Buckets:
| # Usage: run_arm.sh <name> <train flags...> -> trains runs/<name>, then evaluates Core and dev at the arm's K. | |
| set -e | |
| cd /workspace/think-decide | |
| export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True | |
| NAME=$1; shift | |
| NT=0; KMAX=1 | |
| for f in "$@"; do case $prev in --n-think) NT=$f;; esac; prev=$f; done | |
| [ "$NT" != "0" ] && KMAX=8 | |
| python src/train.py --out runs/$NAME "$@" > logs/${NAME}_train.log 2>&1 | |
| python src/eval.py --model runs/$NAME/model.safetensors --n-think $NT --kmax $KMAX --out runs/$NAME/core > logs/${NAME}_eval.log 2>&1 | |
| python src/eval.py --model runs/$NAME/model.safetensors --n-think $NT --kmax $KMAX --split dev --out runs/$NAME/dev >> logs/${NAME}_eval.log 2>&1 | |
| grep -E "eval_done|^K=" logs/${NAME}_eval.log | |
Xet Storage Details
- Size:
- 739 Bytes
- Xet hash:
- 0d120846349b80b70b6f5e9832cba025145653de0481bc69979bdf4a10d6347f
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.