Upload srun_83m.sh

srun_83m.sh

@@ -1,19 +1,5 @@
-#!/bin/bash
-#SBATCH --exclude=nid007571,nid007112,nid006774,nid007502,nid007506,nid007507,nid005145,nid006692,nid007218,nid007123,nid006124,nid006123,nid007496,nid007237,nid006852,nid007206,nid006947,nid007212,nid006977,nid007222,nid005444,nid007219,nid007493,nid007221,nid005300,nid005619,nid006118,nid005203,nid006113,nid006481,nid007077,nid005208,nid005207,nid005879,nid005901
-#SBATCH --nodes=8
-#SBATCH --ntasks-per-node=1
-#SBATCH --cpus-per-task=32
-#SBATCH --mem=256G
-#SBATCH -p eap
-#SBATCH -t 2-0:00:00
-#SBATCH --gpus-per-node=mi250:8
-#SBATCH --exclusive=user
-#SBATCH --hint=nomultithread
-#SBATCH --account=project_462000119
-#SBATCH -o logs/%j.out
-#SBATCH -e logs/%j.err
-
 SLURM_JOB_NUM_NODES=32
+VARIANT=83m
 
 # if run without sbatch, invoke here
 if [ -z $SLURM_JOB_ID ]; then
@@ -28,18 +14,16 @@ set -euo pipefail
 ln -f -s $SLURM_JOB_ID.out logs/latest.out
 ln -f -s $SLURM_JOB_ID.err logs/latest.err
 
-KILL_SWITCH_PATH=kill-switch-1
-CHECKPOINT_PATH=checkpoints_83m
-TENSORBOARD_PATH=tensorboard_83m
-# Start from scratch
-rm -rf "$CHECKPOINT_PATH" "$TENSORBOARD_PATH"
+KILL_SWITCH_PATH=kill-switch-$VARIANT
+CHECKPOINT_PATH=checkpoints_$VARIANT
+TENSORBOARD_PATH=tensorboard_$VARIANT
 mkdir -p $CHECKPOINT_PATH
 mkdir -p $TENSORBOARD_PATH
 
 # Data
 VOCAB_FILE="gpt2/vocab.json"
 MERGE_FILE="gpt2/merges.txt"
-DATA_PATH="/scratch/project_462000119/data/pile/megatron_data"
+DATA_PATH="/scratch/project_462000119/data/pile/megatron_data/meg-gpt2_pile_text_document"
 
 PP_SIZE=1
 TP_SIZE=1