Skip to content

v3.0.0 — Distilled production model

Choose a tag to compare

View all tags
@arnavd371 arnavd371 released this 04 Aug 09:10
v3.0.0
b978720

Summary

  • Distilled production CNN from gbdt + extratrees + cnn_deep soft labels (Kaggle T4)
  • Seed-42 val: 97.15% (macro F1 0.977) vs hard-only 95.57% and teacher ensemble 95.89%
  • Shipped INT8 TFLite (~333 KB) + ONNX for web/
  • Apache LICENSE / NOTICE filled with Arnav Dhiman contact details
  • Distill pipeline: model/distill.py, model/export_deploy.py, experiments/kaggle/push_distill.sh

Report

See experiments/results/distill/distill_report.md.

Not in this release

Field microphone corpus / phone-on-bag calibration (still desk-bound IRRI + ambient clean).

Assets 6
Loading