CoolFace
Modelpublic

StatisticsNetherlands/GECKOv2-ColBERT-NL-prod-v2

sourceHugging Facemitupdated 7d agoView on Hugging Face
0likes38downloads
Model Card

GECKOv2-ColBERT-NL-prod-v2

ColBERT retriever for GECKOv2 (nl), generated 2026-09-16 10:25.

Training configuration

SettingValue
Base modelBAAI/bge-m3
Modeall
Training strategysingle-round with cosine-sim negatives
Negative samples10 per (query, positive) pair
nway11 (1 positive + 10 negatives per step)
Learning rate1e-05
Batch size16 (accumsteps 1, 16 per forward pass)
Epochs10
Search depth (k)200
Mix ratio0.0
Quantization (nbits)4
Embedding modelsentence-transformers/all-MiniLM-L12-v2
Benchmark tablesincluded (production model)

Data

  • Split: 1113 train / 148 validation questions, 214 train tables / 24 validation tables, 0 overlapping (should be 0)
  • Collection: 169,224 documents

Rounds

RoundNegatives fromTriplesStepsdoc_maxlendimFinished
1cosine similarity of embedded descriptions64,3304,0212561282026-09-11 15:57

Evaluation

CheckpointDatasetEMacc@k
colbertv2-dutch-all-nodes-production-dim128complex_nl_test.jsonl0.7730.998 (@50)