Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
step
int64
1
128
loss
float64
0.33
4.35
lr
float64
0
0
step_seconds
float64
2.82
10
1
2.26754
0.00005
7.636
2
2.77037
0.000075
7.305
3
3.44304
0.0001
7.524
4
2.01007
0.0001
8.392
5
1.52428
0.0001
7.703
6
1.34784
0.0001
7.546
7
1.15978
0.0001
7.606
8
1.26929
0.0001
7.263
9
1.21324
0.0001
7.949
10
1.14342
0.0001
7.246
11
1.08036
0.0001
7.52
12
1.66396
0.0001
7.616
13
1.0817
0.0001
7.574
14
1.06903
0.0001
7.266
15
1.20863
0.0001
7.414
16
1.08114
0.0001
7.5
17
1.19709
0.0001
7.453
18
1.50059
0.0001
7.326
19
1.10264
0.0001
7.301
20
1.01059
0.0001
7.252
21
1.43661
0.0001
7.465
22
1.02281
0.0001
7.557
23
1.28334
0.0001
7.535
24
1.17756
0.0001
7.388
25
1.45638
0.0001
7.656
26
1.36685
0.0001
7.555
27
1.51333
0.0001
7.831
28
1.06454
0.0001
7.791
29
1.16708
0.0001
7.681
30
1.1096
0.0001
7.486
31
1.48827
0.0001
7.721
32
0.89666
0.0001
7.365
33
1.09755
0.0001
7.142
34
0.85675
0.0001
7.355
35
0.98193
0.0001
7.333
36
1.18483
0.0001
7.474
37
1.09043
0.0001
7.464
38
1.12239
0.0001
7.865
39
0.68153
0.0001
7.37
40
1.36117
0.0001
7.733
41
1.12459
0.0001
7.446
42
0.88978
0.0001
7.253
43
1.03473
0.0001
7.521
44
0.94089
0.0001
7.676
45
1.07587
0.0001
7.259
46
0.75171
0.0001
7.202
47
1.07973
0.0001
7.435
48
0.96565
0.0001
7.302
49
0.81388
0.0001
7.569
50
1.09303
0.0001
7.653
51
0.90451
0.0001
7.915
52
1.07928
0.0001
7.88
53
0.66469
0.0001
7.587
54
0.99557
0.0001
7.473
55
1.19017
0.0001
7.155
56
0.72787
0.0001
7.604
57
0.99016
0.0001
7.195
58
1.09042
0.0001
7.594
59
0.99903
0.0001
7.449
60
1.46142
0.0001
8.081
61
1.10621
0.0001
7.566
62
1.04697
0.0001
7.869
63
1.14799
0.0001
7.536
64
0.76476
0.0001
7.04
65
0.98751
0.0001
7.54
66
0.97197
0.0001
7.995
67
0.545
0.0001
7.318
68
0.6394
0.0001
7.003
69
0.92697
0.0001
7.548
70
0.9064
0.0001
7.321
71
0.68441
0.0001
7.327
72
0.90356
0.0001
7.459
73
0.90905
0.0001
7.938
74
0.784
0.0001
7.621
75
0.57032
0.0001
7.232
76
0.90228
0.0001
7.435
77
0.85901
0.0001
7.38
78
0.6258
0.0001
7.407
79
0.86136
0.0001
7.517
80
0.9097
0.0001
7.661
81
0.80019
0.0001
7.438
82
0.92413
0.0001
7.48
83
0.72252
0.0001
7.268
84
0.59392
0.0001
7.307
85
0.70131
0.0001
7.64
86
0.89729
0.0001
7.509
87
0.74044
0.0001
7.502
88
0.61589
0.0001
7.223
89
0.68688
0.0001
7.393
90
0.92137
0.0001
7.404
91
0.6323
0.0001
7.31
92
0.81812
0.0001
7.352
93
0.98413
0.0001
7.906
94
0.90494
0.0001
7.643
95
0.74468
0.0001
7.323
96
0.97946
0.0001
7.515
1
2.63934
0.00005
7.448
2
2.26824
0.000075
7.728
3
1.96916
0.0001
7.929
4
2.0487
0.0001
7.877
End of preview. Expand in Data Studio

Brink experiment results

Every experiment's logs from Brink, which detects and prevents LLM safety collapse during fine-tuning (Lexsi AI Safety & Alignment Hackathon 2026), on Cohere Tiny Aya. Code: https://github.com/andrew1234-arch/brink

One folder per experiment (go_nogo, gate, repair, multilingual, pipeline, live): REPORT.md, summary.json, plots, and the CSV logs behind them (probe signals per training step, evaluations, alarm events, per-sample gate scores).

Only rates, scores and labels are published. Model replies to harmful prompts are never published.

Downloads last month
52

Space using cmgdevl/brink-results 1

Collection including cmgdevl/brink-results