2 Commits

Author	SHA1	Message	Date
Codex	2ea14bd465	docs(model): training results addendum for LoRA sentiment + emotion ... - FinBERT: 5 epochs, early stop at ~4.8, 6.2MB, F1=0.907, acc=0.955 - DistilRoBERTa: 5 epochs, early stop at ~4.8, 8.1MB, F1=0.594, acc=0.865 - Both with early stopping (patience=3) - Known issues documented for next iteration	2026-09-26 19:59:56 +02:00
Codex	ddf94b1704	docs(model): model card with training status + event classifier validation ... - Documented exact training state of all 4 ONNX models - FinBERT: base ProsusAI/finbert (no crypto fine-tune) - bert-base-event: VALIDATED fine-tuned on crypto events (8/8 test cases pass) - DistilRoBERTa emotion: base j-hartmann model - MiniLM-L6-v2: base sentence transformer - Backup created at /tmp/models_backup_20260926_154811.tar.gz - Addendum log for append-only tracking	2026-09-26 17:43:55 +02:00