Buckets:
| #!/bin/bash | |
| #SBATCH --job-name=27b-scratch-glmfiltered-full-ddp8 | |
| #SBATCH --partition=dev | |
| #SBATCH --nodes=1 | |
| #SBATCH --ntasks=1 | |
| #SBATCH --gpus=8 | |
| #SBATCH --cpus-per-task=96 | |
| #SBATCH --mem=1500G | |
| #SBATCH --time=2-00:00:00 | |
| #SBATCH --output=/traindata/bo/autojev-27b-scratch-glmfiltered-full-20261004/train-%j.log | |
| set -euo pipefail | |
| root=/traindata/bo/autojev-27b-scratch-glmfiltered-full-20261004 | |
| python_bin=/traindata/bo/autojev-monorepo-import/.venv/bin/python | |
| test_python=/traindata/bo/autojev-lora-head-20261002/env/bin/python | |
| cd "$root/source" | |
| export PYTHONPATH="$PWD/src" PYTHONUNBUFFERED=1 TOKENIZERS_PARALLELISM=false OMP_NUM_THREADS=2 | |
| export HF_HOME=/traindata/bo/autojev-attention-ablation-20261002/hf-cache | |
| export HF_HUB_OFFLINE=1 HF_HUB_DISABLE_PROGRESS_BARS=1 | |
| export AUTOJEV_EVENTS="$root/run/events.jsonl" | |
| unset LD_LIBRARY_PATH | |
| "$python_bin" "$root/verify_inputs.py" | |
| AUTOJEV_TEST_DEVICE=cuda "$test_python" -m pytest -p no:cacheprovider -c /dev/null tests/test_distributed_training.py tests/test_warmup.py -q > "$root/gpu-preflight.log" 2>&1 | |
| cat "$root/gpu-preflight.log" | |
| "$python_bin" -m torch.distributed.run --standalone --nnodes=1 --nproc-per-node=8 --module autojev.train \ | |
| --train "$root/data/train.jsonl" --temperature "$root/data/temperature.jsonl" \ | |
| --development "$root/data/dev.jsonl" --reference "$root/data/reference.jsonl" \ | |
| --public "$root/data/public.jsonl" --run "$root/run" --output "$root/checkpoints" \ | |
| --base-model Qwen/Qwen3.8-27B --revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0 \ | |
| --epochs 1 --seed 20260920 --lr 2e-6 --warmup-ratio 0.15 --weight-decay 0.01 \ | |
| --batch-size 32 --effective-batch-size 256 --token-budget 8192 --max-length 8192 --cpu-threads 32 \ | |
| --eval-every 50 --public-eval-every 50 --resume-every 50 \ | |
| --attention-mode noncausal_full_attention --pooling last --brier-weight 0 --save-every 500 | |
Xet Storage Details
- Size:
- 1.86 kB
- Xet hash:
- 9b9791cd00a37040359ebfdeda3bf9e24dbc8f0f0783d6c4895e6b1a29c00ed5
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.