Commit
•
9149d87
1
Parent(s):
7b58bc5
Update sbatch_8b7_178b_25b_jz_tmp.sh
Browse files
sbatch_8b7_178b_25b_jz_tmp.sh
CHANGED
|
@@ -25,10 +25,16 @@ TENSORBOARD_PATH=tensorboard_$VARIANT
|
|
| 25 |
set -x -e
|
| 26 |
source $six_ALL_CCFRWORK/start-tr13f-6B3-ml-t0
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
|
| 30 |
### LUMI / AMD ###
|
| 31 |
# if run without sbatch, invoke here
|
|
|
|
| 32 |
if [ -z $SLURM_JOB_ID ]; then
|
| 33 |
mkdir -p logs
|
| 34 |
sbatch "$0"
|
|
@@ -41,12 +47,13 @@ set -euo pipefail
|
|
| 41 |
ln -f -s $SLURM_JOB_ID.out logs/latest.out
|
| 42 |
ln -f -s $SLURM_JOB_ID.err logs/latest.err
|
| 43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
# Data
|
| 45 |
VOCAB_FILE="gpt2/vocab.json"
|
| 46 |
MERGE_FILE="gpt2/merges.txt"
|
| 47 |
-
DATA_PATH="/scratch/project_462000119/data/pile/megatron_data/meg-gpt2_pile_text_document"
|
| 48 |
-
DATA_PATH="/scratch/project_462000119/data"
|
| 49 |
-
DATA_PATH="/scratch/project_462000119/data/oscar_megatron/gpt2tok_oscar_text_document"
|
| 50 |
|
| 51 |
PP_SIZE=2
|
| 52 |
TP_SIZE=1
|
|
|
|
| 25 |
set -x -e
|
| 26 |
source $six_ALL_CCFRWORK/start-tr13f-6B3-ml-t0
|
| 27 |
|
| 28 |
+
GPUS_PER_NODE=8
|
| 29 |
+
NNODES=$SLURM_NNODES
|
| 30 |
+
|
| 31 |
+
DATA_PATH="XXXX"
|
| 32 |
+
|
| 33 |
|
| 34 |
|
| 35 |
### LUMI / AMD ###
|
| 36 |
# if run without sbatch, invoke here
|
| 37 |
+
: '
|
| 38 |
if [ -z $SLURM_JOB_ID ]; then
|
| 39 |
mkdir -p logs
|
| 40 |
sbatch "$0"
|
|
|
|
| 47 |
ln -f -s $SLURM_JOB_ID.out logs/latest.out
|
| 48 |
ln -f -s $SLURM_JOB_ID.err logs/latest.err
|
| 49 |
|
| 50 |
+
DATA_PATH="/scratch/project_462000119/data/oscar_megatron/gpt2tok_oscar_text_document"
|
| 51 |
+
'
|
| 52 |
+
|
| 53 |
+
|
| 54 |
# Data
|
| 55 |
VOCAB_FILE="gpt2/vocab.json"
|
| 56 |
MERGE_FILE="gpt2/merges.txt"
|
|
|
|
|
|
|
|
|
|
| 57 |
|
| 58 |
PP_SIZE=2
|
| 59 |
TP_SIZE=1
|