Buckets:
| # Routed S1MB: V2 + RL fast readout + its own thinking on the lowest-margin 30%; then the 2B hybrid | |
| # (readout-only LoRA fast readout + untrained MiniCPM5-2B thinking); then the S1MB score table. | |
| cd /workspace/decision-reasoner | |
| source .venv/bin/activate | |
| ev() { echo "{\"time\": \"$(date -u +%FT%TZ)\", \"event\": \"$1\", \"stage\": \"$2\"}" >> logs/v2_timeline.jsonl; } | |
| ev start s1mb_routed_rl60 | |
| env MAX_NUM_SEQS=160 GPU_MEM=0.92 python -u s1mb_think.py run --name rl60_routed30 --fast rl60 --reason-ref rl60 \ | |
| --model runs/v2/checkpoints/rl60_merged >> logs/s1mb_routed.log 2>&1 && ev done s1mb_routed_rl60 || { ev failed s1mb_routed_rl60; echo "FAILED rl60_routed30" >> logs/s1mb_routed.log; } | |
| ev start s1mb_routed_2b | |
| env MAX_NUM_SEQS=96 GPU_MEM=0.92 python -u s1mb_think.py run --name v2_2b_hybrid30 --fast v2_2b_ro --reason-ref base_2b \ | |
| --model openbmb/MiniCPM5-2B >> logs/s1mb_routed.log 2>&1 && ev done s1mb_routed_2b || { ev failed s1mb_routed_2b; echo "FAILED v2_2b_hybrid30" >> logs/s1mb_routed.log; } | |
| python -u s1mb_eval.py score > logs/s1mb_score.log 2>&1 && echo "SCORED" >> logs/s1mb_routed.log | |
| echo "ROUTED_DONE" >> logs/s1mb_routed.log | |
Xet Storage Details
- Size:
- 1.18 kB
- Xet hash:
- 2e0def4d02bfdd0e96fabcbb71a7fab26a506ab2c1c295a3425973c7cb8e0bb0
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.