cleaning

bench/check_status.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-
-# Initialize counters
-declare -A counts
-statuses=("init" "pending" "running" "fail" "oom" "timeout" "completed")
-for status in "${statuses[@]}"; do
-    counts[$status]=0
-done
-
-# Find and process all status.txt files
-while IFS= read -r -d '' file; do
-    status=$(cat "$file" | tr -d '[:space:]')
-
-    if [[ " ${statuses[@]} " =~ " ${status} " ]]; then
-        ((counts[$status]++))
-    fi
-done < <(find "$1" -name "status.txt" -print0)
-
-# Calculate total
-total=0
-for count in "${counts[@]}"; do
-    ((total += count))
-done
-
-# Print the results
-echo "Status     | Count"
-echo "-----------|---------"
-for status in "${statuses[@]}"; do
-    printf "%-10s | %d\n" "$status" "${counts[$status]}"
-done
-echo "-----------|---------"
-echo "Total      | $total"

submit_slurm_jobs.py

@@ -91,21 +91,21 @@ class Scheduler:
             "qos": job.qos,
         }
         
-        #TODO: don't hardcode the base_bench.slurm path. Should be #HOME/bench_cluster/template/base_bench.slurm
+        #TODO: don't hardcode the base_job.slurm path. Should be #HOME/bench_cluster/template/base_job.slurm
         if cluster == "hf":
-            base_path = "/fsx/ferdinandmom/ferdinand-hf/picotron/bench/template/base_bench.slurm"
+            base_path = "/fsx/ferdinandmom/ferdinand-hf/picotron/bench/template/base_job.slurm"
         else:
             raise ValueError("Invalid cluster")
         
         with open(base_path, 'r') as file:
-            base_bench_file = file.read()
+            base_job_file = file.read()
         
-        base_bench_template = Template(base_bench_file)
+        base_job_template = Template(base_job_file)
                 
         # Write the rendered script to a new file located at the job root_path
-        output_file_path = os.path.join(job.root_path, "bench.slurm")
+        output_file_path = os.path.join(job.root_path, "job.slurm")
         with open(output_file_path, 'w') as file:
-            file.write(base_bench_template.render(context_bench))
+            file.write(base_job_template.render(context_bench))
 
         print(f"Slurm script created at {output_file_path}")
     

template/base_job.slurm

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-#SBATCH --job-name=bench-picotron
+#SBATCH --job-name=job-picotron
 #SBATCH --time=00:30:00
 #SBATCH --partition=hopper-prod
 #SBATCH --nodes={{ nodes }}