#!/bin/bash
export CUDA_LAUNCH_BLOCKING=0
export NCCL_DEBUG=WARN
export TORCH_DISTRIBUTED_DEBUG=OFF

export OMP_NUM_THREADS=1
export HOSTNAMES=`scontrol show hostnames "$SLURM_JOB_NODELIST"`
export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
export MASTER_PORT=12835
export COUNT_NODE=`scontrol show hostnames "$SLURM_JOB_NODELIST" | wc -l`
export TRITON_CACHE_DIR=/tmp/.triton_cache_tony

TRAIN_PATH=/mnt/round-surf/home/tony/Work/glockenspiel/sunoGPT
echo working from $TRAIN_PATH
cd $TRAIN_PATH
echo starting to kill $TRAIN_PATH
kill  -9  -1
echo finish killing