| # Train Arcspan on CyNER 5-class data. | |
| # Usage: bash scripts/train_cyner.sh [EPOCHS] [DEVICE] | |
| set -euo pipefail | |
| EPOCHS="${1:-40}" | |
| DEVICE="${2:-cuda}" | |
| BATCH_SIZE="${3:-8}" | |
| LR="2e-4" | |
| ROOT="$(cd "$(dirname "$0")/.." && pwd)" | |
| echo "=== Arcspan CyNER Training ===" | |
| echo " Epochs: $EPOCHS | Device: $DEVICE | Batch: $BATCH_SIZE | LR: $LR" | |
| echo " Data: $ROOT/data/processed/cyner_train.jsonl" | |
| echo " Output: $ROOT/checkpoints/cyner_v1" | |
| echo "" | |
| opf train "$ROOT/data/processed/cyner_train.jsonl" \ | |
| --validation-dataset "$ROOT/data/processed/cyner_valid.jsonl" \ | |
| --output-dir "$ROOT/checkpoints/cyner_v1" \ | |
| --label-space-json "$ROOT/data/label_spaces/cyner_5class.json" \ | |
| --device "$DEVICE" \ | |
| --epochs "$EPOCHS" \ | |
| --batch-size "$BATCH_SIZE" \ | |
| --learning-rate "$LR" \ | |
| --overwrite-output | |