68
+ "blocks.8.feed_forward.w2.weight": 0.10933306068181992,
69
+ "blocks.8.feed_forward.w3.weight": 0.07172390073537827
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": 6,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.904266601452955
96
+ },
97
+ "base_loss": 3.904266601452955,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.9049811009812228,
101
+ 3.9050922656963607,
102
+ 3.9045786482495832
103
+ ],
104
+ "0.01": [
105
+ 3.9052768881679367,
106
+ 3.9059409226557116,
107
+ 3.9047522105870813
108
+ ],
109
+ "0.02": [
110
+ 3.9056773221414636,
111
+ 3.906056570891774,
112
+ 3.906350495813252
113
+ ],
114
+ "0.05": [
115
+ 3.9104463643199243,
116
+ 3.909633035627544,
117
+ 3.9087067855874817
118
+ ],
119
+ "0.1": [
120
+ 3.927618077318498,
121
+ 3.925788155248195,
122
+ 3.9249304138773375
123
+ ],
124
+ "0.15": [
125
+ 3.955862181699748,
126
+ 3.952781071692016,
127
+ 3.952633987949297
128
+ ],
129
+ "0.2": [
130
+ 4.0011487398755365,
131
+ 3.9967307581679266,
132
+ 3.995774160246703
133
+ ],
134
+ "0.3": [
135
+ 4.152392294090428,
136
+ 4.146487391400203,
137
+ 4.144532861876632
138
+ ],
139
+ "0.4": [
140
+ 4.442498969337913,
141
+ 4.438886389958466,
142
+ 4.428868185455077
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.9049811009812228,
149
+ 3.9050922656963607,
150
+ 3.9045786482495832
151
+ ],
152
+ "0.01": [
153
+ 3.9052768881679367,
154
+ 3.9059409226557116,
155
+ 3.9047522105870813
156
+ ],
157
+ "0.02": [
158
+ 3.9056773221414636,
159
+ 3.906056570891774,
160
+ 3.906350495813252
161
+ ],
162
+ "0.05": [
163
+ 3.9104463643199243,
164
+ 3.909633035627544,
165
+ 3.9087067855874817
166
+ ],
167
+ "0.1": [
168
+ 3.927618077318498,
169
+ 3.925788155248195,
170
+ 3.9249304138773375
171
+ ],
172
+ "0.15": [
173
+ 3.955862181699748,
174
+ 3.952781071692016,
175
+ 3.952633987949297
176
+ ],
177
+ "0.2": [
178
+ 4.0011487398755365,
179
+ 3.9967307581679266,
180
+ 3.995774160246703
181
+ ],
182
+ "0.3": [
183
+ 4.152392294090428,
184
+ 4.146487391400203,
185
+ 4.144532861876632
186
+ ],
187
+ "0.4": [
188
+ 4.442498969337913,
189
+ 4.438886389958466,
190
+ 4.428868185455077
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.0006174035227672903,
197
+ "0.01": 0.0010567390172884217,
198
+ "0.02": 0.0017615281625413814,
199
+ "0.05": 0.00532879372536188,
200
+ "0.1": 0.02184561402838847,
201
+ "0.15": 0.04949247899406517,
202
+ "0.2": 0.09361795131043449,
203
+ "0.3": 0.24353758100280043,
204
+ "0.4": 0.5324845801308635
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.0006174035227672903,
247
+ "0.01": 0.0010567390172884217,
248
+ "0.02": 0.0017615281625413814,
249
+ "0.05": 0.00532879372536188,
250
+ "0.1": 0.02184561402838847,
251
+ "0.15": 0.04949247899406517,
252
+ "0.2": 0.09361795131043449,
253
+ "0.3": 0.24353758100280043,
254
+ "0.4": 0.5324845801308635
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 72.9
263
+ }
pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr2.4e-03-s101",
3
+ "w_bits": 6,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.904266601452955
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.34406060706216
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 3.9607311680540698
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.923247943308014
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.904266601452955
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.9050877665261488
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 0.43979400560920556
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.056464566601114985
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.018981341855059064
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.0
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0008211650731939635
42
+ }
43
+ },
44
+ "own_grid_max_rel_weight_change": 2.2933392074264702e-07,
45
+ "own_grid_delta_loss": 0.0,
46
+ "checks": {
47
+ "weights_restored": true,
48
+ "own_grid_weights_unchanged": true,
49
+ "own_grid_is_fixed_point": true
50
+ },
51
+ "ok": true
52
+ }
pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/finalize.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101",
3
+ "w_bits": 6,
4
+ "a_bits": 6,
5
+ "qkv_bits": 6,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.9273053028299922
13
+ },
14
+ "losses_fakequant_master": {
15
+ "val_dolma_cc": 3.9291688200086536
16
+ },
17
+ "losses_eval_model": {
18
+ "val_dolma_cc": 3.9291688200086536
19
+ },
20
+ "max_distinct_values_per_row": 62,
21
+ "rel_quant_noise_per_layer": {
22
+ "blocks.0.attention.w_q": 0.029958514496684074,
23
+ "blocks.0.attention.w_k": 0.030053334310650826,
24
+ "blocks.0.attention.w_v": 0.03284899517893791,
25
+ "blocks.0.attention.w_out": 0.03345663100481033,
26
+ "blocks.0.feed_forward.w1": 0.029894156381487846,
27
+ "blocks.0.feed_forward.w2": 0.03646356984972954,
28
+ "blocks.0.feed_forward.w3": 0.029902366921305656,
29
+ "blocks.1.attention.w_q": 0.03054300881922245,
30
+ "blocks.1.attention.w_k": 0.03423832729458809,
31
+ "blocks.1.attention.w_v": 0.03337271139025688,
32
+ "blocks.1.attention.w_out": 0.03670801594853401,
33
+ "blocks.1.feed_forward.w1": 0.03054652363061905,
34
+ "blocks.1.feed_forward.w2": 0.03641407936811447,
35
+ "blocks.1.feed_forward.w3": 0.0308720413595438,
36
+ "blocks.2.attention.w_q": 0.03073035180568695,
37
+ "blocks.2.attention.w_k": 0.0352826826274395,
38
+ "blocks.2.attention.w_v": 0.03186545521020889,
39
+ "blocks.2.attention.w_out": 0.03409954160451889,
40
+ "blocks.2.feed_forward.w1": 0.030620494857430458,
41
+ "blocks.2.feed_forward.w2": 0.039687350392341614,
42
+ "blocks.2.feed_forward.w3": 0.030994972214102745,
43
+ "blocks.3.attention.w_q": 0.03141950070858002,
44
+ "blocks.3.attention.w_k": 0.04081984609365463,
45
+ "blocks.3.attention.w_v": 0.032464489340782166,
46
+ "blocks.3.attention.w_out": 0.03311280161142349,
47
+ "blocks.3.feed_forward.w1": 0.030588535591959953,
48
+ "blocks.3.feed_forward.w2": 0.0373603031039238,
49
+ "blocks.3.feed_forward.w3": 0.031179826706647873,
50
+ "blocks.4.attention.w_q": 0.032007645815610886,
51
+ "blocks.4.attention.w_k": 0.04560428112745285,
52
+ "blocks.4.attention.w_v": 0.0322083942592144,
53
+ "blocks.4.attention.w_out": 0.03341170400381088,
54
+ "blocks.4.feed_forward.w1": 0.030829299241304398,
55
+ "blocks.4.feed_forward.w2": 0.03711560368537903,
56
+ "blocks.4.feed_forward.w3": 0.03147368133068085,
57
+ "blocks.5.attention.w_q": 0.03180655464529991,
58
+ "blocks.5.attention.w_k": 0.043339189141988754,
59
+ "blocks.5.attention.w_v": 0.032215967774391174,
60
+ "blocks.5.attention.w_out": 0.03398735448718071,
61
+ "blocks.5.feed_forward.w1": 0.03098096139729023,
62
+ "blocks.5.feed_forward.w2": 0.03750373050570488,
63
+ "blocks.5.feed_forward.w3": 0.031507816165685654,
64
+ "blocks.6.attention.w_q": 0.03224027529358864,
65
+ "blocks.6.attention.w_k": 0.038728248327970505,
66
+ "blocks.6.attention.w_v": 0.03304196894168854,
67
+ "blocks.6.attention.w_out": 0.036059584468603134,
68
+ "blocks.6.feed_forward.w1": 0.030856654047966003,
69
+ "blocks.6.feed_forward.w2": 0.038444433361291885,
70
+ "blocks.6.feed_forward.w3": 0.031284697353839874,
71
+ "blocks.7.attention.w_q": 0.031709592789411545,
72
+ "blocks.7.attention.w_k": 0.04747031256556511,
73
+ "blocks.7.attention.w_v": 0.032274387776851654,
74
+ "blocks.7.attention.w_out": 0.0332445427775383,
75
+ "blocks.7.feed_forward.w1": 0.030889054760336876,
76
+ "blocks.7.feed_forward.w2": 0.040012676268815994,
77
+ "blocks.7.feed_forward.w3": 0.03180297836661339,
78
+ "blocks.8.attention.w_q": 0.031886614859104156,
79
+ "blocks.8.attention.w_k": 0.03702918440103531,
80
+ "blocks.8.attention.w_v": 0.03320465236902237,
81
+ "blocks.8.attention.w_out": 0.03827323764562607,
82
+ "blocks.8.feed_forward.w1": 0.03127485141158104,
83
+ "blocks.8.feed_forward.w2": 0.04559102654457092,
84
+ "blocks.8.feed_forward.w3": 0.03272668644785881
85
+ },
86
+ "rel_quant_noise_mean": 0.03408787733623906,
87
+ "n_quantized_params": 30670848,
88
+ "n_params": 82191872,
89
+ "param_dtypes": [
90
+ "torch.float32"
91
+ ],
92
+ "norm_gain_max_abs_dev_from_1": 0.9966011643409729,
93
+ "loss_eval_model": 3.9291688200086536,
94
+ "checks": {
95
+ "snap_reproduces_fakequant": true,
96
+ "grid_levels_ok": true,
97
+ "master_weights_fp32": true,
98
+ "norm_gains_trained": true,
99
+ "loss_finite": true
100
+ },
101
+ "ok": true
102
+ }
pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.10295666009187698,
8
+ "blocks.0.attention.w_out.weight": 0.07433991879224777,
9
+ "blocks.0.attention.w_q.weight": 0.10253647714853287,
10
+ "blocks.0.attention.w_v.weight": 0.2449776977300644,
11
+ "blocks.0.feed_forward.w1.weight": 0.09247293323278427,
12
+ "blocks.0.feed_forward.w2.weight": 0.06845320016145706,
13
+ "blocks.0.feed_forward.w3.weight": 0.09594220668077469,
14
+ "blocks.1.attention.w_k.weight": 0.0940186083316803,
15
+ "blocks.1.attention.w_out.weight": 0.059166502207517624,
16
+ "blocks.1.attention.w_q.weight": 0.09678329527378082,
17
+ "blocks.1.attention.w_v.weight": 0.12398216873407364,
18
+ "blocks.1.feed_forward.w1.weight": 0.09601962566375732,
19
+ "blocks.1.feed_forward.w2.weight": 0.07874780148267746,
20
+ "blocks.1.feed_forward.w3.weight": 0.1040811538696289,
21
+ "blocks.2.attention.w_k.weight": 0.09405351430177689,
22
+ "blocks.2.attention.w_out.weight": 0.06419295817613602,
23
+ "blocks.2.attention.w_q.weight": 0.09307216107845306,
24
+ "blocks.2.attention.w_v.weight": 0.1580667495727539,
25
+ "blocks.2.feed_forward.w1.weight": 0.09151057153940201,
26
+ "blocks.2.feed_forward.w2.weight": 0.08589061349630356,
27
+ "blocks.2.feed_forward.w3.weight": 0.10345649719238281,
28
+ "blocks.3.attention.w_k.weight": 0.05170898512005806,
29
+ "blocks.3.attention.w_out.weight": 0.07905052602291107,
30
+ "blocks.3.attention.w_q.weight": 0.039911024272441864,
31
+ "blocks.3.attention.w_v.weight": 0.13418684899806976,
32
+ "blocks.3.feed_forward.w1.weight": 0.08949227631092072,
33
+ "blocks.3.feed_forward.w2.weight": 0.08585387468338013,
34
+ "blocks.3.feed_forward.w3.weight": 0.10214531421661377,
35
+ "blocks.4.attention.w_k.weight": 0.09089221805334091,
36
+ "blocks.4.attention.w_out.weight": 0.08829694986343384,
37
+ "blocks.4.attention.w_q.weight": 0.08905211836099625,
38
+ "blocks.4.attention.w_v.weight": 0.10798521339893341,
39
+ "blocks.4.feed_forward.w1.weight": 0.08801241219043732,
40
+ "blocks.4.feed_forward.w2.weight": 0.086526058614254,
41
+ "blocks.4.feed_forward.w3.weight": 0.09784262627363205,
42
+ "blocks.5.attention.w_k.weight": 0.05476023629307747,
43
+ "blocks.5.attention.w_out.weight": 0.09816227108240128,
44
+ "blocks.5.attention.w_q.weight": 0.03678838536143303,
45
+ "blocks.5.attention.w_v.weight": 0.14953851699829102,
46
+ "blocks.5.feed_forward.w1.weight": 0.08689950406551361,
47
+ "blocks.5.feed_forward.w2.weight": 0.09530370682477951,
48
+ "blocks.5.feed_forward.w3.weight": 0.09842880815267563,
49
+ "blocks.6.attention.w_k.weight": 0.046556148678064346,
50
+ "blocks.6.attention.w_out.weight": 0.14122274518013,
51
+ "blocks.6.attention.w_q.weight": 0.03567918762564659,
52
+ "blocks.6.attention.w_v.weight": 0.17810435593128204,
53
+ "blocks.6.feed_forward.w1.weight": 0.08768820762634277,
54
+ "blocks.6.feed_forward.w2.weight": 0.11142566800117493,
55
+ "blocks.6.feed_forward.w3.weight": 0.10013778507709503,
56
+ "blocks.7.attention.w_k.weight": 0.06066993623971939,
57
+ "blocks.7.attention.w_out.weight": 0.14124080538749695,
58
+ "blocks.7.attention.w_q.weight": 0.032068297266960144,
59
+ "blocks.7.attention.w_v.weight": 0.13702520728111267,
60
+ "blocks.7.feed_forward.w1.weight": 0.08460602909326553,
61
+ "blocks.7.feed_forward.w2.weight": 0.13135100901126862,
62
+ "blocks.7.feed_forward.w3.weight": 0.09995918720960617,
63
+ "blocks.8.attention.w_k.weight": 0.04496841877698898,
64
+ "blocks.8.attention.w_out.weight": 0.20446938276290894,
65
+ "blocks.8.attention.w_q.weight": 0.03546059504151344,
66
+ "blocks.8.attention.w_v.weight": 0.1397651731967926,
67
+ "blocks.8.feed_forward.w1.weight": 0.08004491031169891,
68
+ "blocks.8.feed_forward.w2.weight": 0.14966775476932526,
69
+ "blocks.8.feed_forward.w3.weight": 0.09858717769384384
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": 6,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.9291688200086536
96
+ },
97
+ "base_loss": 3.9291688200086536,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.929990462385122,
101
+ 3.9300811226210577,
102
+ 3.9286797484696248
103
+ ],
104
+ "0.01": [
105
+ 3.9299742810216456,
106
+ 3.929023837182808,
107
+ 3.9296316007443552
108
+ ],
109
+ "0.02": [
110
+ 3.9296030525312786,
111
+ 3.929805515190826,
112
+ 3.930205507071976
113
+ ],
114
+ "0.05": [
115
+ 3.9350386917683746,
116
+ 3.934420873539483,
117
+ 3.935254033791695
118
+ ],
119
+ "0.1": [
120
+ 3.951707530556173,
121
+ 3.952103301041967,
122
+ 3.953444804931003
123
+ ],
124
+ "0.15": [
125
+ 3.984590967994828,
126
+ 3.9838400564864167,
127
+ 3.9871740785449017
128
+ ],
129
+ "0.2": [
130
+ 4.031742510469999,
131
+ 4.0354458752796845,
132
+ 4.036791694459802
133
+ ],
134
+ "0.3": [
135
+ 4.207225285518124,
136
+ 4.217448648270866,
137
+ 4.214980704741259
138
+ ],
139
+ "0.4": [
140
+ 4.549954341638222,
141
+ 4.537920336737617,
142
+ 4.557709361804521
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.929990462385122,
149
+ 3.9300811226210577,
150
+ 3.9286797484696248
151
+ ],
152
+ "0.01": [
153
+ 3.9299742810216456,
154
+ 3.929023837182808,
155
+ 3.9296316007443552
156
+ ],
157
+ "0.02": [
158
+ 3.9296030525312786,
159
+ 3.929805515190826,
160
+ 3.930205507071976
161
+ ],
162
+ "0.05": [
163
+ 3.9350386917683746,
164
+ 3.934420873539483,
165
+ 3.935254033791695
166
+ ],
167
+ "0.1": [
168
+ 3.951707530556173,
169
+ 3.952103301041967,
170
+ 3.953444804931003
171
+ ],
172
+ "0.15": [
173
+ 3.984590967994828,
174
+ 3.9838400564864167,
175
+ 3.9871740785449017
176
+ ],
177
+ "0.2": [
178
+ 4.031742510469999,
179
+ 4.0354458752796845,
180
+ 4.036791694459802
181
+ ],
182
+ "0.3": [
183
+ 4.207225285518124,
184
+ 4.217448648270866,
185
+ 4.214980704741259
186
+ ],
187
+ "0.4": [
188
+ 4.549954341638222,
189
+ 4.537920336737617,
190
+ 4.557709361804521
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.00041495781661460285,
197
+ "0.01": 0.0003744196409489753,
198
+ "0.02": 0.0007025382560397198,
199
+ "0.05": 0.0057357130245310195,
200
+ "0.1": 0.023249725501060414,
201
+ "0.15": 0.056032881000061874,
202
+ "0.2": 0.10549120672784174,
203
+ "0.3": 0.2840493928347625,
204
+ "0.4": 0.6193591933847999
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.00041495781661460285,
247
+ "0.01": 0.0003744196409489753,
248
+ "0.02": 0.0007025382560397198,
249
+ "0.05": 0.0057357130245310195,
250
+ "0.1": 0.023249725501060414,
251
+ "0.15": 0.056032881000061874,
252
+ "0.2": 0.10549120672784174,
253
+ "0.3": 0.2840493928347625,
254
+ "0.4": 0.6193591933847999
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 71.7
263
+ }
pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101/ptq.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-adamw-lr4.8e-03-s101",
3
+ "w_bits": 6,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.9291688200086536
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.449217951886328
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 3.9960233321176966
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.953879643863526
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.929098875517303
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.9300679185345655
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 0.5200491318776743
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.06685451210904292
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.024710823854872555
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": -6.994449135078185e-05
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.0008990985259118567
42
+ }
43
+ },
44
+ "own_grid_max_rel_weight_change": 2.2763482832033333e-07,
45
+ "own_grid_delta_loss": -6.994449135078185e-05,
46
+ "checks": {
47
+ "weights_restored": true,
48
+ "own_grid_weights_unchanged": true,
49
+ "own_grid_is_fixed_point": true
50
+ },
51
+ "ok": true
52
+ }
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": 6,
8
+ "a": 6,
9
+ "qkv": 6
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 3.7739331285905724,
25
+ "heldout_cc_en_tail_4M": 4.356792957085902
26
+ },
27
+ "loss_bf16_master": 3.7739331285905724,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 3.8165860759648975
30
+ },
31
+ "bf16_minus_quant": -0.04265294737432512,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-m5-20",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 18.2
40
+ }
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/finalize.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "w_bits": 6,
4
+ "a_bits": 6,
5
+ "qkv_bits": 6,
6
+ "pre_norm": true,
7
+ "primary": "val_dolma_cc",
8
+ "eval_tokens": {
9
+ "val_dolma_cc": 499712
10
+ },
11
+ "losses_master_plain": {
12
+ "val_dolma_cc": 3.816500349166907
13
+ },
14
+ "losses_fakequant_master": {
15
+ "val_dolma_cc": 3.8165860759648975
16
+ },
17
+ "losses_eval_model": {
18
+ "val_dolma_cc": 3.8165860759648975
19
+ },
20
+ "max_distinct_values_per_row": 62,
21
+ "rel_quant_noise_per_layer": {
22
+ "blocks.0.attention.w_q": 0.030932195484638214,
23
+ "blocks.0.attention.w_k": 0.031067954376339912,
24
+ "blocks.0.attention.w_v": 0.031908489763736725,
25
+ "blocks.0.attention.w_out": 0.03268344700336456,
26
+ "blocks.0.feed_forward.w1": 0.030635656788945198,
27
+ "blocks.0.feed_forward.w2": 0.03498230129480362,
28
+ "blocks.0.feed_forward.w3": 0.030625954270362854,
29
+ "blocks.1.attention.w_q": 0.030392387881875038,
30
+ "blocks.1.attention.w_k": 0.030469106510281563,
31
+ "blocks.1.attention.w_v": 0.03046298213303089,
32
+ "blocks.1.attention.w_out": 0.03174380585551262,
33
+ "blocks.1.feed_forward.w1": 0.030810555443167686,
34
+ "blocks.1.feed_forward.w2": 0.03493629768490791,
35
+ "blocks.1.feed_forward.w3": 0.030692772939801216,
36
+ "blocks.2.attention.w_q": 0.03051871620118618,
37
+ "blocks.2.attention.w_k": 0.030241983011364937,
38
+ "blocks.2.attention.w_v": 0.030938543379306793,
39
+ "blocks.2.attention.w_out": 0.033308736979961395,
40
+ "blocks.2.feed_forward.w1": 0.030609318986535072,
41
+ "blocks.2.feed_forward.w2": 0.03414329141378403,
42
+ "blocks.2.feed_forward.w3": 0.030738672241568565,
43
+ "blocks.3.attention.w_q": 0.030858557671308517,
44
+ "blocks.3.attention.w_k": 0.030566640198230743,
45
+ "blocks.3.attention.w_v": 0.030706897377967834,
46
+ "blocks.3.attention.w_out": 0.032133810222148895,
47
+ "blocks.3.feed_forward.w1": 0.030591895803809166,
48
+ "blocks.3.feed_forward.w2": 0.03722628951072693,
49
+ "blocks.3.feed_forward.w3": 0.030861755833029747,
50
+ "blocks.4.attention.w_q": 0.030701402574777603,
51
+ "blocks.4.attention.w_k": 0.03091822750866413,
52
+ "blocks.4.attention.w_v": 0.031039241701364517,
53
+ "blocks.4.attention.w_out": 0.032929953187704086,
54
+ "blocks.4.feed_forward.w1": 0.030669309198856354,
55
+ "blocks.4.feed_forward.w2": 0.034518878906965256,
56
+ "blocks.4.feed_forward.w3": 0.030812233686447144,
57
+ "blocks.5.attention.w_q": 0.030762795358896255,
58
+ "blocks.5.attention.w_k": 0.0308926273137331,
59
+ "blocks.5.attention.w_v": 0.030903514474630356,
60
+ "blocks.5.attention.w_out": 0.033717043697834015,
61
+ "blocks.5.feed_forward.w1": 0.03070686198771,
62
+ "blocks.5.feed_forward.w2": 0.03584417328238487,
63
+ "blocks.5.feed_forward.w3": 0.030798662453889847,
64
+ "blocks.6.attention.w_q": 0.03058581054210663,
65
+ "blocks.6.attention.w_k": 0.030940260738134384,
66
+ "blocks.6.attention.w_v": 0.031143827363848686,
67
+ "blocks.6.attention.w_out": 0.03557548299431801,
68
+ "blocks.6.feed_forward.w1": 0.030763618648052216,
69
+ "blocks.6.feed_forward.w2": 0.0363113172352314,
70
+ "blocks.6.feed_forward.w3": 0.030922498553991318,
71
+ "blocks.7.attention.w_q": 0.031031783670186996,
72
+ "blocks.7.attention.w_k": 0.031099854037165642,
73
+ "blocks.7.attention.w_v": 0.03084469400346279,
74
+ "blocks.7.attention.w_out": 0.033200811594724655,
75
+ "blocks.7.feed_forward.w1": 0.03116525150835514,
76
+ "blocks.7.feed_forward.w2": 0.041475873440504074,
77
+ "blocks.7.feed_forward.w3": 0.031072761863470078,
78
+ "blocks.8.attention.w_q": 0.03149639070034027,
79
+ "blocks.8.attention.w_k": 0.03132655471563339,
80
+ "blocks.8.attention.w_v": 0.03121422417461872,
81
+ "blocks.8.attention.w_out": 0.037734467536211014,
82
+ "blocks.8.feed_forward.w1": 0.031303565949201584,
83
+ "blocks.8.feed_forward.w2": 0.04681382700800896,
84
+ "blocks.8.feed_forward.w3": 0.03129962459206581
85
+ },
86
+ "rel_quant_noise_mean": 0.03219565781690772,
87
+ "n_quantized_params": 30670848,
88
+ "n_params": 82191872,
89
+ "param_dtypes": [
90
+ "torch.float32"
91
+ ],
92
+ "norm_gain_max_abs_dev_from_1": 0.8291723132133484,
93
+ "loss_eval_model": 3.8165860759648975,
94
+ "checks": {
95
+ "snap_reproduces_fakequant": true,
96
+ "grid_levels_ok": true,
97
+ "master_weights_fp32": true,
98
+ "norm_gains_trained": true,
99
+ "loss_finite": true
100
+ },
101
+ "ok": true
102
+ }
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/launch.yaml ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ d_model: 512
3
+ vocab_size: 50304
4
+ n_layers: 9
5
+ block:
6
+ attention:
7
+ name: default
8
+ n_heads: 8
9
+ bias: false
10
+ rope:
11
+ name: default
12
+ theta: 10000
13
+ full_precision: true
14
+ _CLASS_: olmo_core.nn.rope.RoPEConfig
15
+ backend: flash_2
16
+ dtype: float32
17
+ _CLASS_: olmo_core.nn.attention.AttentionConfig
18
+ layer_norm:
19
+ name: rms
20
+ eps: 1.0e-06
21
+ bias: false
22
+ dtype: float32
23
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
24
+ feed_forward:
25
+ hidden_size: 1536
26
+ name: default
27
+ bias: false
28
+ dtype: float32
29
+ _CLASS_: olmo_core.nn.feed_forward.FeedForwardConfig
30
+ name: default
31
+ _CLASS_: olmo_core.nn.transformer.config.TransformerBlockConfig
32
+ lm_head:
33
+ name: default
34
+ layer_norm:
35
+ name: rms
36
+ eps: 1.0e-06
37
+ bias: false
38
+ dtype: float32
39
+ _CLASS_: olmo_core.nn.layer_norm.LayerNormConfig
40
+ bias: false
41
+ dtype: float32
42
+ loss_implementation: default
43
+ _CLASS_: olmo_core.nn.lm_head.LMHeadConfig
44
+ name: default
45
+ dtype: float32
46
+ init_method: normal
47
+ init_seed: 101
48
+ init_std: 0.02
49
+ _CLASS_: olmo_core.nn.transformer.config.TransformerConfig
50
+ dataset:
51
+ _CLASS_: olmo_core.data.NumpyFSLDatasetConfig
52
+ tokenizer:
53
+ _CLASS_: olmo_core.data.TokenizerConfig
54
+ identifier: allenai/gpt-neox-olmo-dolma-v1_5
55
+ vocab_size: 50280
56
+ eos_token_id: 50279
57
+ pad_token_id: 1
58
+ sequence_length: 1024
59
+ work_dir: /data/user_data/jgai/precision-optimizer/work/dataset-cache
60
+ paths:
61
+ - /data/user_data/jgai/precision-perturb/dolma/cc_en_head/part-000-00000.npy
62
+ data_loader:
63
+ _CLASS_: olmo_core.data.NumpyDataLoaderConfig
64
+ global_batch_size: 262144
65
+ seed: 0
66
+ num_workers: 4
67
+ train_module_type: normal
68
+ train_module:
69
+ _CLASS_: precq.train_module.FakeQuantTransformerTrainModuleConfig
70
+ fq_weight_bits: 6
71
+ fq_act_bits: 6
72
+ fq_qkv_bits: 6
73
+ fq_pre_norm: true
74
+ rank_microbatch_size: 32768
75
+ max_sequence_length: 1024
76
+ optim:
77
+ _CLASS_: precq.muon26.MuonConfig
78
+ lr: 0.005
79
+ weight_decay: 0.1
80
+ momentum: 0.95
81
+ nesterov: true
82
+ ns_steps: 5
83
+ eps: 1.0e-07
84
+ adjust_lr_fn: null
85
+ adamw_lr: 0.0024
86
+ adamw_betas:
87
+ - 0.9
88
+ - 0.95
89
+ adamw_eps: 1.0e-15
90
+ adamw_weight_decay: 0.1
91
+ adamw_fused: true
92
+ muon_on_lm_head: false
93
+ scheduler:
94
+ _CLASS_: olmo_core.optim.CosWithWarmup
95
+ warmup: 701
96
+ alpha_f: 0.1
97
+ compile_model: true
98
+ dp_config:
99
+ _CLASS_: olmo_core.train.train_module.transformer.config.TransformerDataParallelConfig
100
+ name: ddp
101
+ param_dtype: float32
102
+ reduce_dtype: float32
103
+ autocast_precision: bfloat16
104
+ max_grad_norm: 1.0
105
+ trainer:
106
+ _CLASS_: olmo_core.train.config.TrainerConfig
107
+ save_folder: /data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/ckpt
108
+ save_overwrite: true
109
+ metrics_collect_interval: 10
110
+ cancel_check_interval: 5
111
+ callbacks:
112
+ config_saver:
113
+ _CLASS_: olmo_core.train.callbacks.config_saver.ConfigSaverCallback
114
+ checkpointer:
115
+ _CLASS_: olmo_core.train.callbacks.checkpointer.CheckpointerCallback
116
+ save_interval: 1007019
117
+ save_async: false
118
+ wandb:
119
+ enabled: false
120
+ init_seed: 101
121
+ n_tokens: 1839988736
122
+ load_path: null
123
+ validation_datasets:
124
+ dolma_cc_val:
125
+ - /data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy
126
+ validation_eval_interval: 500
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/perturb_all.json ADDED
@@ -0,0 +1,284 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "scope": "all",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.033884573727846146,
8
+ "blocks.0.attention.w_out.weight": 0.021545495837926865,
9
+ "blocks.0.attention.w_q.weight": 0.032501790672540665,
10
+ "blocks.0.attention.w_v.weight": 0.026910170912742615,
11
+ "blocks.0.attention_norm.weight": 0.3879252076148987,
12
+ "blocks.0.feed_forward.w1.weight": 0.02422112412750721,
13
+ "blocks.0.feed_forward.w2.weight": 0.008776023052632809,
14
+ "blocks.0.feed_forward.w3.weight": 0.024775676429271698,
15
+ "blocks.0.feed_forward_norm.weight": 0.4009745121002197,
16
+ "blocks.1.attention.w_k.weight": 0.027325544506311417,
17
+ "blocks.1.attention.w_out.weight": 0.017743699252605438,
18
+ "blocks.1.attention.w_q.weight": 0.026878027245402336,
19
+ "blocks.1.attention.w_v.weight": 0.02216923050582409,
20
+ "blocks.1.attention_norm.weight": 0.3890649080276489,
21
+ "blocks.1.feed_forward.w1.weight": 0.0247330479323864,
22
+ "blocks.1.feed_forward.w2.weight": 0.009560699574649334,
23
+ "blocks.1.feed_forward.w3.weight": 0.024706684052944183,
24
+ "blocks.1.feed_forward_norm.weight": 0.4063830077648163,
25
+ "blocks.2.attention.w_k.weight": 0.02852986566722393,
26
+ "blocks.2.attention.w_out.weight": 0.018322842195630074,
27
+ "blocks.2.attention.w_q.weight": 0.02715517021715641,
28
+ "blocks.2.attention.w_v.weight": 0.023582633584737778,
29
+ "blocks.2.attention_norm.weight": 0.3945653438568115,
30
+ "blocks.2.feed_forward.w1.weight": 0.02440229430794716,
31
+ "blocks.2.feed_forward.w2.weight": 0.010609530843794346,
32
+ "blocks.2.feed_forward.w3.weight": 0.024790987372398376,
33
+ "blocks.2.feed_forward_norm.weight": 0.40593409538269043,
34
+ "blocks.3.attention.w_k.weight": 0.029783764854073524,
35
+ "blocks.3.attention.w_out.weight": 0.02107260376214981,
36
+ "blocks.3.attention.w_q.weight": 0.027262361720204353,
37
+ "blocks.3.attention.w_v.weight": 0.024727392941713333,
38
+ "blocks.3.attention_norm.weight": 0.39631086587905884,
39
+ "blocks.3.feed_forward.w1.weight": 0.023838182911276817,
40
+ "blocks.3.feed_forward.w2.weight": 0.011221504770219326,
41
+ "blocks.3.feed_forward.w3.weight": 0.024804741144180298,
42
+ "blocks.3.feed_forward_norm.weight": 0.4052497446537018,
43
+ "blocks.4.attention.w_k.weight": 0.029391638934612274,
44
+ "blocks.4.attention.w_out.weight": 0.02368864417076111,
45
+ "blocks.4.attention.w_q.weight": 0.030474061146378517,
46
+ "blocks.4.attention.w_v.weight": 0.0250234492123127,
47
+ "blocks.4.attention_norm.weight": 0.40151700377464294,
48
+ "blocks.4.feed_forward.w1.weight": 0.023672061040997505,
49
+ "blocks.4.feed_forward.w2.weight": 0.012774117290973663,
50
+ "blocks.4.feed_forward.w3.weight": 0.024888446554541588,
51
+ "blocks.4.feed_forward_norm.weight": 0.40640202164649963,
52
+ "blocks.5.attention.w_k.weight": 0.0299989040941,
53
+ "blocks.5.attention.w_out.weight": 0.02527996338903904,
54
+ "blocks.5.attention.w_q.weight": 0.03006657026708126,
55
+ "blocks.5.attention.w_v.weight": 0.02713628113269806,
56
+ "blocks.5.attention_norm.weight": 0.40378764271736145,
57
+ "blocks.5.feed_forward.w1.weight": 0.023739339783787727,
58
+ "blocks.5.feed_forward.w2.weight": 0.01463079173117876,
59
+ "blocks.5.feed_forward.w3.weight": 0.025103239342570305,
60
+ "blocks.5.feed_forward_norm.weight": 0.4088801145553589,
61
+ "blocks.6.attention.w_k.weight": 0.029504111036658287,
62
+ "blocks.6.attention.w_out.weight": 0.02975505404174328,
63
+ "blocks.6.attention.w_q.weight": 0.02926630713045597,
64
+ "blocks.6.attention.w_v.weight": 0.027235006913542747,
65
+ "blocks.6.attention_norm.weight": 0.40453341603279114,
66
+ "blocks.6.feed_forward.w1.weight": 0.02357933484017849,
67
+ "blocks.6.feed_forward.w2.weight": 0.016710400581359863,
68
+ "blocks.6.feed_forward.w3.weight": 0.024904726073145866,
69
+ "blocks.6.feed_forward_norm.weight": 0.4096214175224304,
70
+ "blocks.7.attention.w_k.weight": 0.030117973685264587,
71
+ "blocks.7.attention.w_out.weight": 0.03328891843557358,
72
+ "blocks.7.attention.w_q.weight": 0.028999419882893562,
73
+ "blocks.7.attention.w_v.weight": 0.025352291762828827,
74
+ "blocks.7.attention_norm.weight": 0.40566807985305786,
75
+ "blocks.7.feed_forward.w1.weight": 0.022943098098039627,
76
+ "blocks.7.feed_forward.w2.weight": 0.019236372783780098,
77
+ "blocks.7.feed_forward.w3.weight": 0.024431584402918816,
78
+ "blocks.7.feed_forward_norm.weight": 0.41024506092071533,
79
+ "blocks.8.attention.w_k.weight": 0.028775880113244057,
80
+ "blocks.8.attention.w_out.weight": 0.039014291018247604,
81
+ "blocks.8.attention.w_q.weight": 0.026884691789746284,
82
+ "blocks.8.attention.w_v.weight": 0.02573145553469658,
83
+ "blocks.8.attention_norm.weight": 0.409767210483551,
84
+ "blocks.8.feed_forward.w1.weight": 0.02186930924654007,
85
+ "blocks.8.feed_forward.w2.weight": 0.023152999579906464,
86
+ "blocks.8.feed_forward.w3.weight": 0.02385779283940792,
87
+ "blocks.8.feed_forward_norm.weight": 0.4094424843788147,
88
+ "embeddings.weight": 0.07002507150173187,
89
+ "lm_head.norm.weight": 1.3853260278701782,
90
+ "lm_head.w_out.weight": 0.07471665740013123
91
+ },
92
+ "gammas": [
93
+ 0.005,
94
+ 0.01,
95
+ 0.02,
96
+ 0.05,
97
+ 0.1,
98
+ 0.15,
99
+ 0.2,
100
+ 0.3,
101
+ 0.4
102
+ ],
103
+ "seeds": [
104
+ 0,
105
+ 1,
106
+ 2
107
+ ],
108
+ "n_tensors": 84,
109
+ "n_perturbed_params": 82191872,
110
+ "eval_tokens": {
111
+ "val_dolma_cc": 499712
112
+ },
113
+ "primary": "val_dolma_cc",
114
+ "w_bits": 6,
115
+ "base_losses": {
116
+ "val_dolma_cc": 3.8165860759648975
117
+ },
118
+ "base_loss": 3.8165860759648975,
119
+ "loss": {
120
+ "0.005": [
121
+ 3.8172678371833086,
122
+ 3.8171116142565764,
123
+ 3.8159586725472936
124
+ ],
125
+ "0.01": [
126
+ 3.8176468903175227,
127
+ 3.818043408056804,
128
+ 3.8168794922845257
129
+ ],
130
+ "0.02": [
131
+ 3.8206880907293117,
132
+ 3.8205451970833733,
133
+ 3.820196484005677
134
+ ],
135
+ "0.05": [
136
+ 3.843052730337484,
137
+ 3.843897525835096,
138
+ 3.8416292241809287
139
+ ],
140
+ "0.1": [
141
+ 3.9249599206092856,
142
+ 3.929138217826567,
143
+ 3.921051436673467
144
+ ],
145
+ "0.15": [
146
+ 4.079257128156649,
147
+ 4.096721040742732,
148
+ 4.074365094313875
149
+ ],
150
+ "0.2": [
151
+ 4.34255490302199,
152
+ 4.388319989029223,
153
+ 4.333827130957245
154
+ ],
155
+ "0.3": [
156
+ 5.484848179687375,
157
+ 5.656619338838878,
158
+ 5.408889064578226
159
+ ],
160
+ "0.4": [
161
+ 7.629972741131737,
162
+ 7.77432493980658,
163
+ 7.303515376176326
164
+ ]
165
+ },
166
+ "loss_by_set": {
167
+ "val_dolma_cc": {
168
+ "0.005": [
169
+ 3.8172678371833086,
170
+ 3.8171116142565764,
171
+ 3.8159586725472936
172
+ ],
173
+ "0.01": [
174
+ 3.8176468903175227,
175
+ 3.818043408056804,
176
+ 3.8168794922845257
177
+ ],
178
+ "0.02": [
179
+ 3.8206880907293117,
180
+ 3.8205451970833733,
181
+ 3.820196484005677
182
+ ],
183
+ "0.05": [
184
+ 3.843052730337484,
185
+ 3.843897525835096,
186
+ 3.8416292241809287
187
+ ],
188
+ "0.1": [
189
+ 3.9249599206092856,
190
+ 3.929138217826567,
191
+ 3.921051436673467
192
+ ],
193
+ "0.15": [
194
+ 4.079257128156649,
195
+ 4.096721040742732,
196
+ 4.074365094313875
197
+ ],
198
+ "0.2": [
199
+ 4.34255490302199,
200
+ 4.388319989029223,
201
+ 4.333827130957245
202
+ ],
203
+ "0.3": [
204
+ 5.484848179687375,
205
+ 5.656619338838878,
206
+ 5.408889064578226
207
+ ],
208
+ "0.4": [
209
+ 7.629972741131737,
210
+ 7.77432493980658,
211
+ 7.303515376176326
212
+ ]
213
+ }
214
+ },
215
+ "mean_delta_by_set": {
216
+ "val_dolma_cc": {
217
+ "0.005": 0.0001932986974959583,
218
+ "0.01": 0.0009371875880535008,
219
+ "0.02": 0.003890514641223586,
220
+ "0.05": 0.026273750819605368,
221
+ "0.1": 0.10846378240487553,
222
+ "0.15": 0.266861678439521,
223
+ "0.2": 0.5383145983712554,
224
+ "0.3": 1.7001994517365953,
225
+ "0.4": 3.7526849430733167
226
+ }
227
+ },
228
+ "rel_noise_over_gamma": {
229
+ "0.005": [
230
+ 0.9965,
231
+ 1.0028
232
+ ],
233
+ "0.01": [
234
+ 0.9965,
235
+ 1.0028
236
+ ],
237
+ "0.02": [
238
+ 0.9965,
239
+ 1.0028
240
+ ],
241
+ "0.05": [
242
+ 0.9965,
243
+ 1.0028
244
+ ],
245
+ "0.1": [
246
+ 0.9965,
247
+ 1.0028
248
+ ],
249
+ "0.15": [
250
+ 0.9965,
251
+ 1.0028
252
+ ],
253
+ "0.2": [
254
+ 0.9965,
255
+ 1.0028
256
+ ],
257
+ "0.3": [
258
+ 0.9965,
259
+ 1.0028
260
+ ],
261
+ "0.4": [
262
+ 0.9965,
263
+ 1.0028
264
+ ]
265
+ },
266
+ "mean_delta": {
267
+ "0.005": 0.0001932986974959583,
268
+ "0.01": 0.0009371875880535008,
269
+ "0.02": 0.003890514641223586,
270
+ "0.05": 0.026273750819605368,
271
+ "0.1": 0.10846378240487553,
272
+ "0.15": 0.266861678439521,
273
+ "0.2": 0.5383145983712554,
274
+ "0.3": 1.7001994517365953,
275
+ "0.4": 3.7526849430733167
276
+ },
277
+ "checks": {
278
+ "base_loss_finite": true,
279
+ "noise_scale_matches_gamma": true,
280
+ "base_restored": true
281
+ },
282
+ "ok": true,
283
+ "t_total_s": 79.2
284
+ }
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/perturb_blocks.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "scope": "blocks",
4
+ "noise_ref": null,
5
+ "noise_scale": "rms",
6
+ "noise_std_per_tensor_over_gamma": {
7
+ "blocks.0.attention.w_k.weight": 0.033884573727846146,
8
+ "blocks.0.attention.w_out.weight": 0.021545495837926865,
9
+ "blocks.0.attention.w_q.weight": 0.032501790672540665,
10
+ "blocks.0.attention.w_v.weight": 0.026910170912742615,
11
+ "blocks.0.feed_forward.w1.weight": 0.02422112412750721,
12
+ "blocks.0.feed_forward.w2.weight": 0.008776023052632809,
13
+ "blocks.0.feed_forward.w3.weight": 0.024775676429271698,
14
+ "blocks.1.attention.w_k.weight": 0.027325544506311417,
15
+ "blocks.1.attention.w_out.weight": 0.017743699252605438,
16
+ "blocks.1.attention.w_q.weight": 0.026878027245402336,
17
+ "blocks.1.attention.w_v.weight": 0.02216923050582409,
18
+ "blocks.1.feed_forward.w1.weight": 0.0247330479323864,
19
+ "blocks.1.feed_forward.w2.weight": 0.009560699574649334,
20
+ "blocks.1.feed_forward.w3.weight": 0.024706684052944183,
21
+ "blocks.2.attention.w_k.weight": 0.02852986566722393,
22
+ "blocks.2.attention.w_out.weight": 0.018322842195630074,
23
+ "blocks.2.attention.w_q.weight": 0.02715517021715641,
24
+ "blocks.2.attention.w_v.weight": 0.023582633584737778,
25
+ "blocks.2.feed_forward.w1.weight": 0.02440229430794716,
26
+ "blocks.2.feed_forward.w2.weight": 0.010609530843794346,
27
+ "blocks.2.feed_forward.w3.weight": 0.024790987372398376,
28
+ "blocks.3.attention.w_k.weight": 0.029783764854073524,
29
+ "blocks.3.attention.w_out.weight": 0.02107260376214981,
30
+ "blocks.3.attention.w_q.weight": 0.027262361720204353,
31
+ "blocks.3.attention.w_v.weight": 0.024727392941713333,
32
+ "blocks.3.feed_forward.w1.weight": 0.023838182911276817,
33
+ "blocks.3.feed_forward.w2.weight": 0.011221504770219326,
34
+ "blocks.3.feed_forward.w3.weight": 0.024804741144180298,
35
+ "blocks.4.attention.w_k.weight": 0.029391638934612274,
36
+ "blocks.4.attention.w_out.weight": 0.02368864417076111,
37
+ "blocks.4.attention.w_q.weight": 0.030474061146378517,
38
+ "blocks.4.attention.w_v.weight": 0.0250234492123127,
39
+ "blocks.4.feed_forward.w1.weight": 0.023672061040997505,
40
+ "blocks.4.feed_forward.w2.weight": 0.012774117290973663,
41
+ "blocks.4.feed_forward.w3.weight": 0.024888446554541588,
42
+ "blocks.5.attention.w_k.weight": 0.0299989040941,
43
+ "blocks.5.attention.w_out.weight": 0.02527996338903904,
44
+ "blocks.5.attention.w_q.weight": 0.03006657026708126,
45
+ "blocks.5.attention.w_v.weight": 0.02713628113269806,
46
+ "blocks.5.feed_forward.w1.weight": 0.023739339783787727,
47
+ "blocks.5.feed_forward.w2.weight": 0.01463079173117876,
48
+ "blocks.5.feed_forward.w3.weight": 0.025103239342570305,
49
+ "blocks.6.attention.w_k.weight": 0.029504111036658287,
50
+ "blocks.6.attention.w_out.weight": 0.02975505404174328,
51
+ "blocks.6.attention.w_q.weight": 0.02926630713045597,
52
+ "blocks.6.attention.w_v.weight": 0.027235006913542747,
53
+ "blocks.6.feed_forward.w1.weight": 0.02357933484017849,
54
+ "blocks.6.feed_forward.w2.weight": 0.016710400581359863,
55
+ "blocks.6.feed_forward.w3.weight": 0.024904726073145866,
56
+ "blocks.7.attention.w_k.weight": 0.030117973685264587,
57
+ "blocks.7.attention.w_out.weight": 0.03328891843557358,
58
+ "blocks.7.attention.w_q.weight": 0.028999419882893562,
59
+ "blocks.7.attention.w_v.weight": 0.025352291762828827,
60
+ "blocks.7.feed_forward.w1.weight": 0.022943098098039627,
61
+ "blocks.7.feed_forward.w2.weight": 0.019236372783780098,
62
+ "blocks.7.feed_forward.w3.weight": 0.024431584402918816,
63
+ "blocks.8.attention.w_k.weight": 0.028775880113244057,
64
+ "blocks.8.attention.w_out.weight": 0.039014291018247604,
65
+ "blocks.8.attention.w_q.weight": 0.026884691789746284,
66
+ "blocks.8.attention.w_v.weight": 0.02573145553469658,
67
+ "blocks.8.feed_forward.w1.weight": 0.02186930924654007,
68
+ "blocks.8.feed_forward.w2.weight": 0.023152999579906464,
69
+ "blocks.8.feed_forward.w3.weight": 0.02385779283940792
70
+ },
71
+ "gammas": [
72
+ 0.005,
73
+ 0.01,
74
+ 0.02,
75
+ 0.05,
76
+ 0.1,
77
+ 0.15,
78
+ 0.2,
79
+ 0.3,
80
+ 0.4
81
+ ],
82
+ "seeds": [
83
+ 0,
84
+ 1,
85
+ 2
86
+ ],
87
+ "n_tensors": 63,
88
+ "n_perturbed_params": 30670848,
89
+ "eval_tokens": {
90
+ "val_dolma_cc": 499712
91
+ },
92
+ "primary": "val_dolma_cc",
93
+ "w_bits": 6,
94
+ "base_losses": {
95
+ "val_dolma_cc": 3.8165860759648975
96
+ },
97
+ "base_loss": 3.8165860759648975,
98
+ "loss": {
99
+ "0.005": [
100
+ 3.817336541468609,
101
+ 3.817538655931506,
102
+ 3.8168800165356633
103
+ ],
104
+ "0.01": [
105
+ 3.8171605613161628,
106
+ 3.816650633188959,
107
+ 3.8168201697471975
108
+ ],
109
+ "0.02": [
110
+ 3.8179068758463135,
111
+ 3.8178052141611283,
112
+ 3.818238981117444
113
+ ],
114
+ "0.05": [
115
+ 3.8249593447154986,
116
+ 3.8250348603532682,
117
+ 3.8262709545527156
118
+ ],
119
+ "0.1": [
120
+ 3.8528304287629402,
121
+ 3.852761274560618,
122
+ 3.854550848854923
123
+ ],
124
+ "0.15": [
125
+ 3.9037916220974953,
126
+ 3.902684196341472,
127
+ 3.9077033297865786
128
+ ],
129
+ "0.2": [
130
+ 3.9867743644386087,
131
+ 3.98497371827326,
132
+ 3.987170623964643
133
+ ],
134
+ "0.3": [
135
+ 4.314607317431654,
136
+ 4.309899569600771,
137
+ 4.3006122689914745
138
+ ],
139
+ "0.4": [
140
+ 5.02026613272549,
141
+ 5.055846697512039,
142
+ 4.985979411922304
143
+ ]
144
+ },
145
+ "loss_by_set": {
146
+ "val_dolma_cc": {
147
+ "0.005": [
148
+ 3.817336541468609,
149
+ 3.817538655931506,
150
+ 3.8168800165356633
151
+ ],
152
+ "0.01": [
153
+ 3.8171605613161628,
154
+ 3.816650633188959,
155
+ 3.8168201697471975
156
+ ],
157
+ "0.02": [
158
+ 3.8179068758463135,
159
+ 3.8178052141611283,
160
+ 3.818238981117444
161
+ ],
162
+ "0.05": [
163
+ 3.8249593447154986,
164
+ 3.8250348603532682,
165
+ 3.8262709545527156
166
+ ],
167
+ "0.1": [
168
+ 3.8528304287629402,
169
+ 3.852761274560618,
170
+ 3.854550848854923
171
+ ],
172
+ "0.15": [
173
+ 3.9037916220974953,
174
+ 3.902684196341472,
175
+ 3.9077033297865786
176
+ ],
177
+ "0.2": [
178
+ 3.9867743644386087,
179
+ 3.98497371827326,
180
+ 3.987170623964643
181
+ ],
182
+ "0.3": [
183
+ 4.314607317431654,
184
+ 4.309899569600771,
185
+ 4.3006122689914745
186
+ ],
187
+ "0.4": [
188
+ 5.02026613272549,
189
+ 5.055846697512039,
190
+ 4.985979411922304
191
+ ]
192
+ }
193
+ },
194
+ "mean_delta_by_set": {
195
+ "val_dolma_cc": {
196
+ "0.005": 0.0006656620136951119,
197
+ "0.01": 0.00029104545254243774,
198
+ "0.02": 0.0013976144100644738,
199
+ "0.05": 0.008835643908929658,
200
+ "0.1": 0.03679477476126225,
201
+ "0.15": 0.08814030677695106,
202
+ "0.2": 0.1697201595939397,
203
+ "0.3": 0.49178697604306887,
204
+ "0.4": 1.2041113380883797
205
+ }
206
+ },
207
+ "rel_noise_over_gamma": {
208
+ "0.005": [
209
+ 0.9966,
210
+ 1.0028
211
+ ],
212
+ "0.01": [
213
+ 0.9966,
214
+ 1.0028
215
+ ],
216
+ "0.02": [
217
+ 0.9966,
218
+ 1.0028
219
+ ],
220
+ "0.05": [
221
+ 0.9966,
222
+ 1.0028
223
+ ],
224
+ "0.1": [
225
+ 0.9966,
226
+ 1.0028
227
+ ],
228
+ "0.15": [
229
+ 0.9966,
230
+ 1.0028
231
+ ],
232
+ "0.2": [
233
+ 0.9966,
234
+ 1.0028
235
+ ],
236
+ "0.3": [
237
+ 0.9966,
238
+ 1.0028
239
+ ],
240
+ "0.4": [
241
+ 0.9966,
242
+ 1.0028
243
+ ]
244
+ },
245
+ "mean_delta": {
246
+ "0.005": 0.0006656620136951119,
247
+ "0.01": 0.00029104545254243774,
248
+ "0.02": 0.0013976144100644738,
249
+ "0.05": 0.008835643908929658,
250
+ "0.1": 0.03679477476126225,
251
+ "0.15": 0.08814030677695106,
252
+ "0.2": 0.1697201595939397,
253
+ "0.3": 0.49178697604306887,
254
+ "0.4": 1.2041113380883797
255
+ },
256
+ "checks": {
257
+ "base_loss_finite": true,
258
+ "noise_scale_matches_gamma": true,
259
+ "base_restored": true
260
+ },
261
+ "ok": true,
262
+ "t_total_s": 71.7
263
+ }
pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101/ptq.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "w_bits": 6,
4
+ "method": "RTN, symmetric absmax per output channel",
5
+ "scope": "transformer-block Linear weights",
6
+ "primary": "val_dolma_cc",
7
+ "base_losses": {
8
+ "val_dolma_cc": 3.8165860759648975
9
+ },
10
+ "losses_by_bits": {
11
+ "3": {
12
+ "val_dolma_cc": 4.549142319661365
13
+ },
14
+ "4": {
15
+ "val_dolma_cc": 3.9001482527445095
16
+ },
17
+ "5": {
18
+ "val_dolma_cc": 3.8357712028443145
19
+ },
20
+ "6": {
21
+ "val_dolma_cc": 3.8165860759648975
22
+ },
23
+ "8": {
24
+ "val_dolma_cc": 3.8169317139575623
25
+ }
26
+ },
27
+ "delta_by_bits": {
28
+ "3": {
29
+ "val_dolma_cc": 0.7325562436964672
30
+ },
31
+ "4": {
32
+ "val_dolma_cc": 0.08356217677961197
33
+ },
34
+ "5": {
35
+ "val_dolma_cc": 0.01918512687941698
36
+ },
37
+ "6": {
38
+ "val_dolma_cc": 0.0
39
+ },
40
+ "8": {
41
+ "val_dolma_cc": 0.00034563799266473794
42
+ }
43
+ },
44
+ "own_grid_max_rel_weight_change": 2.2763893525734602e-07,
45
+ "own_grid_delta_loss": 0.0,
46
+ "checks": {
47
+ "weights_restored": true,
48
+ "own_grid_weights_unchanged": true,
49
+ "own_grid_is_fixed_point": true
50
+ },
51
+ "ok": true
52
+ }
pom-30M-D1.84B-f6B-muon-lr5.0e-03-alr2.4e-03-s101/bf16_start.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "pom-30M-D1.84B-f6B-muon-lr5.0e-03-alr2.4e-03-s101",
3
+ "run_dir": "/data/user_data/jgai/precision-optimizer/runs/pom-30M-D1.84B-f6B-muon-lr5.0e-03-alr2.4e-03-s101",
4
+ "weights": "final-unsharded (fp32 master)",
5
+ "quantization": "none (weights, block-Linear inputs, q/k/v); pre-norm kept; BF16 autocast; uncompiled",
6
+ "run_bits": {
7
+ "w": 6,
8
+ "a": 6,
9
+ "qkv": 6
10
+ },
11
+ "pre_norm": true,
12
+ "val_files": [
13
+ "/data/user_data/jgai/precision-perturb/dolma/val_dolma_cc.npy",
14
+ "/data/user_data/jgai/precision-perturb/dolma/heldout_cc_en_tail_4M.npy"
15
+ ],
16
+ "n_inst": null,
17
+ "batch": 8,
18
+ "primary": "val_dolma_cc",
19
+ "n_seqs": {
20
+ "val_dolma_cc": 488,
21
+ "heldout_cc_en_tail_4M": 4096
22
+ },
23
+ "losses_bf16_master": {
24
+ "val_dolma_cc": 4.706428408139432,
25
+ "heldout_cc_en_tail_4M": 5.2704657109951345
26
+ },
27
+ "loss_bf16_master": 4.706428408139432,
28
+ "losses_eval_model": {
29
+ "val_dolma_cc": 4.668224568905942
30
+ },
31
+ "bf16_minus_quant": 0.03820383923349002,
32
+ "checks": {
33
+ "master_weights_fp32": true,
34
+ "loss_finite": true
35
+ },
36
+ "ok": true,
37
+ "host": "babel-o9-28",
38
+ "gpu": "NVIDIA L40S",
39
+ "t_s": 18.3
40
+ }