Hugging Face's logo

AbstractPhil
/
beatrix-tokenizers

geolip-alephllm
byte-level
adapter
tokenizer
beatrix
Model card Files
xet
 Community
AbstractPhil commited on
Commit
fe18f65
·
verified ·
1 Parent(s): 17354da

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

Browse files
Files changed (10) hide show
  1. surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close1_s12_clip.safetensors +3 -0
  2. surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close2_s12_clip.safetensors +3 -0
  3. surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close3_s12_clip.safetensors +3 -0
  4. surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close4_s12_clip.safetensors +3 -0
  5. surface/clip/mse_solo_o0/manifest.json +20 -0
  6. surface/clip/mse_solo_o0/readout.md +127 -0
  7. surface/clip/mse_solo_o0/results.json +0 -0
  8. surface/clip/mse_solo_o0/s12_clip.safetensors +3 -0
  9. surface/clip/mse_solo_o0/stats.pt +3 -0
  10. surface/clip/mse_solo_o0/train.log +160 -0
surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close1_s12_clip.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e780ba3e5a70792bf5b05bafff862051776e5d98eedfa704c1a76de755ad717
3
+ size 54819312
surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close2_s12_clip.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25899a7ccc4d969be79022af5d5f46942dfeea7dd6b851f32259bee768fbbcd1
3
+ size 54819312
surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close3_s12_clip.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b6c2dea1756424ac463e490611c075f62644819df9552cece9adc9183ec5359
3
+ size 54819312
surface/clip/mse_solo_o0/closes/sarm_clip_mse_solo_o0_close4_s12_clip.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:afae92bf86cd6ea9b675dd80b53dff475d96936d36d8a5a86c7b45f84daff031
3
+ size 54819312
surface/clip/mse_solo_o0/manifest.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "sarm_clip_mse_solo_o0",
3
+ "run": "mse_solo_o0",
4
+ "convention": "clip",
5
+ "member": "s12_clip",
6
+ "final_anchor": "sarm_clip_mse_solo_o0_close4_s12_clip.safetensors",
7
+ "closes": [
8
+ "sarm_clip_mse_solo_o0_close1_s12_clip.safetensors",
9
+ "sarm_clip_mse_solo_o0_close2_s12_clip.safetensors",
10
+ "sarm_clip_mse_solo_o0_close3_s12_clip.safetensors",
11
+ "sarm_clip_mse_solo_o0_close4_s12_clip.safetensors"
12
+ ],
13
+ "finished": true,
14
+ "form": "mse",
15
+ "base": "solo",
16
+ "steps": 4000,
17
+ "trunk_step": 245674,
18
+ "shipped": "2026-10-08 17:37 local",
19
+ "note": "every close's anchor ships (mids included); the final close's anchor is the run's s12_clip.safetensors"
20
+ }
surface/clip/mse_solo_o0/readout.md ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # The surface arm's readout: clip-vit-large-patch14's spelling (CLIP's 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_clip_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_clip_mse_solo_o0
12
+
13
+ - B1 the gap: NOT MET over every served block >= 12 (the lowest 0.784 at block 12 on draw 1, 0.785 at block 12 on draw 2; bar 0.85); over blocks 12-28 alone: NOT MET
14
+ - B2 silence, whitened: NOT MET (the lowest A alignment 0.9843 at block 31 on draw 1, 0.9819 on draw 2; bar 0.995)
15
+ - B2 silence, in bits: MET (web +0.00029 bpb; bar +0.012)
16
+
17
+ ## The closes
18
+
19
+ ### sarm_clip_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 | 0.289 / 0.297 | 0.431 / 0.373 / 0.310 | 1.0000 / 1.0000 | +0.00000 | 0.047 / 0.047 |
24
+ | 1 | 1000 | 0.674 / 0.674 | 0.674 / 0.761 / 0.762 | 0.9695 / 0.9639 | +0.00009 | 0.140 / 0.165 |
25
+ | 2 | 2000 | 0.723 / 0.722 | 0.723 / 0.839 / 0.846 | 0.9755 / 0.9750 | +0.00027 | 0.202 / 0.282 |
26
+ | 3 | 3000 | 0.752 / 0.753 | 0.752 / 0.873 / 0.879 | 0.9829 / 0.9792 | +0.00014 | 0.282 / 0.513 |
27
+ | 4 | 4000 | 0.784 / 0.785 | 0.784 / 0.896 / 0.902 | 0.9843 / 0.9819 | +0.00029 | 0.365 / 0.705 |
28
+
29
+ ## Every block at the final close
30
+
31
+ ### sarm_clip_mse_solo_o0, draw1 (512 captions, 4,871 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 | 0.240 | 0.232 | 0.232 | 0.9997 | 1.179 | 0.198 |
36
+ | 4 | 0.511 | 0.595 | 0.597 | 0.9949 | 0.477 | 0.691 |
37
+ | 8 | 0.479 | 0.676 | 0.680 | 0.9944 | 0.382 | 0.777 |
38
+ | 12 | 0.431 | 0.784 | 0.789 | 0.9937 | 0.304 | 0.842 |
39
+ | 16 | 0.400 | 0.864 | 0.868 | 0.9932 | 0.214 | 0.894 |
40
+ | 18 | 0.388 | 0.884 | 0.886 | 0.9922 | 0.186 | 0.909 |
41
+ | 20 | 0.373 | 0.896 | 0.898 | 0.9915 | 0.167 | 0.920 |
42
+ | 22 | 0.386 | 0.907 | 0.909 | 0.9907 | 0.149 | 0.932 |
43
+ | 24 | 0.383 | 0.910 | 0.910 | 0.9929 | 0.139 | 0.935 |
44
+ | 26 | 0.349 | 0.910 | 0.911 | 0.9914 | 0.133 | 0.936 |
45
+ | 28 | 0.310 | 0.902 | 0.903 | 0.9897 | 0.137 | 0.931 |
46
+ | 29 | 0.294 | 0.896 | 0.896 | 0.9866 | 0.140 | 0.928 |
47
+ | 30 | 0.289 | 0.886 | 0.889 | 0.9881 | 0.142 | 0.922 |
48
+ | 31 | 0.292 | 0.886 | 0.889 | 0.9843 | 0.139 | 0.916 |
49
+
50
+ ### sarm_clip_mse_solo_o0, draw2 (512 captions, 5,023 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 | 0.259 | 0.251 | 0.250 | 0.9997 | 1.180 | 0.192 |
55
+ | 4 | 0.510 | 0.601 | 0.603 | 0.9930 | 0.475 | 0.689 |
56
+ | 8 | 0.484 | 0.680 | 0.685 | 0.9923 | 0.372 | 0.780 |
57
+ | 12 | 0.438 | 0.785 | 0.790 | 0.9905 | 0.294 | 0.845 |
58
+ | 16 | 0.412 | 0.867 | 0.870 | 0.9914 | 0.206 | 0.896 |
59
+ | 18 | 0.398 | 0.886 | 0.889 | 0.9923 | 0.181 | 0.911 |
60
+ | 20 | 0.380 | 0.893 | 0.895 | 0.9918 | 0.165 | 0.920 |
61
+ | 22 | 0.394 | 0.902 | 0.904 | 0.9905 | 0.150 | 0.930 |
62
+ | 24 | 0.387 | 0.904 | 0.908 | 0.9899 | 0.141 | 0.934 |
63
+ | 26 | 0.352 | 0.901 | 0.904 | 0.9917 | 0.138 | 0.934 |
64
+ | 28 | 0.314 | 0.895 | 0.899 | 0.9887 | 0.143 | 0.927 |
65
+ | 29 | 0.300 | 0.890 | 0.893 | 0.9876 | 0.147 | 0.923 |
66
+ | 30 | 0.297 | 0.879 | 0.884 | 0.9845 | 0.153 | 0.916 |
67
+ | 31 | 0.301 | 0.873 | 0.878 | 0.9819 | 0.149 | 0.910 |
68
+
69
+ The member's share in bits per byte (all on minus the member masked): web +0.00029
70
+
71
+ ## The training curves
72
+
73
+ ### sarm_clip_mse_solo_o0
74
+
75
+ | step | task loss | mse | cos | InfoNCE | quiet on A | quiet off-domain |
76
+ |---|---|---|---|---|---|---|
77
+ | 1 | 2.1248 | 2.125 | 0.119 | None | 0.00000 | 0.00000 |
78
+ | 50 | 2.1218 | 2.122 | 0.115 | None | 0.00030 | 0.00023 |
79
+ | 250 | 1.9374 | 1.937 | 0.156 | None | 0.00149 | 0.00098 |
80
+ | 500 | 1.3421 | 1.342 | 0.364 | None | 0.00130 | 0.00049 |
81
+ | 1000 | 0.8953 | 0.895 | 0.552 | None | 0.00108 | 0.00056 |
82
+ | 1500 | 0.6977 | 0.698 | 0.638 | None | 0.00106 | 0.00066 |
83
+ | 2000 | 0.5651 | 0.565 | 0.703 | None | 0.00114 | 0.00071 |
84
+ | 2500 | 0.4772 | 0.477 | 0.745 | None | 0.00114 | 0.00068 |
85
+ | 3000 | 0.4200 | 0.420 | 0.774 | None | 0.00117 | 0.00076 |
86
+ | 3500 | 0.3590 | 0.359 | 0.803 | None | 0.00113 | 0.00072 |
87
+ | 4000 | 0.3151 | 0.315 | 0.826 | None | 0.00110 | 0.00068 |
88
+ | 1 | 2.1248 | 2.125 | 0.119 | None | 0.00000 | 0.00000 |
89
+ | 50 | 2.1218 | 2.122 | 0.115 | None | 0.00030 | 0.00023 |
90
+ | 250 | 1.9374 | 1.937 | 0.156 | None | 0.00149 | 0.00098 |
91
+ | 500 | 1.3421 | 1.342 | 0.364 | None | 0.00130 | 0.00049 |
92
+ | 1000 | 0.8953 | 0.895 | 0.552 | None | 0.00108 | 0.00056 |
93
+ | 1500 | 0.6977 | 0.698 | 0.638 | None | 0.00106 | 0.00066 |
94
+ | 2000 | 0.5651 | 0.565 | 0.703 | None | 0.00114 | 0.00071 |
95
+ | 2500 | 0.4772 | 0.477 | 0.745 | None | 0.00114 | 0.00068 |
96
+ | 3000 | 0.4200 | 0.420 | 0.774 | None | 0.00117 | 0.00076 |
97
+ | 3500 | 0.3590 | 0.359 | 0.803 | None | 0.00113 | 0.00072 |
98
+ | 4000 | 0.3151 | 0.315 | 0.826 | None | 0.00110 | 0.00068 |
99
+ | 1 | 2.1248 | 2.125 | 0.119 | None | 0.00000 | 0.00000 |
100
+ | 50 | 2.1218 | 2.122 | 0.115 | None | 0.00030 | 0.00023 |
101
+ | 250 | 1.9374 | 1.937 | 0.156 | None | 0.00149 | 0.00098 |
102
+ | 500 | 1.3421 | 1.342 | 0.364 | None | 0.00130 | 0.00049 |
103
+ | 1000 | 0.8953 | 0.895 | 0.552 | None | 0.00108 | 0.00056 |
104
+ | 1500 | 0.6977 | 0.698 | 0.638 | None | 0.00106 | 0.00066 |
105
+ | 2000 | 0.5651 | 0.565 | 0.703 | None | 0.00114 | 0.00071 |
106
+ | 2500 | 0.4772 | 0.477 | 0.745 | None | 0.00114 | 0.00068 |
107
+ | 3000 | 0.4200 | 0.420 | 0.774 | None | 0.00117 | 0.00076 |
108
+ | 3500 | 0.3590 | 0.359 | 0.803 | None | 0.00113 | 0.00072 |
109
+ | 4000 | 0.3151 | 0.315 | 0.826 | None | 0.00110 | 0.00068 |
110
+ | 1 | 2.1248 | 2.125 | 0.119 | None | 0.00000 | 0.00000 |
111
+ | 50 | 2.1218 | 2.122 | 0.115 | None | 0.00030 | 0.00023 |
112
+ | 250 | 1.9374 | 1.937 | 0.156 | None | 0.00149 | 0.00098 |
113
+ | 500 | 1.3421 | 1.342 | 0.364 | None | 0.00130 | 0.00049 |
114
+ | 1000 | 0.8953 | 0.895 | 0.552 | None | 0.00108 | 0.00056 |
115
+ | 1500 | 0.6977 | 0.698 | 0.638 | None | 0.00106 | 0.00066 |
116
+ | 2000 | 0.5651 | 0.565 | 0.703 | None | 0.00114 | 0.00071 |
117
+ | 2500 | 0.4772 | 0.477 | 0.745 | None | 0.00114 | 0.00068 |
118
+ | 3000 | 0.4200 | 0.420 | 0.774 | None | 0.00117 | 0.00076 |
119
+ | 3500 | 0.3590 | 0.359 | 0.803 | None | 0.00113 | 0.00072 |
120
+ | 4000 | 0.3151 | 0.315 | 0.826 | None | 0.00110 | 0.00068 |
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: clip-vit-large-patch14's spelling (CLIP's convention; sha256 4a07de0310c7cbea..., vocab 49,408; 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/clip/mse_solo_o0/results.json ADDED
The diff for this file is too large to render. See raw diff
 
surface/clip/mse_solo_o0/s12_clip.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:afae92bf86cd6ea9b675dd80b53dff475d96936d36d8a5a86c7b45f84daff031
3
+ size 54819312
surface/clip/mse_solo_o0/stats.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d59c176ef1cb2f93c2314b8ff1884509cc99303dbd31b3300b54d479d6d5267
3
+ size 264325
surface/clip/mse_solo_o0/train.log ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [ 0.3s] THE JOB: sarm_clip_mse_solo_o0: the surface arm s12_clip (clip spelling by openai/clip-vit-large-patch14; 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
+ [ 16.2s] 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
+ [ 16.5s] tokenizer openai/clip-vit-large-patch14 (clip spelling; vocab 49408; sha256 4a07de0310c7cbea): 'a taco truck' -> ['a</w>', 'taco</w>', 'truck</w>']; A 12 bytes, B 22 bytes; sites [(1, 6, 13)]
5
+ [ 16.5s] 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
+ [ 29.6s] 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 3.6 GB
7
+ [ 29.7s] streams: web text; partners none (two web rows a quiet chunk); held-out sets ['web'] x 8 batches
8
+ [ 29.8s] read rows draw1: 512 captions (0 unspellable), A rows (512, 99), B rows (512, 169), 4,871 sites
9
+ [ 29.9s] read rows draw2: 512 captions (0 unspellable), A rows (512, 151), B rows (512, 246), 5,023 sites
10
+ [arms] attached 's12_clip' for phase 'clip_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
+ [transformers] Token indices sequence length is longer than the specified maximum sequence length for this model (98 > 77). Running this sequence through the model will result in indexing errors
12
+ [ 41.3s] 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)
13
+ [ 44.5s] standardization statistics from 8,173 teacher sites on 8 paired batches (a mean and a spread per block and dimension); sigma range at block 20 0.745..1.103; saved /workspace/qarm/out/anchors/sarm_clip_mse_solo_o0_stats.pt
14
+ [ 248.1s] CLOSE 0 (step 0): the member's alignment read on draw 1 (512 captions, 4,871 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
15
+ [ 248.1s] block 8: 0.479 -> 0.479 / 0.479; A 1.0000; mse 1.928 cos 0.173
16
+ [ 248.1s] block 12: 0.431 -> 0.431 / 0.431; A 1.0000; mse 1.812 cos 0.159
17
+ [ 248.1s] block 16: 0.400 -> 0.400 / 0.400; A 1.0000; mse 1.715 cos 0.144
18
+ [ 248.1s] block 18: 0.388 -> 0.388 / 0.388; A 1.0000; mse 1.681 cos 0.144
19
+ [ 248.1s] block 20: 0.373 -> 0.373 / 0.373; A 1.0000; mse 1.715 cos 0.137
20
+ [ 248.1s] block 24: 0.383 -> 0.383 / 0.383; A 1.0000; mse 1.944 cos 0.140
21
+ [ 248.1s] block 28: 0.310 -> 0.310 / 0.310; A 1.0000; mse 2.120 cos 0.106
22
+ [ 248.1s] block 31: 0.292 -> 0.292 / 0.292; A 1.0000; mse 1.688 cos 0.077
23
+ [ 248.1s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.438 / 0.412 / 0.380 / 0.387 / 0.314
24
+ [ 248.1s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.289 against the bar .85; B2: the lowest A alignment 1.0000 against .995; fineweb share +0.00000 +- 0.00000 against +.012
25
+ [ 248.1s] gate mean 0.047 / widest 0.047; compiled chain: 2 cache entries (limit 256), 0 limit warnings; docs drawn 266 (skipped {'test': 2, 'spelling': 5, 'empty': 0, 'short': 0}); card peak 2.3 GB; host RSS 10.2 GB
26
+ [stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) …
27
+ [ 256.2s] step 1/4000: task 2.1248 (mse 2.125 cos 0.119); quiet A 0.00000 off 0.00000; 1,021 sites; 8.05 s/step; the job: 4.3 min spent, about 536.7 min of steps left; card peak 13.5 GB
28
+ [ 376.3s] step 50/4000: task 2.1218 (mse 2.122 cos 0.115); quiet A 0.00030 off 0.00023; 1,035 sites; 2.56 s/step; the job: 6.3 min spent, about 168.8 min of steps left; card peak 13.6 GB
29
+ [ 498.7s] step 100/4000: task 2.1029 (mse 2.103 cos 0.117); quiet A 0.00040 off 0.00029; 977 sites; 2.51 s/step; the job: 8.3 min spent, about 162.9 min of steps left; card peak 13.6 GB
30
+ [ 620.9s] step 150/4000: task 2.0638 (mse 2.064 cos 0.125); quiet A 0.00058 off 0.00038; 1,034 sites; 2.48 s/step; the job: 10.3 min spent, about 159.4 min of steps left; card peak 13.6 GB
31
+ [ 743.1s] step 200/4000: task 2.0085 (mse 2.009 cos 0.139); quiet A 0.00100 off 0.00062; 1,011 sites; 2.47 s/step; the job: 12.4 min spent, about 156.7 min of steps left; card peak 13.6 GB
32
+ [ 865.2s] step 250/4000: task 1.9374 (mse 1.937 cos 0.156); quiet A 0.00149 off 0.00098; 1,027 sites; 2.47 s/step; the job: 14.4 min spent, about 154.3 min of steps left; card peak 13.6 GB
33
+ [ 987.5s] step 300/4000: task 1.8408 (mse 1.841 cos 0.187); quiet A 0.00176 off 0.00112; 1,019 sites; 2.46 s/step; the job: 16.5 min spent, about 152.0 min of steps left; card peak 13.6 GB
34
+ [ 1109.9s] step 350/4000: task 1.7177 (mse 1.718 cos 0.225); quiet A 0.00172 off 0.00100; 1,024 sites; 2.46 s/step; the job: 18.5 min spent, about 149.8 min of steps left; card peak 13.6 GB
35
+ [ 1232.2s] step 400/4000: task 1.5776 (mse 1.578 cos 0.268); quiet A 0.00144 off 0.00080; 1,023 sites; 2.46 s/step; the job: 20.5 min spent, about 147.6 min of steps left; card peak 13.6 GB
36
+ [ 1354.6s] step 450/4000: task 1.4383 (mse 1.438 cos 0.321); quiet A 0.00136 off 0.00055; 1,038 sites; 2.46 s/step; the job: 22.6 min spent, about 145.5 min of steps left; card peak 13.6 GB
37
+ [ 1476.9s] step 500/4000: task 1.3421 (mse 1.342 cos 0.364); quiet A 0.00130 off 0.00049; 1,030 sites; 2.46 s/step; the job: 24.6 min spent, about 143.4 min of steps left; card peak 13.6 GB
38
+ [ 1599.2s] step 550/4000: task 1.2645 (mse 1.265 cos 0.398); quiet A 0.00118 off 0.00049; 1,045 sites; 2.46 s/step; the job: 26.7 min spent, about 141.2 min of steps left; card peak 13.6 GB
39
+ [ 1721.7s] step 600/4000: task 1.1985 (mse 1.199 cos 0.427); quiet A 0.00110 off 0.00052; 1,019 sites; 2.46 s/step; the job: 28.7 min spent, about 139.2 min of steps left; card peak 13.6 GB
40
+ [ 1844.1s] step 650/4000: task 1.1410 (mse 1.141 cos 0.451); quiet A 0.00105 off 0.00047; 1,003 sites; 2.46 s/step; the job: 30.7 min spent, about 137.1 min of steps left; card peak 13.6 GB
41
+ [ 1966.4s] step 700/4000: task 1.0896 (mse 1.090 cos 0.473); quiet A 0.00104 off 0.00048; 1,035 sites; 2.45 s/step; the job: 32.8 min spent, about 135.0 min of steps left; card peak 13.6 GB
42
+ [ 2088.7s] step 750/4000: task 1.0472 (mse 1.047 cos 0.491); quiet A 0.00099 off 0.00051; 1,035 sites; 2.45 s/step; the job: 34.8 min spent, about 132.9 min of steps left; card peak 13.6 GB
43
+ [ 2210.9s] step 800/4000: task 1.0097 (mse 1.010 cos 0.506); quiet A 0.00100 off 0.00052; 1,013 sites; 2.45 s/step; the job: 36.8 min spent, about 130.9 min of steps left; card peak 13.6 GB
44
+ [ 2333.6s] step 850/4000: task 0.9793 (mse 0.979 cos 0.519); quiet A 0.00107 off 0.00053; 1,021 sites; 2.45 s/step; the job: 38.9 min spent, about 128.8 min of steps left; card peak 13.6 GB
45
+ [ 2456.2s] step 900/4000: task 0.9506 (mse 0.951 cos 0.531); quiet A 0.00114 off 0.00062; 1,038 sites; 2.45 s/step; the job: 40.9 min spent, about 126.8 min of steps left; card peak 13.6 GB
46
+ [ 2578.4s] step 950/4000: task 0.9233 (mse 0.923 cos 0.542); quiet A 0.00114 off 0.00054; 1,022 sites; 2.45 s/step; the job: 43.0 min spent, about 124.7 min of steps left; card peak 13.6 GB
47
+ [ 2700.7s] step 1000/4000: task 0.8953 (mse 0.895 cos 0.552); quiet A 0.00108 off 0.00056; 1,012 sites; 2.45 s/step; the job: 45.0 min spent, about 122.6 min of steps left; card peak 13.6 GB
48
+ [ 2700.9s] anchor saved: /workspace/qarm/out/anchors/sarm_clip_mse_solo_o0_close1_s12_clip.safetensors
49
+ [ 2855.7s] CLOSE 1 (step 1000): the member's alignment read on draw 1 (512 captions, 4,871 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
50
+ [ 2855.7s] block 8: 0.479 -> 0.608 / 0.614; A 0.9810; mse 0.839 cos 0.515
51
+ [ 2855.7s] block 12: 0.431 -> 0.674 / 0.683; A 0.9783; mse 0.695 cos 0.619
52
+ [ 2855.7s] block 16: 0.400 -> 0.715 / 0.719; A 0.9786; mse 0.570 cos 0.699
53
+ [ 2855.7s] block 18: 0.388 -> 0.741 / 0.745; A 0.9816; mse 0.528 cos 0.726
54
+ [ 2855.7s] block 20: 0.373 -> 0.761 / 0.766; A 0.9792; mse 0.484 cos 0.754
55
+ [ 2855.7s] block 24: 0.383 -> 0.787 / 0.793; A 0.9796; mse 0.412 cos 0.798
56
+ [ 2855.7s] block 28: 0.310 -> 0.762 / 0.767; A 0.9761; mse 0.400 cos 0.787
57
+ [ 2855.7s] block 31: 0.292 -> 0.733 / 0.741; A 0.9695; mse 0.381 cos 0.750
58
+ [ 2855.7s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.674 / 0.713 / 0.761 / 0.781 / 0.760
59
+ [ 2855.7s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.674 against the bar .85; B2: the lowest A alignment 0.9695 against .995; fineweb share +0.00009 +- 0.00021 against +.012
60
+ [ 2855.7s] gate mean 0.140 / widest 0.165; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 32,969 (skipped {'test': 45, 'spelling': 549, 'empty': 0, 'short': 0}); card peak 13.6 GB; host RSS 13.4 GB
61
+ [ 2978.1s] step 1050/4000: task 0.8707 (mse 0.871 cos 0.562); quiet A 0.00113 off 0.00059; 1,006 sites; 2.45 s/step; the job: 49.6 min spent, about 120.6 min of steps left; card peak 13.6 GB
62
+ [ 3100.4s] step 1100/4000: task 0.8467 (mse 0.847 cos 0.571); quiet A 0.00110 off 0.00053; 1,066 sites; 2.45 s/step; the job: 51.7 min spent, about 118.5 min of steps left; card peak 13.6 GB
63
+ [ 3222.9s] step 1150/4000: task 0.8257 (mse 0.826 cos 0.581); quiet A 0.00114 off 0.00055; 1,019 sites; 2.45 s/step; the job: 53.7 min spent, about 116.5 min of steps left; card peak 13.6 GB
64
+ [ 3345.1s] step 1200/4000: task 0.8041 (mse 0.804 cos 0.590); quiet A 0.00114 off 0.00059; 1,038 sites; 2.45 s/step; the job: 55.8 min spent, about 114.4 min of steps left; card peak 13.6 GB
65
+ [ 3467.3s] step 1250/4000: task 0.7856 (mse 0.786 cos 0.598); quiet A 0.00112 off 0.00064; 991 sites; 2.45 s/step; the job: 57.8 min spent, about 112.3 min of steps left; card peak 13.6 GB
66
+ [ 3589.7s] step 1300/4000: task 0.7713 (mse 0.771 cos 0.604); quiet A 0.00109 off 0.00058; 1,030 sites; 2.45 s/step; the job: 59.8 min spent, about 110.3 min of steps left; card peak 13.6 GB
67
+ [ 3712.3s] step 1350/4000: task 0.7478 (mse 0.748 cos 0.614); quiet A 0.00111 off 0.00059; 990 sites; 2.45 s/step; the job: 61.9 min spent, about 108.3 min of steps left; card peak 13.6 GB
68
+ [ 3834.7s] step 1400/4000: task 0.7340 (mse 0.734 cos 0.621); quiet A 0.00107 off 0.00059; 1,007 sites; 2.45 s/step; the job: 63.9 min spent, about 106.2 min of steps left; card peak 13.6 GB
69
+ [ 3957.3s] step 1450/4000: task 0.7183 (mse 0.718 cos 0.628); quiet A 0.00108 off 0.00060; 1,047 sites; 2.45 s/step; the job: 66.0 min spent, about 104.2 min of steps left; card peak 13.6 GB
70
+ [ 4079.9s] step 1500/4000: task 0.6977 (mse 0.698 cos 0.638); quiet A 0.00106 off 0.00066; 1,011 sites; 2.45 s/step; the job: 68.0 min spent, about 102.1 min of steps left; card peak 13.6 GB
71
+ [ 4202.3s] step 1550/4000: task 0.6829 (mse 0.683 cos 0.645); quiet A 0.00105 off 0.00060; 1,012 sites; 2.45 s/step; the job: 70.0 min spent, about 100.1 min of steps left; card peak 13.6 GB
72
+ [ 4324.8s] step 1600/4000: task 0.6673 (mse 0.667 cos 0.652); quiet A 0.00106 off 0.00057; 1,042 sites; 2.45 s/step; the job: 72.1 min spent, about 98.0 min of steps left; card peak 13.6 GB
73
+ [ 4447.2s] step 1650/4000: task 0.6550 (mse 0.655 cos 0.659); quiet A 0.00106 off 0.00063; 1,022 sites; 2.45 s/step; the job: 74.1 min spent, about 96.0 min of steps left; card peak 13.6 GB
74
+ [ 4569.7s] step 1700/4000: task 0.6438 (mse 0.644 cos 0.664); quiet A 0.00110 off 0.00066; 1,026 sites; 2.45 s/step; the job: 76.2 min spent, about 93.9 min of steps left; card peak 13.6 GB
75
+ [ 4692.1s] step 1750/4000: task 0.6264 (mse 0.626 cos 0.672); quiet A 0.00112 off 0.00067; 993 sites; 2.45 s/step; the job: 78.2 min spent, about 91.9 min of steps left; card peak 13.6 GB
76
+ [ 4814.5s] step 1800/4000: task 0.6161 (mse 0.616 cos 0.678); quiet A 0.00112 off 0.00061; 1,000 sites; 2.45 s/step; the job: 80.2 min spent, about 89.9 min of steps left; card peak 13.6 GB
77
+ [ 4937.0s] step 1850/4000: task 0.5983 (mse 0.598 cos 0.686); quiet A 0.00111 off 0.00064; 999 sites; 2.45 s/step; the job: 82.3 min spent, about 87.8 min of steps left; card peak 13.6 GB
78
+ [ 5059.6s] step 1900/4000: task 0.5841 (mse 0.584 cos 0.692); quiet A 0.00109 off 0.00072; 992 sites; 2.45 s/step; the job: 84.3 min spent, about 85.8 min of steps left; card peak 13.6 GB
79
+ [ 5182.4s] step 1950/4000: task 0.5771 (mse 0.577 cos 0.697); quiet A 0.00111 off 0.00068; 1,030 sites; 2.45 s/step; the job: 86.4 min spent, about 83.7 min of steps left; card peak 13.6 GB
80
+ [ 5305.5s] step 2000/4000: task 0.5651 (mse 0.565 cos 0.703); quiet A 0.00114 off 0.00071; 1,034 sites; 2.45 s/step; the job: 88.4 min spent, about 81.7 min of steps left; card peak 13.6 GB
81
+ [ 5305.8s] anchor saved: /workspace/qarm/out/anchors/sarm_clip_mse_solo_o0_close2_s12_clip.safetensors
82
+ [ 5460.7s] CLOSE 2 (step 2000): the member's alignment read on draw 1 (512 captions, 4,871 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
83
+ [ 5460.7s] block 8: 0.479 -> 0.644 / 0.646; A 0.9892; mse 0.567 cos 0.662
84
+ [ 5460.7s] block 12: 0.431 -> 0.723 / 0.729; A 0.9836; mse 0.495 cos 0.732
85
+ [ 5460.7s] block 16: 0.400 -> 0.785 / 0.789; A 0.9890; mse 0.381 cos 0.805
86
+ [ 5460.7s] block 18: 0.388 -> 0.815 / 0.819; A 0.9858; mse 0.332 cos 0.833
87
+ [ 5460.7s] block 20: 0.373 -> 0.839 / 0.844; A 0.9863; mse 0.291 cos 0.856
88
+ [ 5460.7s] block 24: 0.383 -> 0.862 / 0.865; A 0.9855; mse 0.236 cos 0.888
89
+ [ 5460.7s] block 28: 0.310 -> 0.846 / 0.849; A 0.9823; mse 0.230 cos 0.882
90
+ [ 5460.7s] block 31: 0.292 -> 0.824 / 0.828; A 0.9755; mse 0.234 cos 0.855
91
+ [ 5460.7s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.722 / 0.785 / 0.838 / 0.856 / 0.843
92
+ [ 5460.7s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.723 against the bar .85; B2: the lowest A alignment 0.9755 against .995; fineweb share +0.00027 +- 0.00026 against +.012
93
+ [ 5460.7s] gate mean 0.202 / widest 0.282; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 65,441 (skipped {'test': 98, 'spelling': 1054, 'empty': 0, 'short': 0}); card peak 13.6 GB; host RSS 13.5 GB
94
+ [ 5583.2s] step 2050/4000: task 0.5548 (mse 0.555 cos 0.708); quiet A 0.00115 off 0.00068; 1,053 sites; 2.45 s/step; the job: 93.1 min spent, about 79.7 min of steps left; card peak 13.6 GB
95
+ [ 5705.7s] step 2100/4000: task 0.5460 (mse 0.546 cos 0.712); quiet A 0.00112 off 0.00070; 1,019 sites; 2.45 s/step; the job: 95.1 min spent, about 77.6 min of steps left; card peak 13.6 GB
96
+ [ 5828.5s] step 2150/4000: task 0.5365 (mse 0.536 cos 0.717); quiet A 0.00116 off 0.00067; 1,006 sites; 2.45 s/step; the job: 97.1 min spent, about 75.6 min of steps left; card peak 13.6 GB
97
+ [ 5951.0s] step 2200/4000: task 0.5290 (mse 0.529 cos 0.721); quiet A 0.00113 off 0.00063; 1,029 sites; 2.45 s/step; the job: 99.2 min spent, about 73.5 min of steps left; card peak 13.6 GB
98
+ [ 6073.5s] step 2250/4000: task 0.5223 (mse 0.522 cos 0.724); quiet A 0.00112 off 0.00070; 1,013 sites; 2.45 s/step; the job: 101.2 min spent, about 71.5 min of steps left; card peak 13.6 GB
99
+ [ 6196.1s] step 2300/4000: task 0.5166 (mse 0.517 cos 0.727); quiet A 0.00113 off 0.00066; 996 sites; 2.45 s/step; the job: 103.3 min spent, about 69.5 min of steps left; card peak 13.6 GB
100
+ [ 6318.5s] step 2350/4000: task 0.4985 (mse 0.499 cos 0.735); quiet A 0.00113 off 0.00070; 1,046 sites; 2.45 s/step; the job: 105.3 min spent, about 67.4 min of steps left; card peak 13.6 GB
101
+ [ 6440.9s] step 2400/4000: task 0.4993 (mse 0.499 cos 0.736); quiet A 0.00116 off 0.00066; 1,035 sites; 2.45 s/step; the job: 107.3 min spent, about 65.4 min of steps left; card peak 13.6 GB
102
+ [ 6563.2s] step 2450/4000: task 0.4849 (mse 0.485 cos 0.742); quiet A 0.00118 off 0.00074; 1,007 sites; 2.45 s/step; the job: 109.4 min spent, about 63.3 min of steps left; card peak 13.6 GB
103
+ [ 6685.8s] step 2500/4000: task 0.4772 (mse 0.477 cos 0.745); quiet A 0.00114 off 0.00068; 1,034 sites; 2.45 s/step; the job: 111.4 min spent, about 61.3 min of steps left; card peak 13.6 GB
104
+ [ 6808.5s] step 2550/4000: task 0.4736 (mse 0.474 cos 0.748); quiet A 0.00116 off 0.00070; 1,039 sites; 2.45 s/step; the job: 113.5 min spent, about 59.2 min of steps left; card peak 13.6 GB
105
+ [ 6931.0s] step 2600/4000: task 0.4674 (mse 0.467 cos 0.750); quiet A 0.00119 off 0.00071; 983 sites; 2.45 s/step; the job: 115.5 min spent, about 57.2 min of steps left; card peak 13.6 GB
106
+ [ 7053.3s] step 2650/4000: task 0.4576 (mse 0.458 cos 0.755); quiet A 0.00118 off 0.00075; 1,034 sites; 2.45 s/step; the job: 117.6 min spent, about 55.1 min of steps left; card peak 13.6 GB
107
+ [ 7175.8s] step 2700/4000: task 0.4502 (mse 0.450 cos 0.758); quiet A 0.00118 off 0.00071; 1,006 sites; 2.45 s/step; the job: 119.6 min spent, about 53.1 min of steps left; card peak 13.6 GB
108
+ [ 7298.3s] step 2750/4000: task 0.4493 (mse 0.449 cos 0.759); quiet A 0.00119 off 0.00074; 1,007 sites; 2.45 s/step; the job: 121.6 min spent, about 51.1 min of steps left; card peak 13.6 GB
109
+ [ 7420.7s] step 2800/4000: task 0.4364 (mse 0.436 cos 0.764); quiet A 0.00117 off 0.00071; 1,036 sites; 2.45 s/step; the job: 123.7 min spent, about 49.0 min of steps left; card peak 13.6 GB
110
+ [ 7543.1s] step 2850/4000: task 0.4331 (mse 0.433 cos 0.766); quiet A 0.00120 off 0.00070; 1,002 sites; 2.45 s/step; the job: 125.7 min spent, about 47.0 min of steps left; card peak 13.6 GB
111
+ [ 7665.8s] step 2900/4000: task 0.4297 (mse 0.430 cos 0.768); quiet A 0.00123 off 0.00071; 1,032 sites; 2.45 s/step; the job: 127.8 min spent, about 44.9 min of steps left; card peak 13.6 GB
112
+ [ 7788.2s] step 2950/4000: task 0.4239 (mse 0.424 cos 0.771); quiet A 0.00119 off 0.00069; 1,061 sites; 2.45 s/step; the job: 129.8 min spent, about 42.9 min of steps left; card peak 13.6 GB
113
+ [ 7910.8s] step 3000/4000: task 0.4200 (mse 0.420 cos 0.774); quiet A 0.00117 off 0.00076; 997 sites; 2.45 s/step; the job: 131.8 min spent, about 40.8 min of steps left; card peak 13.6 GB
114
+ [ 7911.0s] anchor saved: /workspace/qarm/out/anchors/sarm_clip_mse_solo_o0_close3_s12_clip.safetensors
115
+ [ 8066.0s] CLOSE 3 (step 3000): the member's alignment read on draw 1 (512 captions, 4,871 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
116
+ [ 8066.0s] block 8: 0.479 -> 0.662 / 0.665; A 0.9941; mse 0.465 cos 0.726
117
+ [ 8066.0s] block 12: 0.431 -> 0.752 / 0.755; A 0.9927; mse 0.384 cos 0.797
118
+ [ 8066.0s] block 16: 0.400 -> 0.828 / 0.831; A 0.9913; mse 0.276 cos 0.861
119
+ [ 8066.0s] block 18: 0.388 -> 0.856 / 0.858; A 0.9919; mse 0.240 cos 0.881
120
+ [ 8066.0s] block 20: 0.373 -> 0.873 / 0.875; A 0.9922; mse 0.213 cos 0.896
121
+ [ 8066.0s] block 24: 0.383 -> 0.892 / 0.894; A 0.9911; mse 0.173 cos 0.919
122
+ [ 8066.0s] block 28: 0.310 -> 0.879 / 0.881; A 0.9886; mse 0.171 cos 0.914
123
+ [ 8066.0s] block 31: 0.292 -> 0.859 / 0.863; A 0.9829; mse 0.174 cos 0.894
124
+ [ 8066.0s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.753 / 0.830 / 0.871 / 0.885 / 0.873
125
+ [ 8066.0s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.752 against the bar .85; B2: the lowest A alignment 0.9829 against .995; fineweb share +0.00014 +- 0.00016 against +.012
126
+ [ 8066.0s] gate mean 0.282 / widest 0.513; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 97,781 (skipped {'test': 158, 'spelling': 1583, 'empty': 0, 'short': 0}); card peak 13.6 GB; host RSS 13.5 GB
127
+ [ 8188.5s] step 3050/4000: task 0.4135 (mse 0.414 cos 0.777); quiet A 0.00118 off 0.00071; 1,012 sites; 2.45 s/step; the job: 136.5 min spent, about 38.8 min of steps left; card peak 13.6 GB
128
+ [ 8311.0s] step 3100/4000: task 0.4026 (mse 0.403 cos 0.783); quiet A 0.00129 off 0.00104; 1,037 sites; 2.45 s/step; the job: 138.5 min spent, about 36.8 min of steps left; card peak 13.6 GB
129
+ [ 8433.4s] step 3150/4000: task 0.4034 (mse 0.403 cos 0.783); quiet A 0.00119 off 0.00071; 1,022 sites; 2.45 s/step; the job: 140.6 min spent, about 34.7 min of steps left; card peak 13.6 GB
130
+ [ 8556.4s] step 3200/4000: task 0.3915 (mse 0.391 cos 0.788); quiet A 0.00117 off 0.00076; 1,030 sites; 2.45 s/step; the job: 142.6 min spent, about 32.7 min of steps left; card peak 13.6 GB
131
+ [ 8679.3s] step 3250/4000: task 0.3933 (mse 0.393 cos 0.788); quiet A 0.00118 off 0.00077; 1,006 sites; 2.45 s/step; the job: 144.7 min spent, about 30.6 min of steps left; card peak 13.6 GB
132
+ [ 8801.6s] step 3300/4000: task 0.3889 (mse 0.389 cos 0.790); quiet A 0.00120 off 0.00076; 1,050 sites; 2.45 s/step; the job: 146.7 min spent, about 28.6 min of steps left; card peak 13.6 GB
133
+ [ 8923.8s] step 3350/4000: task 0.3755 (mse 0.375 cos 0.796); quiet A 0.00116 off 0.00070; 1,030 sites; 2.45 s/step; the job: 148.7 min spent, about 26.6 min of steps left; card peak 13.6 GB
134
+ [ 9046.2s] step 3400/4000: task 0.3721 (mse 0.372 cos 0.798); quiet A 0.00115 off 0.00069; 1,018 sites; 2.45 s/step; the job: 150.8 min spent, about 24.5 min of steps left; card peak 13.6 GB
135
+ [ 9168.6s] step 3450/4000: task 0.3653 (mse 0.365 cos 0.801); quiet A 0.00114 off 0.00073; 1,016 sites; 2.45 s/step; the job: 152.8 min spent, about 22.5 min of steps left; card peak 13.6 GB
136
+ [ 9291.1s] step 3500/4000: task 0.3590 (mse 0.359 cos 0.803); quiet A 0.00113 off 0.00072; 1,008 sites; 2.45 s/step; the job: 154.9 min spent, about 20.4 min of steps left; card peak 13.6 GB
137
+ [ 9413.5s] step 3550/4000: task 0.3552 (mse 0.355 cos 0.806); quiet A 0.00113 off 0.00065; 1,011 sites; 2.45 s/step; the job: 156.9 min spent, about 18.4 min of steps left; card peak 13.6 GB
138
+ [ 9535.9s] step 3600/4000: task 0.3527 (mse 0.353 cos 0.807); quiet A 0.00113 off 0.00066; 1,022 sites; 2.45 s/step; the job: 158.9 min spent, about 16.3 min of steps left; card peak 13.6 GB
139
+ [ 9658.2s] step 3650/4000: task 0.3470 (mse 0.347 cos 0.809); quiet A 0.00111 off 0.00074; 1,026 sites; 2.45 s/step; the job: 161.0 min spent, about 14.3 min of steps left; card peak 13.6 GB
140
+ [ 9780.6s] step 3700/4000: task 0.3401 (mse 0.340 cos 0.813); quiet A 0.00109 off 0.00071; 1,034 sites; 2.45 s/step; the job: 163.0 min spent, about 12.3 min of steps left; card peak 13.6 GB
141
+ [ 9903.1s] step 3750/4000: task 0.3407 (mse 0.341 cos 0.813); quiet A 0.00117 off 0.00080; 1,025 sites; 2.45 s/step; the job: 165.1 min spent, about 10.2 min of steps left; card peak 13.6 GB
142
+ [10026.3s] step 3800/4000: task 0.3318 (mse 0.332 cos 0.817); quiet A 0.00111 off 0.00066; 1,001 sites; 2.45 s/step; the job: 167.1 min spent, about 8.2 min of steps left; card peak 13.6 GB
143
+ [10149.2s] step 3850/4000: task 0.3199 (mse 0.320 cos 0.823); quiet A 0.00110 off 0.00064; 1,025 sites; 2.45 s/step; the job: 169.2 min spent, about 6.1 min of steps left; card peak 13.6 GB
144
+ [10272.2s] step 3900/4000: task 0.3173 (mse 0.317 cos 0.824); quiet A 0.00114 off 0.00067; 1,020 sites; 2.45 s/step; the job: 171.2 min spent, about 4.1 min of steps left; card peak 13.6 GB
145
+ [10395.1s] step 3950/4000: task 0.3144 (mse 0.314 cos 0.826); quiet A 0.00110 off 0.00067; 1,013 sites; 2.45 s/step; the job: 173.3 min spent, about 2.0 min of steps left; card peak 13.6 GB
146
+ [10517.9s] step 4000/4000: task 0.3151 (mse 0.315 cos 0.826); quiet A 0.00110 off 0.00068; 1,023 sites; 2.45 s/step; the job: 175.3 min spent, about 0.0 min of steps left; card peak 13.6 GB
147
+ [10518.2s] anchor saved: /workspace/qarm/out/anchors/sarm_clip_mse_solo_o0_close4_s12_clip.safetensors
148
+ [10673.3s] CLOSE 4 (step 4000): the member's alignment read on draw 1 (512 captions, 4,871 sites); per block: before -> armed B vs masked A (target) / armed B vs armed A; A armed vs masked
149
+ [10673.3s] block 8: 0.479 -> 0.676 / 0.680; A 0.9944; mse 0.382 cos 0.777
150
+ [10673.3s] block 12: 0.431 -> 0.784 / 0.789; A 0.9937; mse 0.304 cos 0.842
151
+ [10673.3s] block 16: 0.400 -> 0.864 / 0.868; A 0.9932; mse 0.214 cos 0.894
152
+ [10673.3s] block 18: 0.388 -> 0.884 / 0.886; A 0.9922; mse 0.186 cos 0.909
153
+ [10673.3s] block 20: 0.373 -> 0.896 / 0.898; A 0.9915; mse 0.167 cos 0.920
154
+ [10673.3s] block 24: 0.383 -> 0.910 / 0.910; A 0.9929; mse 0.139 cos 0.935
155
+ [10673.3s] block 28: 0.310 -> 0.902 / 0.903; A 0.9897; mse 0.137 cos 0.931
156
+ [10673.3s] block 31: 0.292 -> 0.886 / 0.889; A 0.9843; mse 0.139 cos 0.916
157
+ [10673.3s] draw 2 (the repeat), target alignment at 12/16/20/24/28: 0.785 / 0.867 / 0.893 / 0.904 / 0.895
158
+ [10673.3s] B1 (served blocks >= 12, draw 1): the lowest target alignment 0.784 against the bar .85; B2: the lowest A alignment 0.9843 against .995; fineweb share +0.00029 +- 0.00019 against +.012
159
+ [10673.3s] gate mean 0.365 / widest 0.705; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 130,476 (skipped {'test': 206, 'spelling': 2100, 'empty': 0, 'short': 0}); card peak 13.6 GB; host RSS 13.6 GB
160
+ [10673.4s] DONE sarm_clip_mse_solo_o0: 4000 steps in 163.4 min of steps (2.45 s/step); 4 closes; 10673 s in all; card peak 13.6 GB; host RSS 13.6 GB; compiled chain: 5 cache entries (limit 256), 0 limit warnings