#!/bin/bash
#SBATCH --job-name="verify"
#SBATCH --output="verify_output.out"
#SBATCH --error="verify_errors.err"
#SBATCH --partition=gpuA40x4
#SBATCH --mem=12G
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=1
#SBATCH --constraint="projects,perf,nvperf"
#SBATCH --gpus-per-node=1
#SBATCH --gpu-bind=closest
#SBATCH --account=bche-delta-gpu
#SBATCH -t 00:10:00
#SBATCH --exclude=gpub029,gpub034,gpub038,gpub043,gpub074,gpub054,gpub012,gpub044,gpub083
#SBATCH --mail-user=asing271@illinois.edu
#SBATCH --mail-type=BEGIN,END,FAIL

module reset
module load cuda

echo -e "job $SLURM_JOBID is starting on `hostname`\n\n"

# ── nsys system-level profile ─────────────────────────────────────
srun nsys profile --stats=true ./test_gpt2 > test_gpt2_nsys_prof_req_perf.out

# ── ncu kernel-level profile ──────────────────────────────────────
# ncu --set full -f -o analysis_file ./test_gpt2 > test_gpt2_ncu.out