Hugging Face's logo

AbstractPhil
/
beatrix-tokenizers

geolip-alephllm
byte-level
adapter
tokenizer
beatrix
Model card Files
xet
 Community
AbstractPhil commited on
Commit
2733a01
·
verified ·
1 Parent(s): 1a64c9c

surface arm bert/mse_solo_o0: anchors (every close), stats, results, log, readout

Browse files
Files changed (10) hide show
  1. surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close1_s13_bert.safetensors +3 -0
  2. surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close2_s13_bert.safetensors +3 -0
  3. surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close3_s13_bert.safetensors +3 -0
  4. surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close4_s13_bert.safetensors +3 -0
  5. surface/bert/mse_solo_o0/manifest.json +20 -0
  6. surface/bert/mse_solo_o0/readout.md +127 -0
  7. surface/bert/mse_solo_o0/results.json +0 -0
  8. surface/bert/mse_solo_o0/s13_bert.safetensors +3 -0
  9. surface/bert/mse_solo_o0/stats.pt +3 -0
  10. surface/bert/mse_solo_o0/train.log +160 -0
surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close1_s13_bert.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:475b89ac3a12286307c6797213b036a445242739475ca64740d3870e405b6aac
3
+ size 54819296
surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close2_s13_bert.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de0265c56030aaae450072999a4b971fb1af66ea22feb39dc10e30dd29d950dd
3
+ size 54819296
surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close3_s13_bert.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a8735c658879a064868f7ec48846e31e392d7608a79038076d27028a20f9a4a
3
+ size 54819296
surface/bert/mse_solo_o0/closes/sarm_bert_mse_solo_o0_close4_s13_bert.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c06f0aebe2bc73a8990c71b6e94805ae978dced253ec090adebe09ac8c7ddcba
3
+ size 54819296
surface/bert/mse_solo_o0/manifest.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "sarm_bert_mse_solo_o0",
3
+ "run": "mse_solo_o0",
4
+ "convention": "bert",
5
+ "member": "s13_bert",
6
+ "final_anchor": "sarm_bert_mse_solo_o0_close4_s13_bert.safetensors",
7
+ "closes": [
8
+ "sarm_bert_mse_solo_o0_close1_s13_bert.safetensors",
9
+ "sarm_bert_mse_solo_o0_close2_s13_bert.safetensors",
10
+ "sarm_bert_mse_solo_o0_close3_s13_bert.safetensors",
11
+ "sarm_bert_mse_solo_o0_close4_s13_bert.safetensors"
12
+ ],
13
+ "finished": true,
14
+ "form": "mse",
15
+ "base": "solo",
16
+ "steps": 4000,
17
+ "trunk_step": 245674,
18
+ "shipped": "2026-10-08 20:36 local",
19
+ "note": "every close's anchor ships (mids included); the final close's anchor is the run's s13_bert.safetensors"
20
+ }
surface/bert/mse_solo_o0/readout.md ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # The surface arm's readout: bert-base-uncased's spelling (the WordPiece convention)
2
+
3
+ A surface arm makes the trunk read a tokenizer's spelling of a text as it reads the text's own bytes. Every number below is read on the dual-extraction read's held-out captions (512 of each COCO draw, never trained on; draw 2 is the repeat), one document a row, at every token's closing byte, as whitened Procrustes alignment (1 = the same geometry up to a rotation; two folds by caption). The pre-registered bars are in the plan; B3 and B4 (against Qwen's own states, the mood axis) are the instrument's read.
4
+
5
+ ## The runs
6
+
7
+ - sarm_bert_mse_solo_o0: per-site MSE; alone on the bare trunk; blocks 0-31; 4000 of 4000 steps; seed offset 0
8
+
9
+ ## The verdicts (the final close of each run)
10
+
11
+ ### sarm_bert_mse_solo_o0
12
+
13
+ - B1 the gap: NOT MET over every served block >= 12 (the lowest 0.742 at block 31 on draw 1, 0.745 at block 31 on draw 2; bar 0.85); over blocks 12-28 alone: NOT MET
14
+ - B2 silence, whitened: NOT MET (the lowest A alignment 0.9867 at block 1 on draw 1, 0.9832 on draw 2; bar 0.995)
15
+ - B2 silence, in bits: MET (web +0.00055 bpb; bar +0.012)
16
+
17
+ ## The closes
18
+
19
+ ### sarm_bert_mse_solo_o0
20
+
21
+ | close | step | B1 lowest (draw 1 / 2) | B1 at 12 / 20 / 28 (draw 1) | B2 lowest (draw 1 / 2) | web bits | gate mean / widest |
22
+ |---|---|---|---|---|---|---|
23
+ | 0 | 0 | 0.674 / 0.690 | 0.872 / 0.854 / 0.786 | 1.0000 / 1.0000 | +0.00000 | 0.047 / 0.047 |
24
+ | 1 | 1000 | 0.709 / 0.723 | 0.894 / 0.886 / 0.827 | 0.9868 / 0.9824 | +0.00048 | 0.164 / 0.221 |
25
+ | 2 | 2000 | 0.725 / 0.738 | 0.901 / 0.896 / 0.843 | 0.9869 / 0.9830 | +0.00084 | 0.259 / 0.512 |
26
+ | 3 | 3000 | 0.740 / 0.744 | 0.905 / 0.903 / 0.852 | 0.9863 / 0.9780 | +0.00049 | 0.352 / 0.683 |
27
+ | 4 | 4000 | 0.742 / 0.745 | 0.907 / 0.906 / 0.854 | 0.9867 / 0.9832 | +0.00055 | 0.435 / 0.834 |
28
+
29
+ ## Every block at the final close
30
+
31
+ ### sarm_bert_mse_solo_o0, draw1 (512 captions, 4,966 sites), close 4 at step 4000
32
+
33
+ | block | gap before | armed B vs masked A | armed B vs armed A | A armed vs masked | mse | cos |
34
+ |---|---|---|---|---|---|---|
35
+ | 0 | 0.799 | 0.799 | 0.799 | 0.9997 | 0.264 | 0.894 |
36
+ | 4 | 0.868 | 0.880 | 0.881 | 0.9979 | 0.132 | 0.940 |
37
+ | 8 | 0.868 | 0.895 | 0.895 | 0.9966 | 0.130 | 0.942 |
38
+ | 12 | 0.872 | 0.907 | 0.908 | 0.9963 | 0.127 | 0.943 |
39
+ | 16 | 0.870 | 0.915 | 0.915 | 0.9962 | 0.123 | 0.942 |
40
+ | 18 | 0.859 | 0.912 | 0.912 | 0.9964 | 0.130 | 0.938 |
41
+ | 20 | 0.854 | 0.906 | 0.907 | 0.9952 | 0.138 | 0.936 |
42
+ | 22 | 0.842 | 0.900 | 0.902 | 0.9950 | 0.162 | 0.930 |
43
+ | 24 | 0.832 | 0.895 | 0.896 | 0.9951 | 0.172 | 0.925 |
44
+ | 26 | 0.810 | 0.873 | 0.876 | 0.9944 | 0.208 | 0.910 |
45
+ | 28 | 0.786 | 0.854 | 0.858 | 0.9923 | 0.242 | 0.893 |
46
+ | 29 | 0.769 | 0.838 | 0.841 | 0.9910 | 0.259 | 0.883 |
47
+ | 30 | 0.728 | 0.801 | 0.804 | 0.9917 | 0.288 | 0.874 |
48
+ | 31 | 0.674 | 0.742 | 0.747 | 0.9888 | 0.333 | 0.861 |
49
+
50
+ ### sarm_bert_mse_solo_o0, draw2 (512 captions, 5,109 sites), close 4 at step 4000
51
+
52
+ | block | gap before | armed B vs masked A | armed B vs armed A | A armed vs masked | mse | cos |
53
+ |---|---|---|---|---|---|---|
54
+ | 0 | 0.776 | 0.776 | 0.777 | 0.9994 | 0.251 | 0.900 |
55
+ | 4 | 0.861 | 0.870 | 0.871 | 0.9943 | 0.129 | 0.942 |
56
+ | 8 | 0.866 | 0.886 | 0.889 | 0.9955 | 0.127 | 0.942 |
57
+ | 12 | 0.873 | 0.903 | 0.904 | 0.9965 | 0.126 | 0.943 |
58
+ | 16 | 0.869 | 0.907 | 0.908 | 0.9962 | 0.124 | 0.941 |
59
+ | 18 | 0.859 | 0.903 | 0.905 | 0.9930 | 0.132 | 0.937 |
60
+ | 20 | 0.849 | 0.898 | 0.900 | 0.9944 | 0.142 | 0.934 |
61
+ | 22 | 0.837 | 0.888 | 0.893 | 0.9929 | 0.166 | 0.928 |
62
+ | 24 | 0.831 | 0.884 | 0.887 | 0.9928 | 0.176 | 0.922 |
63
+ | 26 | 0.806 | 0.861 | 0.866 | 0.9921 | 0.211 | 0.908 |
64
+ | 28 | 0.788 | 0.847 | 0.851 | 0.9889 | 0.244 | 0.891 |
65
+ | 29 | 0.774 | 0.831 | 0.836 | 0.9888 | 0.260 | 0.880 |
66
+ | 30 | 0.737 | 0.797 | 0.801 | 0.9879 | 0.288 | 0.870 |
67
+ | 31 | 0.690 | 0.745 | 0.750 | 0.9832 | 0.318 | 0.858 |
68
+
69
+ The member's share in bits per byte (all on minus the member masked): web +0.00055
70
+
71
+ ## The training curves
72
+
73
+ ### sarm_bert_mse_solo_o0
74
+
75
+ | step | task loss | mse | cos | InfoNCE | quiet on A | quiet off-domain |
76
+ |---|---|---|---|---|---|---|
77
+ | 1 | 0.6231 | 0.623 | 0.698 | None | 0.00000 | 0.00000 |
78
+ | 50 | 0.6639 | 0.664 | 0.677 | None | 0.00034 | 0.00025 |
79
+ | 250 | 0.5834 | 0.583 | 0.716 | None | 0.00149 | 0.00072 |
80
+ | 500 | 0.5191 | 0.519 | 0.749 | None | 0.00118 | 0.00050 |
81
+ | 1000 | 0.3609 | 0.361 | 0.832 | None | 0.00177 | 0.00092 |
82
+ | 1500 | 0.3010 | 0.301 | 0.862 | None | 0.00189 | 0.00104 |
83
+ | 2000 | 0.2712 | 0.271 | 0.877 | None | 0.00181 | 0.00127 |
84
+ | 2500 | 0.2601 | 0.260 | 0.883 | None | 0.00188 | 0.00097 |
85
+ | 3000 | 0.2567 | 0.257 | 0.885 | None | 0.00215 | 0.00112 |
86
+ | 3500 | 0.2391 | 0.239 | 0.893 | None | 0.00175 | 0.00097 |
87
+ | 4000 | 0.2343 | 0.234 | 0.896 | None | 0.00172 | 0.00091 |
88
+ | 1 | 0.6231 | 0.623 | 0.698 | None | 0.00000 | 0.00000 |
89
+ | 50 | 0.6639 | 0.664 | 0.677 | None | 0.00034 | 0.00025 |
90
+ | 250 | 0.5834 | 0.583 | 0.716 | None | 0.00149 | 0.00072 |
91
+ | 500 | 0.5191 | 0.519 | 0.749 | None | 0.00118 | 0.00050 |
92
+ | 1000 | 0.3609 | 0.361 | 0.832 | None | 0.00177 | 0.00092 |
93
+ | 1500 | 0.3010 | 0.301 | 0.862 | None | 0.00189 | 0.00104 |
94
+ | 2000 | 0.2712 | 0.271 | 0.877 | None | 0.00181 | 0.00127 |
95
+ | 2500 | 0.2601 | 0.260 | 0.883 | None | 0.00188 | 0.00097 |
96
+ | 3000 | 0.2567 | 0.257 | 0.885 | None | 0.00215 | 0.00112 |
97
+ | 3500 | 0.2391 | 0.239 | 0.893 | None | 0.00175 | 0.00097 |
98
+ | 4000 | 0.2343 | 0.234 | 0.896 | None | 0.00172 | 0.00091 |
99
+ | 1 | 0.6231 | 0.623 | 0.698 | None | 0.00000 | 0.00000 |
100
+ | 50 | 0.6639 | 0.664 | 0.677 | None | 0.00034 | 0.00025 |
101
+ | 250 | 0.5834 | 0.583 | 0.716 | None | 0.00149 | 0.00072 |
102
+ | 500 | 0.5191 | 0.519 | 0.749 | None | 0.00118 | 0.00050 |
103
+ | 1000 | 0.3609 | 0.361 | 0.832 | None | 0.00177 | 0.00092 |
104
+ | 1500 | 0.3010 | 0.301 | 0.862 | None | 0.00189 | 0.00104 |
105
+ | 2000 | 0.2712 | 0.271 | 0.877 | None | 0.00181 | 0.00127 |
106
+ | 2500 | 0.2601 | 0.260 | 0.883 | None | 0.00188 | 0.00097 |
107
+ | 3000 | 0.2567 | 0.257 | 0.885 | None | 0.00215 | 0.00112 |
108
+ | 3500 | 0.2391 | 0.239 | 0.893 | None | 0.00175 | 0.00097 |
109
+ | 4000 | 0.2343 | 0.234 | 0.896 | None | 0.00172 | 0.00091 |
110
+ | 1 | 0.6231 | 0.623 | 0.698 | None | 0.00000 | 0.00000 |
111
+ | 50 | 0.6639 | 0.664 | 0.677 | None | 0.00034 | 0.00025 |
112
+ | 250 | 0.5834 | 0.583 | 0.716 | None | 0.00149 | 0.00072 |
113
+ | 500 | 0.5191 | 0.519 | 0.749 | None | 0.00118 | 0.00050 |
114
+ | 1000 | 0.3609 | 0.361 | 0.832 | None | 0.00177 | 0.00092 |
115
+ | 1500 | 0.3010 | 0.301 | 0.862 | None | 0.00189 | 0.00104 |
116
+ | 2000 | 0.2712 | 0.271 | 0.877 | None | 0.00181 | 0.00127 |
117
+ | 2500 | 0.2601 | 0.260 | 0.883 | None | 0.00188 | 0.00097 |
118
+ | 3000 | 0.2567 | 0.257 | 0.885 | None | 0.00215 | 0.00112 |
119
+ | 3500 | 0.2391 | 0.239 | 0.893 | None | 0.00175 | 0.00097 |
120
+ | 4000 | 0.2343 | 0.234 | 0.896 | None | 0.00172 | 0.00091 |
121
+
122
+ ## The recipe
123
+
124
+ - per step: the teacher (member masked) reads the A rows: block outputs at the sites + logits; the armed model reads the B rows: per-site MSE of standardized LayerNorm'd block outputs against the teacher's at each token's closing byte, mean over served blocks; quiet on A: lambda x KL(teacher || armed) on the A rows; quiet off-domain: lambda x KL(member masked || armed) on one web row + one partner row
125
+ - lambda 2.0; pure Adam lr 0.001 wd 0; 2 x 4096-byte paired rows a step; the arm LAWFUL_16x8 at every block ({'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}); the trunk mini-beatrix-3 at step 245,674; alephllm 0.10.7, amoe-lora 0.2.11, torch 2.8.0+cu128
126
+ - the surface: bert-base-uncased's spelling (the WordPiece convention; sha256 9114256b104cca80..., vocab 30,522; A = the text written; sites = tokens); documents = one caption-like field of a caption-pack row (caption, short, long, tags, medium), sources drawn by {'dff': 0.161, 'scf': 0.161, 'danbooru': 0.164, 'cc12m': 0.406, 'coco': 0.108}
127
+ - the test set: the first 512 captions of each draw of coco_caps_2x2048.json (every caption of the file kept out of training)
surface/bert/mse_solo_o0/results.json ADDED
The diff for this file is too large to render. See raw diff
 
surface/bert/mse_solo_o0/s13_bert.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c06f0aebe2bc73a8990c71b6e94805ae978dced253ec090adebe09ac8c7ddcba
3
+ size 54819296
surface/bert/mse_solo_o0/stats.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15a35318cb48c90c0af571381ce2f81c4b8be8b886641f8e8f1915b4938c0fbf
3
+ size 264325
surface/bert/mse_solo_o0/train.log ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [ 0.4s] THE JOB: sarm_bert_mse_solo_o0: the surface arm s13_bert (wordpiece spelling by bert-base-uncased; A = the text written; sites = tokens) alone on the bare trunk; form mse; blocks all; 4000 steps, a close every 1000; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; card 30.5 GB with 30.4 GB free; host RSS 0.6 GB
2
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
3
+ [ 17.4s] mini-beatrix-3 @ 245,674: trunk loaded from /workspace/qarm/dl/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); alephllm 0.10.7 at /workspace/qarm/alephllm/geolip/alephllm; amoe-lora 0.2.11; torch 2.8.0+cu128
4
+ [ 18.0s] tokenizer bert-base-uncased (wordpiece spelling; vocab 30522; sha256 9114256b104cca80): 'a taco truck' -> ['a', 'ta', '##co', 'truck']; A 12 bytes, B 15 bytes; sites [(1, 4, 4), (2, 6, 9)]
5
+ [ 18.0s] the read's captions: 512 of each draw of /workspace/qarm/dl/coco_caps_2x2048.json (4,090 captions of both draws kept out of training)
6
+ [ 31.3s] caption pack (/workspace/qarm/dl/mini-beatrix-3/caption_pack/pack_v2, train): dff 121,153 rows, scf 42,080 rows, danbooru 603,026 rows, cc12m 1,242,176 rows, coco 19,227 rows; host RSS 3.7 GB
7
+ [ 31.4s] streams: web text; partners none (two web rows a quiet chunk); held-out sets ['web'] x 8 batches
8
+ [ 31.5s] read rows draw1: 512 captions (0 unspellable), A rows (512, 99), B rows (512, 108), 4,966 sites
9
+ [ 31.5s] read rows draw2: 512 captions (0 unspellable), A rows (512, 151), B rows (512, 161), 5,109 sites
10
+ [arms] attached 's13_bert' for phase 'wordpiece_spelling': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on
11
+ [ 44.1s] BIRTH: the member attached over the trunk: web rows read exactly as before; the hand-run trunk against the model's forward: armed 0.0e+00, masked 0.0e+00 (gates: 0.0)
12
+ [ 48.4s] standardization statistics from 12,850 teacher sites on 8 paired batches (a mean and a spread per block and dimension); sigma range at block 20 0.748..1.117; saved /workspace/qarm/out/anchors/sarm_bert_mse_solo_o0_stats.pt
13
+ [ 255.0s] CLOSE 0 (step 0): the member's alignment read on draw 1 (512 captions, 4,966 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
14
+ [ 255.0s] block 8: 0.868 -> 0.868 / 0.868; A 1.0000; mse 0.190 cos 0.914
15
+ [ 255.0s] block 12: 0.872 -> 0.872 / 0.872; A 1.0000; mse 0.197 cos 0.910
16
+ [ 255.0s] block 16: 0.870 -> 0.870 / 0.870; A 1.0000; mse 0.203 cos 0.903
17
+ [ 255.0s] block 18: 0.859 -> 0.859 / 0.859; A 1.0000; mse 0.225 cos 0.892
18
+ [ 255.0s] block 20: 0.854 -> 0.854 / 0.854; A 1.0000; mse 0.246 cos 0.884
19
+ [ 255.0s] block 24: 0.832 -> 0.832 / 0.832; A 1.0000; mse 0.305 cos 0.864
20
+ [ 255.0s] block 28: 0.786 -> 0.786 / 0.786; A 1.0000; mse 0.403 cos 0.814
21
+ [ 255.0s] block 31: 0.674 -> 0.674 / 0.674; A 1.0000; mse 0.484 cos 0.772
22
+ [ 255.0s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.873 / 0.869 / 0.849 / 0.831 / 0.788
23
+ [ 255.0s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.674 against the bar .85; B2: the lowest A alignment 1.0000 against .995; fineweb share +0.00000 +- 0.00000 against +.012
24
+ [ 255.0s] gate mean 0.047 / widest 0.047; compiled chain: 2 cache entries (limit 256), 0 limit warnings; docs drawn 266 (skipped {'test': 2, 'spelling': 1, 'empty': 0, 'short': 0}); card peak 2.2 GB; host RSS 10.6 GB
25
+ [stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) …
26
+ [ 262.3s] step 1/4000: task 0.6231 (mse 0.623 cos 0.698); quiet A 0.00000 off 0.00000; 1,604 sites; 7.34 s/step; the job: 4.4 min spent, about 488.9 min of steps left; card peak 13.7 GB
27
+ [ 383.4s] step 50/4000: task 0.6639 (mse 0.664 cos 0.677); quiet A 0.00034 off 0.00025; 1,600 sites; 2.57 s/step; the job: 6.4 min spent, about 169.1 min of steps left; card peak 13.9 GB
28
+ [ 506.7s] step 100/4000: task 0.6441 (mse 0.644 cos 0.687); quiet A 0.00085 off 0.00049; 1,583 sites; 2.52 s/step; the job: 8.4 min spent, about 163.6 min of steps left; card peak 13.9 GB
29
+ [ 629.9s] step 150/4000: task 0.6262 (mse 0.626 cos 0.696); quiet A 0.00119 off 0.00063; 1,597 sites; 2.50 s/step; the job: 10.5 min spent, about 160.4 min of steps left; card peak 13.9 GB
30
+ [ 753.2s] step 200/4000: task 0.6075 (mse 0.608 cos 0.705); quiet A 0.00123 off 0.00060; 1,579 sites; 2.49 s/step; the job: 12.6 min spent, about 157.8 min of steps left; card peak 13.9 GB
31
+ [ 876.5s] step 250/4000: task 0.5834 (mse 0.583 cos 0.716); quiet A 0.00149 off 0.00072; 1,588 sites; 2.49 s/step; the job: 14.6 min spent, about 155.4 min of steps left; card peak 13.9 GB
32
+ [ 999.7s] step 300/4000: task 0.5750 (mse 0.575 cos 0.721); quiet A 0.00168 off 0.00085; 1,626 sites; 2.48 s/step; the job: 16.7 min spent, about 153.1 min of steps left; card peak 13.9 GB
33
+ [ 1122.7s] step 350/4000: task 0.5633 (mse 0.563 cos 0.727); quiet A 0.00167 off 0.00081; 1,656 sites; 2.48 s/step; the job: 18.7 min spent, about 150.8 min of steps left; card peak 13.9 GB
34
+ [ 1245.9s] step 400/4000: task 0.5261 (mse 0.526 cos 0.744); quiet A 0.00153 off 0.00063; 1,576 sites; 2.48 s/step; the job: 20.8 min spent, about 148.6 min of steps left; card peak 13.9 GB
35
+ [ 1369.3s] step 450/4000: task 0.5328 (mse 0.533 cos 0.742); quiet A 0.00135 off 0.00058; 1,588 sites; 2.48 s/step; the job: 22.8 min spent, about 146.5 min of steps left; card peak 13.9 GB
36
+ [ 1492.5s] step 500/4000: task 0.5191 (mse 0.519 cos 0.749); quiet A 0.00118 off 0.00050; 1,656 sites; 2.48 s/step; the job: 24.9 min spent, about 144.4 min of steps left; card peak 13.9 GB
37
+ [ 1615.9s] step 550/4000: task 0.4910 (mse 0.491 cos 0.763); quiet A 0.00128 off 0.00057; 1,615 sites; 2.47 s/step; the job: 26.9 min spent, about 142.3 min of steps left; card peak 13.9 GB
38
+ [ 1739.5s] step 600/4000: task 0.4825 (mse 0.482 cos 0.768); quiet A 0.00114 off 0.00059; 1,591 sites; 2.47 s/step; the job: 29.0 min spent, about 140.2 min of steps left; card peak 13.9 GB
39
+ [ 1862.9s] step 650/4000: task 0.4749 (mse 0.475 cos 0.772); quiet A 0.00114 off 0.00055; 1,615 sites; 2.47 s/step; the job: 31.0 min spent, about 138.1 min of steps left; card peak 13.9 GB
40
+ [transformers] Token indices sequence length is longer than the specified maximum sequence length for this model (558 > 512). Running this sequence through the model will result in indexing errors
41
+ [ 1986.2s] step 700/4000: task 0.4531 (mse 0.453 cos 0.783); quiet A 0.00111 off 0.00058; 1,596 sites; 2.47 s/step; the job: 33.1 min spent, about 136.0 min of steps left; card peak 13.9 GB
42
+ [ 2109.4s] step 750/4000: task 0.4355 (mse 0.436 cos 0.792); quiet A 0.00116 off 0.00064; 1,637 sites; 2.47 s/step; the job: 35.2 min spent, about 133.9 min of steps left; card peak 13.9 GB
43
+ [ 2232.8s] step 800/4000: task 0.4083 (mse 0.408 cos 0.805); quiet A 0.00122 off 0.00071; 1,602 sites; 2.47 s/step; the job: 37.2 min spent, about 131.9 min of steps left; card peak 13.9 GB
44
+ [ 2356.5s] step 850/4000: task 0.3983 (mse 0.398 cos 0.811); quiet A 0.00134 off 0.00070; 1,603 sites; 2.47 s/step; the job: 39.3 min spent, about 129.8 min of steps left; card peak 13.9 GB
45
+ [ 2480.1s] step 900/4000: task 0.3827 (mse 0.383 cos 0.819); quiet A 0.00142 off 0.00083; 1,588 sites; 2.47 s/step; the job: 41.3 min spent, about 127.7 min of steps left; card peak 13.9 GB
46
+ [ 2603.3s] step 950/4000: task 0.3721 (mse 0.372 cos 0.826); quiet A 0.00152 off 0.00072; 1,593 sites; 2.47 s/step; the job: 43.4 min spent, about 125.7 min of steps left; card peak 13.9 GB
47
+ [ 2726.9s] step 1000/4000: task 0.3609 (mse 0.361 cos 0.832); quiet A 0.00177 off 0.00092; 1,636 sites; 2.47 s/step; the job: 45.4 min spent, about 123.6 min of steps left; card peak 13.9 GB
48
+ [ 2727.1s] anchor saved: /workspace/qarm/out/anchors/sarm_bert_mse_solo_o0_close1_s13_bert.safetensors
49
+ [ 2881.7s] CLOSE 1 (step 1000): the member's alignment read on draw 1 (512 captions, 4,966 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
50
+ [ 2881.7s] block 8: 0.868 -> 0.881 / 0.881; A 0.9975; mse 0.153 cos 0.931
51
+ [ 2881.7s] block 12: 0.872 -> 0.894 / 0.895; A 0.9966; mse 0.152 cos 0.931
52
+ [ 2881.7s] block 16: 0.870 -> 0.896 / 0.898; A 0.9948; mse 0.152 cos 0.927
53
+ [ 2881.7s] block 18: 0.859 -> 0.891 / 0.893; A 0.9955; mse 0.164 cos 0.922
54
+ [ 2881.7s] block 20: 0.854 -> 0.886 / 0.890; A 0.9953; mse 0.175 cos 0.918
55
+ [ 2881.7s] block 24: 0.832 -> 0.869 / 0.873; A 0.9950; mse 0.217 cos 0.904
56
+ [ 2881.7s] block 28: 0.786 -> 0.827 / 0.831; A 0.9930; mse 0.297 cos 0.867
57
+ [ 2881.7s] block 31: 0.674 -> 0.709 / 0.715; A 0.9868; mse 0.386 cos 0.829
58
+ [ 2881.7s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.891 / 0.892 / 0.879 / 0.862 / 0.826
59
+ [ 2881.7s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.709 against the bar .85; B2: the lowest A alignment 0.9868 against .995; fineweb share +0.00048 +- 0.00014 against +.012
60
+ [ 2881.7s] gate mean 0.164 / widest 0.221; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 32,563 (skipped {'test': 44, 'spelling': 6, 'empty': 0, 'short': 0}); card peak 13.9 GB; host RSS 13.0 GB
61
+ [ 3004.9s] step 1050/4000: task 0.3473 (mse 0.347 cos 0.838); quiet A 0.00186 off 0.00094; 1,663 sites; 2.46 s/step; the job: 50.1 min spent, about 121.5 min of steps left; card peak 13.9 GB
62
+ [ 3128.2s] step 1100/4000: task 0.3355 (mse 0.335 cos 0.844); quiet A 0.00183 off 0.00094; 1,626 sites; 2.46 s/step; the job: 52.1 min spent, about 119.4 min of steps left; card peak 13.9 GB
63
+ [ 3251.9s] step 1150/4000: task 0.3274 (mse 0.327 cos 0.848); quiet A 0.00207 off 0.00107; 1,598 sites; 2.47 s/step; the job: 54.2 min spent, about 117.4 min of steps left; card peak 13.9 GB
64
+ [ 3375.2s] step 1200/4000: task 0.3229 (mse 0.323 cos 0.851); quiet A 0.00184 off 0.00099; 1,612 sites; 2.47 s/step; the job: 56.3 min spent, about 115.3 min of steps left; card peak 13.9 GB
65
+ [ 3498.6s] step 1250/4000: task 0.3245 (mse 0.325 cos 0.851); quiet A 0.00192 off 0.00098; 1,628 sites; 2.47 s/step; the job: 58.3 min spent, about 113.3 min of steps left; card peak 13.9 GB
66
+ [ 3622.1s] step 1300/4000: task 0.3207 (mse 0.321 cos 0.853); quiet A 0.00184 off 0.00095; 1,600 sites; 2.47 s/step; the job: 60.4 min spent, about 111.2 min of steps left; card peak 13.9 GB
67
+ [ 3745.4s] step 1350/4000: task 0.3139 (mse 0.314 cos 0.856); quiet A 0.00191 off 0.00100; 1,581 sites; 2.47 s/step; the job: 62.4 min spent, about 109.1 min of steps left; card peak 13.9 GB
68
+ [ 3868.6s] step 1400/4000: task 0.3092 (mse 0.309 cos 0.858); quiet A 0.00196 off 0.00100; 1,579 sites; 2.47 s/step; the job: 64.5 min spent, about 107.1 min of steps left; card peak 13.9 GB
69
+ [ 3992.2s] step 1450/4000: task 0.3118 (mse 0.312 cos 0.857); quiet A 0.00197 off 0.00099; 1,619 sites; 2.47 s/step; the job: 66.5 min spent, about 105.0 min of steps left; card peak 13.9 GB
70
+ [ 4115.7s] step 1500/4000: task 0.3010 (mse 0.301 cos 0.862); quiet A 0.00189 off 0.00104; 1,597 sites; 2.47 s/step; the job: 68.6 min spent, about 102.9 min of steps left; card peak 13.9 GB
71
+ [ 4239.2s] step 1550/4000: task 0.3001 (mse 0.300 cos 0.863); quiet A 0.00189 off 0.00096; 1,579 sites; 2.47 s/step; the job: 70.7 min spent, about 100.9 min of steps left; card peak 13.9 GB
72
+ [ 4362.5s] step 1600/4000: task 0.2945 (mse 0.294 cos 0.866); quiet A 0.00179 off 0.00087; 1,604 sites; 2.47 s/step; the job: 72.7 min spent, about 98.8 min of steps left; card peak 13.9 GB
73
+ [ 4485.7s] step 1650/4000: task 0.2933 (mse 0.293 cos 0.866); quiet A 0.00184 off 0.00093; 1,612 sites; 2.47 s/step; the job: 74.8 min spent, about 96.8 min of steps left; card peak 13.9 GB
74
+ [ 4609.1s] step 1700/4000: task 0.2956 (mse 0.296 cos 0.866); quiet A 0.00188 off 0.00102; 1,572 sites; 2.47 s/step; the job: 76.8 min spent, about 94.7 min of steps left; card peak 13.9 GB
75
+ [ 4732.8s] step 1750/4000: task 0.2833 (mse 0.283 cos 0.871); quiet A 0.00194 off 0.00099; 1,624 sites; 2.47 s/step; the job: 78.9 min spent, about 92.6 min of steps left; card peak 13.9 GB
76
+ [ 4856.3s] step 1800/4000: task 0.2820 (mse 0.282 cos 0.872); quiet A 0.00198 off 0.00095; 1,635 sites; 2.47 s/step; the job: 80.9 min spent, about 90.6 min of steps left; card peak 13.9 GB
77
+ [ 4979.8s] step 1850/4000: task 0.2900 (mse 0.290 cos 0.868); quiet A 0.00187 off 0.00097; 1,611 sites; 2.47 s/step; the job: 83.0 min spent, about 88.5 min of steps left; card peak 13.9 GB
78
+ [ 5103.2s] step 1900/4000: task 0.2818 (mse 0.282 cos 0.872); quiet A 0.00186 off 0.00095; 1,583 sites; 2.47 s/step; the job: 85.1 min spent, about 86.5 min of steps left; card peak 13.9 GB
79
+ [ 5226.6s] step 1950/4000: task 0.2772 (mse 0.277 cos 0.874); quiet A 0.00188 off 0.00103; 1,681 sites; 2.47 s/step; the job: 87.1 min spent, about 84.4 min of steps left; card peak 13.9 GB
80
+ [ 5349.9s] step 2000/4000: task 0.2712 (mse 0.271 cos 0.877); quiet A 0.00181 off 0.00127; 1,597 sites; 2.47 s/step; the job: 89.2 min spent, about 82.3 min of steps left; card peak 13.9 GB
81
+ [ 5350.1s] anchor saved: /workspace/qarm/out/anchors/sarm_bert_mse_solo_o0_close2_s13_bert.safetensors
82
+ [ 5504.9s] CLOSE 2 (step 2000): the member's alignment read on draw 1 (512 captions, 4,966 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
83
+ [ 5504.9s] block 8: 0.868 -> 0.888 / 0.889; A 0.9956; mse 0.138 cos 0.938
84
+ [ 5504.9s] block 12: 0.872 -> 0.901 / 0.903; A 0.9961; mse 0.136 cos 0.938
85
+ [ 5504.9s] block 16: 0.870 -> 0.905 / 0.908; A 0.9959; mse 0.135 cos 0.936
86
+ [ 5504.9s] block 18: 0.859 -> 0.901 / 0.903; A 0.9957; mse 0.145 cos 0.931
87
+ [ 5504.9s] block 20: 0.854 -> 0.896 / 0.899; A 0.9948; mse 0.155 cos 0.928
88
+ [ 5504.9s] block 24: 0.832 -> 0.883 / 0.885; A 0.9942; mse 0.193 cos 0.915
89
+ [ 5504.9s] block 28: 0.786 -> 0.843 / 0.847; A 0.9915; mse 0.268 cos 0.881
90
+ [ 5504.9s] block 31: 0.674 -> 0.725 / 0.729; A 0.9869; mse 0.360 cos 0.846
91
+ [ 5504.9s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.898 / 0.901 / 0.890 / 0.876 / 0.839
92
+ [ 5504.9s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.725 against the bar .85; B2: the lowest A alignment 0.9869 against .995; fineweb share +0.00084 +- 0.00010 against +.012
93
+ [ 5504.9s] gate mean 0.259 / widest 0.512; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 64,613 (skipped {'test': 97, 'spelling': 14, 'empty': 0, 'short': 0}); card peak 13.9 GB; host RSS 13.0 GB
94
+ [ 5628.8s] step 2050/4000: task 0.2738 (mse 0.274 cos 0.876); quiet A 0.00179 off 0.00096; 1,631 sites; 2.48 s/step; the job: 93.8 min spent, about 80.3 min of steps left; card peak 13.9 GB
95
+ [ 5752.0s] step 2100/4000: task 0.2765 (mse 0.276 cos 0.875); quiet A 0.00188 off 0.00098; 1,623 sites; 2.47 s/step; the job: 95.9 min spent, about 78.2 min of steps left; card peak 13.9 GB
96
+ [ 5875.4s] step 2150/4000: task 0.2635 (mse 0.263 cos 0.881); quiet A 0.00175 off 0.00094; 1,617 sites; 2.47 s/step; the job: 97.9 min spent, about 76.2 min of steps left; card peak 13.9 GB
97
+ [ 5998.8s] step 2200/4000: task 0.2767 (mse 0.277 cos 0.875); quiet A 0.00194 off 0.00092; 1,592 sites; 2.47 s/step; the job: 100.0 min spent, about 74.1 min of steps left; card peak 13.9 GB
98
+ [ 6122.1s] step 2250/4000: task 0.2769 (mse 0.277 cos 0.875); quiet A 0.00191 off 0.00099; 1,591 sites; 2.47 s/step; the job: 102.0 min spent, about 72.0 min of steps left; card peak 13.9 GB
99
+ [ 6245.7s] step 2300/4000: task 0.2682 (mse 0.268 cos 0.879); quiet A 0.00184 off 0.00094; 1,609 sites; 2.47 s/step; the job: 104.1 min spent, about 70.0 min of steps left; card peak 13.9 GB
100
+ [ 6369.0s] step 2350/4000: task 0.2658 (mse 0.266 cos 0.880); quiet A 0.00174 off 0.00092; 1,635 sites; 2.47 s/step; the job: 106.2 min spent, about 67.9 min of steps left; card peak 13.9 GB
101
+ [ 6492.3s] step 2400/4000: task 0.2538 (mse 0.254 cos 0.886); quiet A 0.00173 off 0.00105; 1,628 sites; 2.47 s/step; the job: 108.2 min spent, about 65.9 min of steps left; card peak 13.9 GB
102
+ [ 6615.6s] step 2450/4000: task 0.2599 (mse 0.260 cos 0.882); quiet A 0.00202 off 0.00103; 1,636 sites; 2.47 s/step; the job: 110.3 min spent, about 63.8 min of steps left; card peak 13.9 GB
103
+ [ 6739.0s] step 2500/4000: task 0.2601 (mse 0.260 cos 0.883); quiet A 0.00188 off 0.00097; 1,581 sites; 2.47 s/step; the job: 112.3 min spent, about 61.7 min of steps left; card peak 13.9 GB
104
+ [ 6862.4s] step 2550/4000: task 0.2629 (mse 0.263 cos 0.881); quiet A 0.00193 off 0.00102; 1,588 sites; 2.47 s/step; the job: 114.4 min spent, about 59.7 min of steps left; card peak 13.9 GB
105
+ [ 6985.9s] step 2600/4000: task 0.2656 (mse 0.266 cos 0.880); quiet A 0.00203 off 0.00104; 1,633 sites; 2.47 s/step; the job: 116.4 min spent, about 57.6 min of steps left; card peak 13.9 GB
106
+ [ 7109.3s] step 2650/4000: task 0.2513 (mse 0.251 cos 0.887); quiet A 0.00175 off 0.00101; 1,601 sites; 2.47 s/step; the job: 118.5 min spent, about 55.6 min of steps left; card peak 13.9 GB
107
+ [ 7232.8s] step 2700/4000: task 0.2589 (mse 0.259 cos 0.884); quiet A 0.00186 off 0.00099; 1,633 sites; 2.47 s/step; the job: 120.5 min spent, about 53.5 min of steps left; card peak 13.9 GB
108
+ [ 7356.4s] step 2750/4000: task 0.2527 (mse 0.253 cos 0.886); quiet A 0.00184 off 0.00101; 1,597 sites; 2.47 s/step; the job: 122.6 min spent, about 51.5 min of steps left; card peak 13.9 GB
109
+ [ 7479.6s] step 2800/4000: task 0.2515 (mse 0.251 cos 0.887); quiet A 0.00186 off 0.00095; 1,634 sites; 2.47 s/step; the job: 124.7 min spent, about 49.4 min of steps left; card peak 13.9 GB
110
+ [ 7603.1s] step 2850/4000: task 0.2528 (mse 0.253 cos 0.886); quiet A 0.00187 off 0.00093; 1,612 sites; 2.47 s/step; the job: 126.7 min spent, about 47.3 min of steps left; card peak 13.9 GB
111
+ [ 7726.4s] step 2900/4000: task 0.2536 (mse 0.254 cos 0.886); quiet A 0.00231 off 0.00096; 1,592 sites; 2.47 s/step; the job: 128.8 min spent, about 45.3 min of steps left; card peak 13.9 GB
112
+ [ 7849.6s] step 2950/4000: task 0.2451 (mse 0.245 cos 0.890); quiet A 0.00183 off 0.00099; 1,611 sites; 2.47 s/step; the job: 130.8 min spent, about 43.2 min of steps left; card peak 13.9 GB
113
+ [ 7972.8s] step 3000/4000: task 0.2567 (mse 0.257 cos 0.885); quiet A 0.00215 off 0.00112; 1,586 sites; 2.47 s/step; the job: 132.9 min spent, about 41.2 min of steps left; card peak 13.9 GB
114
+ [ 7973.1s] anchor saved: /workspace/qarm/out/anchors/sarm_bert_mse_solo_o0_close3_s13_bert.safetensors
115
+ [ 8127.8s] CLOSE 3 (step 3000): the member's alignment read on draw 1 (512 captions, 4,966 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
116
+ [ 8127.8s] block 8: 0.868 -> 0.893 / 0.893; A 0.9964; mse 0.132 cos 0.941
117
+ [ 8127.8s] block 12: 0.872 -> 0.905 / 0.907; A 0.9953; mse 0.129 cos 0.942
118
+ [ 8127.8s] block 16: 0.870 -> 0.912 / 0.913; A 0.9958; mse 0.127 cos 0.940
119
+ [ 8127.8s] block 18: 0.859 -> 0.908 / 0.908; A 0.9962; mse 0.135 cos 0.936
120
+ [ 8127.8s] block 20: 0.854 -> 0.903 / 0.905; A 0.9950; mse 0.144 cos 0.933
121
+ [ 8127.8s] block 24: 0.832 -> 0.892 / 0.894; A 0.9943; mse 0.178 cos 0.922
122
+ [ 8127.8s] block 28: 0.786 -> 0.852 / 0.855; A 0.9916; mse 0.249 cos 0.889
123
+ [ 8127.8s] block 31: 0.674 -> 0.740 / 0.744; A 0.9864; mse 0.335 cos 0.856
124
+ [ 8127.8s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.902 / 0.907 / 0.897 / 0.882 / 0.845
125
+ [ 8127.8s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.740 against the bar .85; B2: the lowest A alignment 0.9863 against .995; fineweb share +0.00049 +- 0.00020 against +.012
126
+ [ 8127.8s] gate mean 0.352 / widest 0.683; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 96,519 (skipped {'test': 156, 'spelling': 31, 'empty': 0, 'short': 0}); card peak 13.9 GB; host RSS 13.1 GB
127
+ [ 8251.0s] step 3050/4000: task 0.2438 (mse 0.244 cos 0.891); quiet A 0.00187 off 0.00094; 1,568 sites; 2.46 s/step; the job: 137.5 min spent, about 39.1 min of steps left; card peak 13.9 GB
128
+ [ 8374.4s] step 3100/4000: task 0.2481 (mse 0.248 cos 0.889); quiet A 0.00180 off 0.00103; 1,569 sites; 2.47 s/step; the job: 139.6 min spent, about 37.0 min of steps left; card peak 13.9 GB
129
+ [ 8498.5s] step 3150/4000: task 0.2510 (mse 0.251 cos 0.888); quiet A 0.00180 off 0.00091; 1,649 sites; 2.47 s/step; the job: 141.6 min spent, about 35.0 min of steps left; card peak 13.9 GB
130
+ [ 8622.0s] step 3200/4000: task 0.2454 (mse 0.245 cos 0.890); quiet A 0.00177 off 0.00102; 1,600 sites; 2.47 s/step; the job: 143.7 min spent, about 32.9 min of steps left; card peak 13.9 GB
131
+ [ 8745.5s] step 3250/4000: task 0.2441 (mse 0.244 cos 0.891); quiet A 0.00220 off 0.00118; 1,636 sites; 2.47 s/step; the job: 145.8 min spent, about 30.9 min of steps left; card peak 13.9 GB
132
+ [ 8869.0s] step 3300/4000: task 0.2365 (mse 0.237 cos 0.894); quiet A 0.00193 off 0.00104; 1,677 sites; 2.47 s/step; the job: 147.8 min spent, about 28.8 min of steps left; card peak 13.9 GB
133
+ [ 8992.5s] step 3350/4000: task 0.2481 (mse 0.248 cos 0.889); quiet A 0.00183 off 0.00091; 1,665 sites; 2.47 s/step; the job: 149.9 min spent, about 26.8 min of steps left; card peak 13.9 GB
134
+ [ 9116.0s] step 3400/4000: task 0.2396 (mse 0.240 cos 0.893); quiet A 0.00173 off 0.00091; 1,647 sites; 2.47 s/step; the job: 151.9 min spent, about 24.7 min of steps left; card peak 13.9 GB
135
+ [ 9239.8s] step 3450/4000: task 0.2393 (mse 0.239 cos 0.893); quiet A 0.00177 off 0.00100; 1,636 sites; 2.47 s/step; the job: 154.0 min spent, about 22.6 min of steps left; card peak 13.9 GB
136
+ [ 9363.3s] step 3500/4000: task 0.2391 (mse 0.239 cos 0.893); quiet A 0.00175 off 0.00097; 1,590 sites; 2.47 s/step; the job: 156.1 min spent, about 20.6 min of steps left; card peak 13.9 GB
137
+ [ 9486.5s] step 3550/4000: task 0.2369 (mse 0.237 cos 0.894); quiet A 0.00171 off 0.00086; 1,607 sites; 2.47 s/step; the job: 158.1 min spent, about 18.5 min of steps left; card peak 13.9 GB
138
+ [ 9609.8s] step 3600/4000: task 0.2345 (mse 0.234 cos 0.895); quiet A 0.00173 off 0.00093; 1,644 sites; 2.47 s/step; the job: 160.2 min spent, about 16.5 min of steps left; card peak 13.9 GB
139
+ [ 9733.3s] step 3650/4000: task 0.2383 (mse 0.238 cos 0.894); quiet A 0.00173 off 0.00096; 1,655 sites; 2.47 s/step; the job: 162.2 min spent, about 14.4 min of steps left; card peak 13.9 GB
140
+ [ 9856.6s] step 3700/4000: task 0.2348 (mse 0.235 cos 0.895); quiet A 0.00176 off 0.00103; 1,578 sites; 2.47 s/step; the job: 164.3 min spent, about 12.3 min of steps left; card peak 13.9 GB
141
+ [ 9980.1s] step 3750/4000: task 0.2293 (mse 0.229 cos 0.898); quiet A 0.00175 off 0.00091; 1,606 sites; 2.47 s/step; the job: 166.3 min spent, about 10.3 min of steps left; card peak 13.9 GB
142
+ [10103.3s] step 3800/4000: task 0.2303 (mse 0.230 cos 0.897); quiet A 0.00171 off 0.00085; 1,597 sites; 2.47 s/step; the job: 168.4 min spent, about 8.2 min of steps left; card peak 13.9 GB
143
+ [10226.6s] step 3850/4000: task 0.2325 (mse 0.233 cos 0.896); quiet A 0.00181 off 0.00087; 1,618 sites; 2.47 s/step; the job: 170.4 min spent, about 6.2 min of steps left; card peak 13.9 GB
144
+ [10349.6s] step 3900/4000: task 0.2251 (mse 0.225 cos 0.900); quiet A 0.00171 off 0.00088; 1,585 sites; 2.47 s/step; the job: 172.5 min spent, about 4.1 min of steps left; card peak 13.9 GB
145
+ [10472.7s] step 3950/4000: task 0.2260 (mse 0.226 cos 0.899); quiet A 0.00166 off 0.00088; 1,578 sites; 2.47 s/step; the job: 174.5 min spent, about 2.1 min of steps left; card peak 13.9 GB
146
+ [10596.2s] step 4000/4000: task 0.2343 (mse 0.234 cos 0.896); quiet A 0.00172 off 0.00091; 1,588 sites; 2.47 s/step; the job: 176.6 min spent, about 0.0 min of steps left; card peak 13.9 GB
147
+ [10596.5s] anchor saved: /workspace/qarm/out/anchors/sarm_bert_mse_solo_o0_close4_s13_bert.safetensors
148
+ [10751.3s] CLOSE 4 (step 4000): the member's alignment read on draw 1 (512 captions, 4,966 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
149
+ [10751.3s] block 8: 0.868 -> 0.895 / 0.895; A 0.9966; mse 0.130 cos 0.942
150
+ [10751.3s] block 12: 0.872 -> 0.907 / 0.908; A 0.9963; mse 0.127 cos 0.943
151
+ [10751.3s] block 16: 0.870 -> 0.915 / 0.915; A 0.9962; mse 0.123 cos 0.942
152
+ [10751.3s] block 18: 0.859 -> 0.912 / 0.912; A 0.9964; mse 0.130 cos 0.938
153
+ [10751.3s] block 20: 0.854 -> 0.906 / 0.907; A 0.9952; mse 0.138 cos 0.936
154
+ [10751.3s] block 24: 0.832 -> 0.895 / 0.896; A 0.9951; mse 0.172 cos 0.925
155
+ [10751.3s] block 28: 0.786 -> 0.854 / 0.858; A 0.9923; mse 0.242 cos 0.893
156
+ [10751.3s] block 31: 0.674 -> 0.742 / 0.747; A 0.9888; mse 0.333 cos 0.861
157
+ [10751.3s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.903 / 0.907 / 0.898 / 0.884 / 0.847
158
+ [10751.3s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.742 against the bar .85; B2: the lowest A alignment 0.9867 against .995; fineweb share +0.00055 +- 0.00018 against +.012
159
+ [10751.3s] gate mean 0.435 / widest 0.834; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 128,678 (skipped {'test': 203, 'spelling': 48, 'empty': 0, 'short': 0}); card peak 13.9 GB; host RSS 13.1 GB
160
+ [10751.4s] DONE sarm_bert_mse_solo_o0: 4000 steps in 164.6 min of steps (2.47 s/step); 4 closes; 10751 s in all; card peak 13.9 GB; host RSS 13.1 GB; compiled chain: 5 cache entries (limit 256), 0 limit warnings