| # Usage: run_main.sh [EFFORT] [MAX_TOKENS] [--with-eval] | |
| # Generates the train split until the daily output-token budget hits the reserve. | |
| # --with-eval first generates teacher references for the held-out eval prompts. | |
| cd "${0:A:h}/.." | |
| e=${1:-medium}; mt=${2:-16384} | |
| if [ "$3" = "--with-eval" ]; then | |
| .venv/bin/python -I scripts/generate.py --pool data/pool.jsonl --ids data/eval_ids.txt --out data/gen/eval.jsonl --effort $e --max-tokens $mt --concurrency 20 --per-minute 55 --log-every 50 --reserve-out 50000 | |
| fi | |
| .venv/bin/python -I scripts/generate.py --pool data/pool.jsonl --ids data/train_ids.txt --out data/gen/train.jsonl --effort $e --max-tokens $mt --concurrency 20 --per-minute 55 --log-every 50 --reserve-out 50000 | |