Buckets:
| # Overnight order: Jevals fast mode on the 4B models -> Jevals thinking mode for V2 + RL -> 2B pipeline -> after_2b.sh | |
| cd /workspace/decision-reasoner | |
| source .venv/bin/activate | |
| echo "{\"time\": \"$(date -u +%FT%TZ)\", \"event\": \"restart\", \"stage\": \"jevals_4b\"}" >> logs/v2_timeline.jsonl | |
| for spec in "base Qwen/Qwen3.5-4B" "v1_joint checkpoints/joint_merged" "v2_joint runs/v2/checkpoints/v2_joint_merged" "rl60 runs/v2/checkpoints/rl60_merged"; do | |
| set -- $spec | |
| [ -f data/jevals/runs/$1.json ] || python -u jevals_eval.py run --name $1 --model $2 >> logs/jevals_run.log 2>&1 || echo "FAILED $1" >> logs/jevals_run.log | |
| done | |
| python -u jevals_eval.py score > logs/jevals_score.log 2>&1 || true | |
| echo "{\"time\": \"$(date -u +%FT%TZ)\", \"event\": \"done\", \"stage\": \"jevals_4b\"}" >> logs/v2_timeline.jsonl | |
| echo "{\"time\": \"$(date -u +%FT%TZ)\", \"event\": \"start\", \"stage\": \"jevals_think_rl60\"}" >> logs/v2_timeline.jsonl | |
| [ -f data/jevals/runs/rl60_think.json ] || python -u jevals_eval.py think --name rl60 --model runs/v2/checkpoints/rl60_merged >> logs/jevals_run.log 2>&1 || echo "FAILED rl60_think" >> logs/jevals_run.log | |
| python -u jevals_eval.py score > logs/jevals_score.log 2>&1 || true | |
| echo "{\"time\": \"$(date -u +%FT%TZ)\", \"event\": \"done\", \"stage\": \"jevals_think_rl60\"}" >> logs/v2_timeline.jsonl | |
| bash runs/v2_2b/run_2b.sh > logs/v2b_driver.log 2>&1 | |
| echo "DRIVER_EXIT $?" >> logs/v2b_driver.log | |
| [ -f /workspace/decision-reasoner/after_2b.sh ] && bash /workspace/decision-reasoner/after_2b.sh > logs/after_2b.log 2>&1 | |
| echo "NIGHT_DONE" >> logs/after_2b.log | |
Xet Storage Details
- Size:
- 1.62 kB
- Xet hash:
- 0c3016ac4b4bccf626b74a59ad18fb1e22448228c2c3a173e778726b90888866
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.