rl4phyx-backup / root_scripts /launch_8gpu.sh
YUNTA88's picture
Upload root_scripts/launch_8gpu.sh with huggingface_hub
fd2ed29 verified
#!/bin/bash
cd /workspace/rl4phyx/RL4Phyx/SFT
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export PYTHONUNBUFFERED=1
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
LOG="/workspace/rl4phyx/logs/train_lora_phyx_math_f_8gpu_${TIMESTAMP}.log"
echo "Starting lora_phyx_math_f (8 GPUs, grad_accum=8, batch=64) at $(date)" > "$LOG"
torchrun --nproc_per_node=8 --master_port=29500 train_sft_phyx_math_lora_freeze.py >> "$LOG" 2>&1
echo "TRAINING_COMPLETE at $(date)" >> "$LOG"