mudassiruddin21 commited on
Commit
461b667
·
verified ·
1 Parent(s): 6671108

Run r3: logs and results

Browse files
runs/r3/logs/ablate.log CHANGED
@@ -198,3 +198,22 @@ downloaded mudassiruddin21/EMC:submissions/r2/main-bf16
198
  [02:45:49] step 150/300 ce 0.806/1.360/1.701 acc 0.760/0.656/0.599 batch-tau~2.557 lr x0.58 1.32s/step w [1.0, 0.63, 0.28]
199
  [02:46:02] step 160/300 ce 0.691/1.161/1.447 acc 0.789/0.684/0.636 batch-tau~2.672 lr x0.53 1.32s/step w [1.0, 0.62, 0.28]
200
  [02:46:15] step 170/300 ce 0.676/1.119/1.324 acc 0.798/0.711/0.701 batch-tau~2.762 lr x0.48 1.31s/step w [1.0, 0.62, 0.27]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198
  [02:45:49] step 150/300 ce 0.806/1.360/1.701 acc 0.760/0.656/0.599 batch-tau~2.557 lr x0.58 1.32s/step w [1.0, 0.63, 0.28]
199
  [02:46:02] step 160/300 ce 0.691/1.161/1.447 acc 0.789/0.684/0.636 batch-tau~2.672 lr x0.53 1.32s/step w [1.0, 0.62, 0.28]
200
  [02:46:15] step 170/300 ce 0.676/1.119/1.324 acc 0.798/0.711/0.701 batch-tau~2.762 lr x0.48 1.31s/step w [1.0, 0.62, 0.27]
201
+ [02:46:28] step 180/300 ce 0.770/1.277/1.636 acc 0.773/0.663/0.604 batch-tau~2.595 lr x0.43 1.31s/step w [1.0, 0.63, 0.28]
202
+ [02:46:40] step 190/300 ce 0.175/0.295/0.391 acc 0.944/0.917/0.900 batch-tau~3.589 lr x0.39 1.31s/step w [1.0, 0.63, 0.28]
203
+ [02:46:53] step 200/300 ce 0.575/0.966/1.181 acc 0.825/0.752/0.718 batch-tau~2.892 lr x0.34 1.31s/step w [1.0, 0.63, 0.28]
204
+ [02:46:59] eval step 200: val tau 2.8444 | dev(1/4) tau 2.9154 (best 2.8376 @ 0)
205
+ [02:47:12] step 210/300 ce 0.222/0.379/0.455 acc 0.931/0.896/0.885 batch-tau~3.503 lr x0.30 1.33s/step w [1.0, 0.63, 0.28]
206
+ [02:47:24] step 220/300 ce 0.493/0.852/1.073 acc 0.854/0.780/0.746 batch-tau~3.018 lr x0.26 1.33s/step w [1.0, 0.63, 0.28]
207
+ [02:47:37] step 230/300 ce 0.796/1.334/1.677 acc 0.764/0.653/0.598 batch-tau~2.563 lr x0.23 1.33s/step w [1.0, 0.63, 0.28]
208
+ [02:47:50] step 240/300 ce 0.803/1.366/1.710 acc 0.757/0.641/0.592 batch-tau~2.530 lr x0.19 1.33s/step w [1.0, 0.63, 0.28]
209
+ [02:48:03] step 250/300 ce 0.143/0.237/0.335 acc 0.955/0.934/0.909 batch-tau~3.659 lr x0.17 1.33s/step w [1.0, 0.63, 0.28]
210
+ [02:48:16] step 260/300 ce 0.778/1.282/1.610 acc 0.762/0.653/0.600 batch-tau~2.558 lr x0.14 1.33s/step w [1.0, 0.63, 0.28]
211
+ [02:48:29] step 270/300 ce 0.146/0.254/0.322 acc 0.953/0.925/0.911 batch-tau~3.636 lr x0.13 1.32s/step w [1.0, 0.63, 0.28]
212
+ [02:48:42] step 280/300 ce 0.807/1.367/1.666 acc 0.759/0.646/0.606 batch-tau~2.546 lr x0.11 1.32s/step w [1.0, 0.63, 0.28]
213
+ [02:48:54] step 290/300 ce 0.078/0.137/0.189 acc 0.973/0.958/0.940 batch-tau~3.780 lr x0.10 1.32s/step w [1.0, 0.63, 0.28]
214
+ [02:49:07] step 300/300 ce 0.175/0.293/0.369 acc 0.943/0.915/0.897 batch-tau~3.581 lr x0.10 1.32s/step w [1.0, 0.63, 0.28]
215
+ [02:49:14] eval step 300: val tau 2.8464 | dev(1/4) tau 2.9072 (best 2.8444 @ 200)
216
+ [02:49:14] best checkpoint: step 300 (val tau 2.8464)
217
+ [02:49:14] ablation run finished (no export)
218
+ [02:49:15] recipe accw_topk_lora: validation tau 2.8376 -> 2.8464
219
+ [02:49:15] chosen recipe: accw_topk_lora (accw_topk with rank-128 LoRA adapters instead of full fine-tuning)
runs/r3/logs/pipeline.log CHANGED
@@ -18,3 +18,4 @@
18
  [2026-11-19 01:28:54] start extract
19
  [2026-11-19 02:10:26] done extract (2492s)
20
  [2026-11-19 02:10:27] start ablate
 
 
18
  [2026-11-19 01:28:54] start extract
19
  [2026-11-19 02:10:26] done extract (2492s)
20
  [2026-11-19 02:10:27] start ablate
21
+ [2026-11-19 02:49:15] done ablate (2328s)
runs/r3/logs/pipeline.out CHANGED
@@ -18,3 +18,4 @@
18
  [2026-11-19 01:28:54] start extract
19
  [2026-11-19 02:10:26] done extract (2492s)
20
  [2026-11-19 02:10:27] start ablate
 
 
18
  [2026-11-19 01:28:54] start extract
19
  [2026-11-19 02:10:26] done extract (2492s)
20
  [2026-11-19 02:10:27] start ablate
21
+ [2026-11-19 02:49:15] done ablate (2328s)
runs/r3/logs/stats.log CHANGED
@@ -32,3 +32,4 @@
32
  [2026-11-19 02:26:17] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 42.5/95.6 GiB, 100%, 408 W | load 0.0 | largest python 80.9 GB, python 0.1 GB, python 0.1 GB
33
  [2026-11-19 02:36:19] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 42.5/95.6 GiB, 100%, 457 W | load 0.0 | largest python 81.0 GB, python 0.1 GB, python 0.1 GB
34
  [2026-11-19 02:46:21] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 35.6/95.6 GiB, 100%, 460 W | load 0.0 | largest python 80.9 GB, python 0.1 GB, python 0.1 GB
 
 
32
  [2026-11-19 02:26:17] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 42.5/95.6 GiB, 100%, 408 W | load 0.0 | largest python 80.9 GB, python 0.1 GB, python 0.1 GB
33
  [2026-11-19 02:36:19] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 42.5/95.6 GiB, 100%, 457 W | load 0.0 | largest python 81.0 GB, python 0.1 GB, python 0.1 GB
34
  [2026-11-19 02:46:21] ram 74.8/172 GB used (cache 0.1, shmem 0.0) | cgroup 74.8/172 GB | disk 141 GB used | gpu 35.6/95.6 GiB, 100%, 460 W | load 0.0 | largest python 80.9 GB, python 0.1 GB, python 0.1 GB
35
+ [2026-11-19 02:49:15] ram 62.8/172 GB used (cache 0.0, shmem 0.0) | cgroup 62.8/172 GB | disk 141 GB used | gpu 0.0/95.6 GiB, 0%, 111 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
runs/r3/logs/sync.log CHANGED
@@ -22,3 +22,4 @@ uploaded logs/results to mudassiruddin21/EMC:runs/r3
22
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
23
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
24
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
 
 
22
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
23
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
24
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
25
+ uploaded logs/results to mudassiruddin21/EMC:runs/r3
runs/r3/results/ablation.json ADDED
@@ -0,0 +1,175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "steps": 300,
3
+ "recipes": {
4
+ "control": {
5
+ "description": "previous recipe",
6
+ "flags": [
7
+ "--step-w",
8
+ "1",
9
+ "0.7",
10
+ "0.5",
11
+ "--topk",
12
+ "0",
13
+ "--floor",
14
+ "0.3",
15
+ "--rel-lr",
16
+ "1e-3"
17
+ ],
18
+ "val_tau_start": 2.837570347370464,
19
+ "val_tau_best": 2.8412747765254567,
20
+ "history": [
21
+ {
22
+ "step": 100,
23
+ "val": 2.8209283481884477,
24
+ "dev_quarter": 2.882645141260478
25
+ },
26
+ {
27
+ "step": 200,
28
+ "val": 2.83268105552757,
29
+ "dev_quarter": 2.8949173682569382
30
+ },
31
+ {
32
+ "step": 300,
33
+ "val": 2.8412747765254567,
34
+ "dev_quarter": 2.895992515203493
35
+ }
36
+ ],
37
+ "seconds": 489.2
38
+ },
39
+ "accw_topk": {
40
+ "description": "acceptance-aware step weights + top-10 distillation",
41
+ "flags": [
42
+ "--step-w",
43
+ "auto",
44
+ "--topk",
45
+ "10",
46
+ "--floor",
47
+ "0.3",
48
+ "--rel-lr",
49
+ "1e-3"
50
+ ],
51
+ "val_tau_start": 2.837570347370464,
52
+ "val_tau_best": 2.840164730564513,
53
+ "history": [
54
+ {
55
+ "step": 100,
56
+ "val": 2.8189774830351633,
57
+ "dev_quarter": 2.881905648665425
58
+ },
59
+ {
60
+ "step": 200,
61
+ "val": 2.835615375664249,
62
+ "dev_quarter": 2.9029232452712206
63
+ },
64
+ {
65
+ "step": 300,
66
+ "val": 2.840164730564513,
67
+ "dev_quarter": 2.9017187499999997
68
+ }
69
+ ],
70
+ "seconds": 461.4
71
+ },
72
+ "accw_topk_mask": {
73
+ "description": "accw_topk + stricter masking of unreachable later steps",
74
+ "flags": [
75
+ "--step-w",
76
+ "auto",
77
+ "--topk",
78
+ "10",
79
+ "--floor",
80
+ "0.1",
81
+ "--rel-lr",
82
+ "1e-3"
83
+ ],
84
+ "val_tau_start": 2.837570347370464,
85
+ "val_tau_best": 2.8404631464428647,
86
+ "history": [
87
+ {
88
+ "step": 100,
89
+ "val": 2.8180696885599756,
90
+ "dev_quarter": 2.8728341584158414
91
+ },
92
+ {
93
+ "step": 200,
94
+ "val": 2.832868325195691,
95
+ "dev_quarter": 2.8929739834865247
96
+ },
97
+ {
98
+ "step": 300,
99
+ "val": 2.8404631464428647,
100
+ "dev_quarter": 2.893736366469305
101
+ }
102
+ ],
103
+ "seconds": 449.8
104
+ },
105
+ "accw_topk_lr2": {
106
+ "description": "accw_topk + twice the learning rate",
107
+ "flags": [
108
+ "--step-w",
109
+ "auto",
110
+ "--topk",
111
+ "10",
112
+ "--floor",
113
+ "0.3",
114
+ "--rel-lr",
115
+ "2e-3"
116
+ ],
117
+ "val_tau_start": 2.837570347370464,
118
+ "val_tau_best": 2.837570347370464,
119
+ "history": [
120
+ {
121
+ "step": 100,
122
+ "val": 2.7792054742444403,
123
+ "dev_quarter": 2.848443012732014
124
+ },
125
+ {
126
+ "step": 200,
127
+ "val": 2.820236855302241,
128
+ "dev_quarter": 2.8944825436408976
129
+ },
130
+ {
131
+ "step": 300,
132
+ "val": 2.8314032301793253,
133
+ "dev_quarter": 2.895245518316446
134
+ }
135
+ ],
136
+ "seconds": 449.1
137
+ },
138
+ "accw_topk_lora": {
139
+ "description": "accw_topk with rank-128 LoRA adapters instead of full fine-tuning",
140
+ "flags": [
141
+ "--step-w",
142
+ "auto",
143
+ "--topk",
144
+ "10",
145
+ "--floor",
146
+ "0.3",
147
+ "--rel-lr",
148
+ "1e-3",
149
+ "--lora",
150
+ "128"
151
+ ],
152
+ "val_tau_start": 2.837570347370464,
153
+ "val_tau_best": 2.846363494782744,
154
+ "history": [
155
+ {
156
+ "step": 100,
157
+ "val": 2.836942180830423,
158
+ "dev_quarter": 2.8956969130028063
159
+ },
160
+ {
161
+ "step": 200,
162
+ "val": 2.8444089954089176,
163
+ "dev_quarter": 2.9153846153846157
164
+ },
165
+ {
166
+ "step": 300,
167
+ "val": 2.846363494782744,
168
+ "dev_quarter": 2.9071696931747026
169
+ }
170
+ ],
171
+ "seconds": 451.9
172
+ }
173
+ },
174
+ "chosen": "accw_topk_lora"
175
+ }
runs/r3/state/ablate.done ADDED
@@ -0,0 +1 @@
 
 
1
+ 2328
runs/r3/state/train_args ADDED
@@ -0,0 +1 @@
 
 
1
+ --step-w auto --topk 10 --floor 0.3 --rel-lr 1e-3 --lora 128