upd

tensorflow/viper/run_undistributed.slurm

@@ -15,12 +15,13 @@ module purge
 module load apptainer/1.4.1
 
 # Get the command line from the job info
-command_line=$(scontrol show job "${SLURM_JOBID}" | grep -oP 'Command=\K.*')
-lower_cell=$(echo "$SLURM_CELL" | tr '[:upper:]' '[:lower:]') # Convert SLURM_CELL to lowercase
-export SBATCH_SCRIPT_PATH="${command_line//\/$lower_cell/}" # Remove /$lower_cell from the path
+export SBATCH_SCRIPT_PATH=$(scontrol show job "$SLURM_JOBID" | grep -oP 'Command=\K.*' | sed -E "s|/$(echo "$SLURM_CELL" | tr '[:upper:]' '[:lower:]')||" | sed -E 's|^/u[12]/|/u/|')
 
-sif_file="$(dirname "$SBATCH_SCRIPT_PATH")/raven/tf-2.16.sif"
-code_dir="$(dirname "$SBATCH_SCRIPT_PATH")/src"
+
+sif_file="$(dirname "$SBATCH_SCRIPT_PATH")/tf-2.16.sif"
+code_dir="$(dirname $(dirname "$SBATCH_SCRIPT_PATH"))/src"
+echo sif_file: $sif_file
+echo code_dir: $code_dir
 
 export TF_FORCE_GPU_ALLOW_GROWTH=true
 export batch_size_per_device="--batch_size_per_device 256"

tensorflow/viper/tf-2.16-recipe.def

@@ -4,6 +4,7 @@ From: rocm/tensorflow:rocm6.3.3-py3.12-tf2.16-dev
 %post
 python -m pip install --upgrade pip
 pip install ipython ipykernel
+pip install click
 
 %environment
   export ROCBLAS_TENSILE_LIBPATH=/opt/rocm-6.3.3/lib/rocblas/library