segment int64 0 19 | lang stringlengths 2 2 | bounds listlengths 2 2 | c float64 0 1 | c_per_tensor listlengths 3 3 | resid float64 0 0.09 | sched float64 0 1 | flagged bool 2
classes |
|---|---|---|---|---|---|---|---|
0 | en | [
0,
8192
] | 0.000034 | [
0.0000343916,
0.0000343982,
0.0000343461
] | 0.054454 | 0.000032 | true |
1 | es | [
8192,
10240
] | 0.000059 | [
0.0000590983,
0.0000590983,
0.0000590983
] | 0.000021 | 0.000055 | false |
2 | pl | [
10240,
12288
] | 0.000101 | [
0.00010084,
0.0001014861,
0.0001013676
] | 0.049975 | 0.000095 | true |
3 | fr | [
12288,
14336
] | 0.00017 | [
0.0001680017,
0.0001719872,
0.0001713246
] | 0.094966 | 0.000163 | true |
4 | de | [
14336,
16384
] | 0.00029 | [
0.0002869113,
0.0002914332,
0.000290674
] | 0.076106 | 0.000281 | true |
5 | nl | [
16384,
18432
] | 0.000498 | [
0.0004974236,
0.0004981958,
0.0004979733
] | 0.0266 | 0.000485 | true |
6 | it | [
18432,
20480
] | 0.000858 | [
0.0008583444,
0.0008583444,
0.0008583444
] | 0.000019 | 0.000836 | false |
7 | sv | [
20480,
22528
] | 0.001473 | [
0.0014687479,
0.0014754486,
0.001474266
] | 0.040485 | 0.001443 | true |
8 | da | [
22528,
24576
] | 0.002519 | [
0.0025099718,
0.0025253938,
0.0025228524
] | 0.047103 | 0.002488 | true |
9 | pt | [
24576,
26624
] | 0.004325 | [
0.0043201044,
0.0043285369,
0.0043256729
] | 0.035235 | 0.004292 | true |
10 | cz | [
26624,
28672
] | 0.007413 | [
0.007402605,
0.0074222788,
0.0074142969
] | 0.043342 | 0.007402 | true |
11 | ro | [
28672,
30720
] | 0.012767 | [
0.0127665547,
0.0127665545,
0.0127665538
] | 0.000016 | 0.012767 | false |
12 | fi | [
30720,
32768
] | 0.02202 | [
0.022019855,
0.0220198566,
0.0220198574
] | 0.000015 | 0.022021 | false |
13 | hu | [
32768,
34816
] | 0.03798 | [
0.0379800276,
0.0379800287,
0.0379800264
] | 0.000014 | 0.037982 | false |
14 | bg | [
34816,
36864
] | 0.065508 | [
0.0655082589,
0.065508257,
0.0655082598
] | 0.000013 | 0.065511 | false |
15 | et | [
36864,
38912
] | 0.112989 | [
0.1129891748,
0.1129891789,
0.1129891752
] | 0.000011 | 0.112993 | false |
16 | el | [
38912,
40960
] | 0.194885 | [
0.1948846654,
0.1948846739,
0.194884656
] | 0.00001 | 0.19489 | false |
17 | sk | [
40960,
43008
] | 0.336139 | [
0.3361386462,
0.3361386141,
0.3361385859
] | 0.000008 | 0.336145 | false |
18 | sl | [
43008,
45056
] | 0.579775 | [
0.5797746718,
0.5797746416,
0.5797745733
] | 0.000006 | 0.57978 | false |
19 | lt | [
45056,
47104
] | 1 | [
1,
1,
1
] | 0 | 1 | false |
BDH Continual-Learning phase-1b checkpoints (RA2, single-copy subset)
17 PyTorch checkpoints from the RA2 route-aware growth ladder — the 20-phase
continual-learning run over 20 European languages (europarl corpora) that
precedes RA2b and whose run names (ladRA2-*) are cited throughout bdh/docs.
This repo contains exactly those files of the RA2 ladder that exist in a single place (on the gx10 training host) and therefore cannot be reclaimed locally until replicated here: the bg, el, et, fi, hu, sk, sl language best+last pairs (14 files), the final-phase lt_best.pt, and the decay-leak boundary-repair artefacts lt_repaired.pt and lt_repaired_repair.json. The remaining 25 RA2 files have byte-identical copies on a second host and are therefore not part of this upload.
Contents
bdh_europarl_ladRA2-<lang>_{best,last}.pt— 7 languages x 2 = 14 filesbdh_europarl_ladRA2-lt_best.pt— final phase (lt) of the ladderbdh_europarl_ladRA2-lt_repaired.pt+.json— boundary-repair artefacts
Total: 17 files, ~88 GiB.
Loading
Pipeline-native PyTorch .pt files, not safetensors, not loadable via
transformers AutoModel. Model architecture and loading code live in
Saga-AI-Labs/bdh-cl (pipeline/, bdh.py).
Provenance
Produced by the Saga AI Labs BDH continual-learning research line (RA2 ladder, final report and decay-leak repair measurement). License: Pathway MIT, see the source repository's LICENSE.md. These are provenance snapshots for reproducing published phase-1 numbers; no independent claims are attached to this set.
- Downloads last month
- 115