-
Fix label leakage: temporal split β use first 70% of events as input, predict purchase in last 30%. Remove n_purchases/purchase_rate from features.
-
Fix model loading: use from_pretrained() instead of torch.load() for safetensors format
-
Add 03_ecommerce_finetune.ipynb β next-purchase prediction with JointFusion, LightGBM baseline comparison
-
Update 02_ecommerce notebook: add HF login, memory-free cell, subsample option for <64GB RAM machines
-
Add 02_ecommerce_pretrain.ipynb β REES46 e-commerce pre-training with sequential entropy check, wandb, push to hub
-
Fix notebook: total_mem β total_memory, add hub_model_id push, add wandb logging support
-
Add 01_finance_pretrain.ipynb β Phase 3.1 notebook for pre-training on 5M Nigerian financial transactionsβ
29 cells (18 code + 11 markdown):
- Loads electricsheepafrica/Nigerian-Financial-Transactions dataset from HF Hub
- Data profiling with matplotlib visualizations
- Converts to FINANCE_SCHEMA, groups by sender_account
- Builds hybrid domain tokenizer (97 special + BPE)
- Packs sequences, trains 24M DomainTransformer
- Loss curves, next-token predictions, t-SNE user embeddings
- Saves artifacts for fine-tuning notebook
Auto-detects GPU (L4/CPU) and adjusts batch size/epochs accordingly.