Update llama3.1-70b run_leonardo.slurm (single GPU)

llama3.1-70b-bnb/run_leonardo.slurm

 #!/bin/bash
 
 #SBATCH --partition=boost_usr_prod
-# #SBATCH --qos=boost_qos_dbg
+#SBATCH --qos=boost_qos_dbg
 
 ## Specify resources:
 ## Leonardo Booster: 32 CPU cores and 4 GPUs per node => 8 cores per GPU
 #SBATCH --nodes=1
+#SBATCH --gpus-per-task=1  # up to 4 on Leonardo
 #SBATCH --ntasks-per-node=1
-#SBATCH --gpus-per-task=1
-#SBATCH --mem-per-gpu=25GB
-#SBATCH --cpus-per-task=8
+#SBATCH --mem-per-gpu=120GB
+#SBATCH --cpus-per-task=8  # should be 8 * gpus-per-task on Leonardo
 
-#SBATCH --time=1:00:00
+#SBATCH --time=0:30:00  # up to 0:30:00 for boost_qos_dbg
 
 # Set conda environment:
 CONDA_ENV=finetuning
@@ -34,5 +34,4 @@ nvidia-smi
 
 # Run AI scripts:
 time conda run -n $CONDA_ENV --no-capture-output python llama3.1-70b_train.py
-time conda run -n $CONDA_ENV --no-capture-output python llama3.1-70b_test.py
-# time conda run -n $CONDA_ENV --no-capture-output torchrun --nproc_per_node 4 llama3.1-70b_train.py
+# time conda run -n $CONDA_ENV --no-capture-output python llama3.1-70b_test.py
\ No newline at end of file