bioX Home EH14 X BGC-1 EH18 Designer API GitHub

BGC-1

180.8 GPU hours training · 274 values · 8 reports · 05 November 2026, 13:04 CET

From the other models: 26 on EH14 X and 9 on EH18 Designer.

taskoursbaseline margin
compound class: BGC-1, 2048-base windows0.74060.4625nearest known neighbour0.2781
compound class: BGC-1, 2048-base windows0.74060.5031EH14 X on the same task0.2375
compound class: BGC-1, 512-base windows0.69370.4625nearest known neighbour0.2312
bgc-330M-run55
nameBGC-1
parameters321,017,986
window2,048
corpusbgcgroot
tokens read5,520,000,000
points[0] › tokens_per_rang40,000,000
points[0] › auc0.4791
points[0] › loss1.2603
points[0] › skipped0
points[0] › minutes73
points[1] › tokens_per_rang80,000,000
points[1] › auc0.4867
points[1] › loss1.2624
points[1] › skipped0
points[1] › minutes145
points[2] › tokens_per_rang160,000,000
points[2] › auc0.4946
points[2] › loss1.2181
points[2] › skipped0
points[2] › minutes291
points[3] › tokens_per_rang240,000,000
points[3] › auc0.4932
points[3] › loss1.1907
points[3] › skipped0
points[3] › minutes436
points[4] › tokens_per_rang320,000,000
points[4] › auc0.4805
points[4] › loss1.1751
points[4] › skipped0
points[4] › minutes581
points[5] › tokens_per_rang410,000,000
points[5] › auc0.4866
points[5] › loss1.1445
points[5] › skipped0
points[5] › minutes726
points[6] › tokens_per_rang490,000,000
points[6] › auc0.4895
points[6] › loss1.114
points[6] › skipped0
points[6] › minutes871
points[7] › tokens_per_rang570,000,000
points[7] › auc0.4907
points[7] › loss1.1126
points[7] › skipped0
points[7] › minutes1,016
points[8] › tokens_per_rang690,000,000
points[8] › auc0.4885
points[8] › loss1.0876
points[8] › skipped0
points[8] › minutes1,233
final loss1.0876
perplexity2.967
best auc0.4946
tokens5,520,000,000
parameters of the log320,753,793
clustermeting15
clusters1,604
families1,534
train1,284
test320
measurement › largest class guessing › correct0.2281
measurement › largest class guessing › interval0.1812 – 0.2751
measurement › 4-mers + linear layer › correct0.4594
measurement › 4-mers + linear layer › interval0.4093 – 0.5125
measurement › EH14 X, 512-base windows › correct0.5031
measurement › EH14 X, 512-base windows › interval0.4531 – 0.5563
measurement › BGC-1, 512-base windows › correct0.6937
measurement › BGC-1, 512-base windows › interval0.6468 – 0.7438
measurement › BGC-1, 2048-base windows › correct0.7406
measurement › BGC-1, 2048-base windows › interval0.6906 – 0.7875
measured withEH14 X, 6.7 M (previous version) – BGC-1
clusterontwerp89
prefix512
written1,536
families1,534
per class40
stops floor random11.74
classes › NRPS › n40
classes › NRPS › model0.325
classes › NRPS › markov0.125
classes › NRPS › real held-out0.375
classes › NRPS › stops model0.65
classes › NRPS › orf_model1,437
classes › NRPS › stops markov5.22
classes › NRPS › orf_markov570
classes › NRPS › stops real0
classes › NRPS › orf real1,533
classes › NRPS › distance model0.65
classes › NRPS › distance markov5.22
classes › PKS › n40
classes › PKS › model0.15
classes › PKS › markov0.225
classes › PKS › real held-out0.375
classes › PKS › stops model1.3
classes › PKS › orf_model1,156
classes › PKS › stops markov5.21
classes › PKS › orf_markov550
classes › PKS › stops real0.65
classes › PKS › orf real1,353
classes › PKS › distance model0.65
classes › PKS › distance markov4.56
classes › other › n40
classes › other › model0.125
classes › other › markov0.2
classes › other › real held-out0.325
classes › other › stops model0
classes › other › orf_model1,533
classes › other › stops markov5.54
classes › other › orf_markov538
classes › other › stops real1.3
classes › other › orf real1,225
classes › other › distance model1.3
classes › other › distance markov4.24
classes › ribosomal › n40
classes › ribosomal › model0.125
classes › ribosomal › markov0.025
classes › ribosomal › real held-out0.35
classes › ribosomal › stops model0.65
classes › ribosomal › orf_model1,341
classes › ribosomal › stops markov9.78
classes › ribosomal › orf_markov418
classes › ribosomal › stops real1.3
classes › ribosomal › orf real1,273
classes › ribosomal › distance model0.65
classes › ribosomal › distance markov8.48
classes › saccharide › n33
classes › saccharide › model0.1818
classes › saccharide › markov0.1212
classes › saccharide › real held-out0.2424
classes › saccharide › stops model0.65
classes › saccharide › orf_model1,389
classes › saccharide › stops markov8.46
classes › saccharide › orf_markov417
classes › saccharide › stops real2.6
classes › saccharide › orf real1,017
classes › saccharide › distance model1.95
classes › saccharide › distance markov5.86
classes › terpene › n19
classes › terpene › model0.2105
classes › terpene › markov0.2105
classes › terpene › real held-out0.2105
classes › terpene › stops model2.61
classes › terpene › orf_model975
classes › terpene › stops markov8.46
classes › terpene › orf_markov441
classes › terpene › stops real7.18
classes › terpene › orf real759
classes › terpene › distance model4.57
classes › terpene › distance markov1.28
model mean0.1862
markov mean0.1511
real mean0.313
orf model mean1,305.1667
orf markov mean489
orf real mean1,193.3333
stops model mean0.9767
stops markov mean7.1117
stops real mean2.1717
distance model mean1.6283
distance markov mean4.94
difference0.0351
schaal-interval11
baseline 4-mers 5120.8227
eh14x_fp320.7063
draws2,000
vol › 6.4M0.6288
vol › 50M0.6542
vol › difference0.0254
vol › interval0.0026 – 0.0494
real › 6.4M0.642
real › 50M0.6931
real › difference0.0511
real › interval0.0371 – 0.0656
ontwerpmeting62
length200
groups › dinoflagellata › taxonomic placement › real held-out0.9667
groups › dinoflagellata › taxonomic placement › markov0.6167
groups › dinoflagellata › taxonomic placement › model0.1
groups › dinoflagellata › similarity › real held-out0.8955
groups › dinoflagellata › similarity › markov0.6347
groups › dinoflagellata › similarity › model0.5789
groups › dinoflagellata › model lands at › fungi21
groups › dinoflagellata › model lands at › apicomplexa10
groups › dinoflagellata › model lands at › dinoflagellata6
groups › dinoflagellata › model lands at › gyrista5
groups › dinoflagellata › model lands at › chlorophyta5
groups › dinoflagellata › model lands at › discoba4
groups › dinoflagellata › model lands at › ciliophora4
groups › dinoflagellata › model lands at › metazoa4
groups › dinoflagellata › model lands at › rhodophyta1
groups › dinoflagellata › too separate › markov › auc0.9983
groups › dinoflagellata › too separate › markov › interval0.995 – 1
groups › dinoflagellata › too separate › model › auc1
groups › dinoflagellata › too separate › model › interval1 – 1
groups › dinoflagellata › model failed0
groups › ciliophora › taxonomic placement › real held-out0.9833
groups › ciliophora › taxonomic placement › markov0.5167
groups › ciliophora › taxonomic placement › model0.1333
groups › ciliophora › similarity › real held-out0.9
groups › ciliophora › similarity › markov0.6845
groups › ciliophora › similarity › model0.6175
groups › ciliophora › model lands at › fungi19
groups › ciliophora › model lands at › apicomplexa17
groups › ciliophora › model lands at › ciliophora8
groups › ciliophora › model lands at › gyrista7
groups › ciliophora › model lands at › rhodophyta2
groups › ciliophora › model lands at › discoba2
groups › ciliophora › model lands at › dinoflagellata2
groups › ciliophora › model lands at › metazoa2
groups › ciliophora › model lands at › chlorophyta1
groups › ciliophora › too separate › markov › auc0.9961
groups › ciliophora › too separate › markov › interval0.9872 – 1
groups › ciliophora › too separate › model › auc1
groups › ciliophora › too separate › model › interval1 – 1
groups › ciliophora › model failed0
groups › metazoa › taxonomic placement › real held-out0.9667
groups › metazoa › taxonomic placement › markov0.7167
groups › metazoa › taxonomic placement › model0.0667
groups › metazoa › similarity › real held-out0.8384
groups › metazoa › similarity › markov0.6636
groups › metazoa › similarity › model0.583
groups › metazoa › model lands at › fungi21
groups › metazoa › model lands at › apicomplexa15
groups › metazoa › model lands at › ciliophora6
groups › metazoa › model lands at › rhodophyta5
groups › metazoa › model lands at › metazoa4
groups › metazoa › model lands at › discoba3
groups › metazoa › model lands at › dinoflagellata2
groups › metazoa › model lands at › gyrista2
groups › metazoa › model lands at › chlorophyta2
groups › metazoa › too separate › markov › auc0.9381
groups › metazoa › too separate › markov › interval0.8902 – 0.9762
groups › metazoa › too separate › model › auc0.9956
groups › metazoa › too separate › model › interval0.9833 – 1
groups › metazoa › model failed0
measured withno model
clusterontwerp-lat31
prefix512
written1,536
families1,534
per class40
classes › NRPS › n40
classes › NRPS › model—
classes › NRPS › markov0.125
classes › NRPS › real held-out0.375
classes › PKS › n40
classes › PKS › model—
classes › PKS › markov0.225
classes › PKS › real held-out0.375
classes › other › n40
classes › other › model—
classes › other › markov0.2
classes › other › real held-out0.325
classes › ribosomal › n40
classes › ribosomal › model—
classes › ribosomal › markov0.025
classes › ribosomal › real held-out0.35
classes › saccharide › n33
classes › saccharide › model—
classes › saccharide › markov0.1212
classes › saccharide › real held-out0.2424
classes › terpene › n19
classes › terpene › model—
classes › terpene › markov0.2105
classes › terpene › real held-out0.2105
model mean—
markov mean0.1511
real mean0.313
corpus_telling4
size7,171,796,732
rules12,824,758
on02-10-2026 13:18
measured withno model
gpu-uren7
bgc-1 › gpu hours180.85
bgc-1 › gpu minutes10,851
bgc-1 › cpu hours0
eh14x › gpu hours0
eh14x › gpu minutes0
eh14x › cpu hours22.2
eh14x › cpu minutes1,331