msulemans/open-model-training-lab

1---2title: Open Model Training Lab3emoji: 🧪4colorFrom: blue5colorTo: yellow6sdk: static7app_file: index.html8pinned: false9license: mit10datasets:11  - PolyAI/banking7712models:13  - Qwen/Qwen3-1.7B-MLX-bf1614  - google-bert/bert-large-cased15  - microsoft/deberta-v3-large16tags:17  - apple-silicon18  - fine-tuning19  - mlx20  - pytorch21  - education22---23 24# Open Model Training Lab25 26An interactive, beginner-friendly account of training Qwen3, BERT-Large and27DeBERTa-v3-large for BANKING77 intent classification on an Apple M2 Max.28 29The guide covers fifteen controlled experiments, including numerical failures,30rejected refinements, data-leakage protections, a searchable glossary, quiz and31interview practice.32 33**Best recorded result:** 92.99% validation accuracy and 94.12%34reporting-only test accuracy with a DeBERTa upper-layer refinement.35 36Source and reproduction instructions:37[msulemans/open-model-training-lab](https://github.com/msulemans/open-model-training-lab)38 39This is educational software, not a production banking classifier.40