File size: 421 Bytes
ce19dee | 1 2 3 4 5 6 7 8 9 10 | #!/bin/bash
cd /workspace/rl4phyx/RL4Phyx/SFT
export CUDA_VISIBLE_DEVICES=1,2,3,6
export PYTHONUNBUFFERED=1
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
LOG="/workspace/rl4phyx/logs/train_fullft_math_f_resume_${TIMESTAMP}.log"
echo "Resuming fullft_math_f (GPUs 1,2,3,6) at $(date)" > "$LOG"
torchrun --nproc_per_node=4 --master_port=29500 train_sft_math_fullft_freeze.py >> "$LOG" 2>&1
echo "TRAINING_COMPLETE at $(date)" >> "$LOG"
|